aeroftp-cli

about

Categories: safe

Summary: Show detailed server info, account, and storage quota

Description: Show detailed server info, account, and storage quota

Syntax: aeroftp-cli about [OPTIONS] [URL]

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Server and account info

aeroftp-cli about --profile ftps-prodServer:        ftps.example.com
Protocol:      FTPS (explicit, TLS 1.3)
Account:       aeroftp@aeroftp.app
Quota:         1.8 / 5.0 GB
Features:      RESUME, MLST, UTF-8, EPSV
Latency:       42 ms (RTT to server)

access

Categories: modify

Summary: Set OpenDrive privacy of an existing file/folder (#252)

Description: Issue #252: set the privacy level of an EXISTING file or folder on providers that model access (OpenDrive). The CLI counterpart of the GUI Properties > Permissions tab; file-vs-folder is detected via `stat`

Syntax: aeroftp-cli access --profile NAME /path --to public|private|hidden

Options

• --to : Privacy level to set: private, public, or hidden Possible values: - private: Not listed, not shared; reachable only by the owner - public: Anyone with the link can access; searchable - hidden: Accessible by direct link only; not searchable
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `access` (offline help)

aeroftp-cli help accessIssue #252: set the privacy level of an EXISTING file or folder on providers that model access (OpenDrive). The CLI counterpart of the GUI Properties > Permissions tab; file-vs-folder is detected via `stat`

Usage: aeroftp-cli access [OPTIONS] --to <TO> [URL] [PATH]

Arguments:
  [URL]
          Server URL (omit when using --profile)

  [PATH]
          Remote path of the file or folder
          
          [default: ""]

Options:
      --to <TO>
          Privacy level to set: private, public, or hidden

          Possible values:
          - private: Not listed, not shared; reachable only by the owner
          - public:  Anyone with the link can access; searchable
          - hidden:  Accessible by direct link only; not searchable

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

aerocloud

Categories: advanced

Summary: AeroCloud folder sync: configure and run the always-on cloud folder from the CLI.

Description: AeroCloud folder sync: configure and run the always-on cloud folder from the CLI.

Syntax: aeroftp-cli aerocloud [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `aerocloud` (offline help)

aeroftp-cli help aerocloudAeroCloud folder sync: configure and run the always-on cloud folder from the CLI.

AeroCloud keeps one local folder mirrored with a remote folder on a saved profile (like Dropbox/Drive). The GUI configures it and a background worker runs it on a schedule; these subcommands expose the same config plus a one-shot `sync` so it is scriptable and headless-testable. `sync` needs the vault open (see `--master-password` / AEROFTP_MASTER_PASSWORD).

Usage: aeroftp-cli aerocloud [OPTIONS] <COMMAND>

Commands:
  show     Print the current AeroCloud configuration (folders, profile, direction, interval, enabled, versioning, last sync)
  set      Mutate the AeroCloud configuration. Only the flags you pass are changed; everything else is preserved. Writes the config to disk on success
  enable   Enable AeroCloud (sets enabled=true and saves)
  disable  Disable AeroCloud (sets enabled=false and saves)
  status   Synthetic, machine-friendly status: enabled, last sync, folder pair, direction
  sync     Run one synchronization cycle now (the same path the background worker uses). Requires the vault open and AeroCloud enabled
  pair     Subcommands for multiple AeroCloud pairs (cloud_pairs.json store). Each pair is independent (own profile, direction, excludes)
  help     Print this message or the help of the given subcommand(s)

Options:
  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

aerorsync

Categories: modify

Summary: AeroRsync delta engine: configuration and live diagnostics

Description: AeroRsync delta engine: configuration and live diagnostics

Syntax: aeroftp-cli aerorsync [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Probe the aerorsync delta engine

aeroftp-cli aerorsync probe --profile sftp-prodMode:     auto (will pick native if both ends support rsync >=3.2)
Probe:    SSH endpoint OK, rsync 3.2.7 detected on remote
Result:   NATIVE rsync available (Z.4.5 R1 transport)
Use:      aeroftp-cli sync --profile sftp-prod ... will accelerate

agent

Categories: advanced

Summary: AeroAgent - AI-powered interactive agent with tool execution

Description: AeroAgent - AI-powered interactive agent with tool execution

Syntax: aeroftp-cli agent [OPTIONS]

Options

• -m, --message : One-shot message (run and exit)
• -p, --provider : AI provider (anthropic, openai, gemini, ollama, etc.)
• --model : Model override
• -c, --connect : Pre-connect to server on startup
• --auto-approve : Auto-approve tool calls: safe, medium, high, all (default: safe)
• --max-steps : Max autonomous steps (default: 10) (default: 10)
• --orchestrate: Orchestration mode (JSON-RPC 2.0 over stdin/stdout)
• --mcp: MCP server mode (Model Context Protocol)
• --stdin: Read message from stdin
• -y, --yes: Auto-approve all tools (equivalent to --auto-approve all)
• --plan-only: Plan only - show execution plan without running
• --cost-limit : Cost limit in USD (stop when exceeded)
• --system : Custom system prompt (or @file.txt to load from file)
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Start the agent surface (MCP stdio)

aeroftp-cli agent --mcpAeroFTP agent v3.8.5 ready
Transport:   stdio (JSON-RPC 2.0, MCP)
Tools:       39 exposed (read 21, write 14, system 4)
Waiting for client...

agent-bootstrap

Categories: advanced

Summary: Show the canonical task-oriented quick-start for AI agents

Description: Show the canonical task-oriented quick-start for AI agents

Syntax: aeroftp-cli agent-bootstrap [OPTIONS]

Options

• --task : Optional task focus for tailored commands [possible values: explore, verify-file, transfer, backup, reconcile]
• --path : Remote path or working path for the task
• --pattern : Optional filename or glob pattern
• --source-profile : Source profile for transfer/backup flows
• --dest-profile : Destination profile for transfer/backup flows
• --source-path : Source remote path for transfer/backup flows
• --dest-path : Destination remote path for transfer/backup flows
• --local-path : Local path for local-vs-remote verification flows
• --remote-path : Remote path for local-vs-remote verification flows
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Canonical task workflows for AI agents

aeroftp-cli agent-bootstrap --json{
  "intent_examples": [
    {"goal":"upload site","workflow":["connect","sync-doctor","sync"]},
    {"goal":"clone bucket","workflow":["transfer-doctor","transfer"]},
    {"goal":"recover space","workflow":["ncdu","cleanup","dedupe"]}
  ]
}

agent-connect

Categories: advanced

Summary: Single-shot connect surface for agents: returns per-block status (connect/capabilities/quota/path) in one JSON call. Replaces the boilerplate `connect → about → df → ls /` sequence.

Description: Single-shot connect surface for agents: returns per-block status (connect/capabilities/quota/path) in one JSON call. Replaces the boilerplate `connect → about → df → ls /` sequence.

Syntax: aeroftp-cli agent-connect [OPTIONS] <PROFILE>

Options

• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Open an ad-hoc connection for the agent

aeroftp-cli agent-connect sftp://ci@build.example.com:/srv/staging --json{
  "session_id":  "s_8c1d4e",
  "protocol":    "sftp",
  "host":        "build.example.com",
  "ttl_s":       900,
  "tools":       ["ls","get","put","sync","rm"]
}

agent-info

Categories: advanced

Summary: Show machine-readable CLI capabilities

Description: Show CLI capabilities for AI agent discovery (always JSON)

Syntax: aeroftp-cli agent-info --json

Options

• --redact-identifiers: Redact per-profile host and username identifiers (PRIV-01). Use when a large agent context only needs tool discovery, not account metadata; the full identifiers stay available via `profiles --json`
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Machine-readable capability discovery

aeroftp-cli agent-info --json{
  "version": "3.8.5",
  "capabilities": {
    "agent_native_tools": 39,
    "main_command_groups": 38
  },
  "commands": { "safe": [...], "modify": [...], ... }
}

ai-models

Categories: advanced

Summary: List configured AI providers and models from the encrypted vault

Description: List configured AI providers and models from the encrypted vault

Syntax: aeroftp-cli ai-models [OPTIONS]

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

List AI models known to the agent layer

aeroftp-cli ai-modelsNAME                          VENDOR       CTX     ROLE
claude-opus-4-7               anthropic    200K    planner
claude-sonnet-4-6             anthropic    200K    actor
claude-haiku-4-5              anthropic    200K    fast-paths
gpt-5o                        openai       128K    fallback
4 models registered

alias

Categories: modify

Summary: Manage CLI aliases stored in config.toml

Description: Manage CLI aliases stored in config.toml

Syntax: aeroftp-cli alias [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

List CLI aliases

aeroftp-cli alias listALIAS         EXPANDS TO
prod          --profile ftps-prod
backup        --profile hetzner-backup
deploy        sync --profile ftps-prod ./_site/ /srv/site/
3 aliases defined

alias-toggle

Categories: modify

Summary: Toggle an opt-in shell alias (default: aero) on or off

Description: Toggle an opt-in shell alias (default: aero) on or off

Syntax: aeroftp-cli alias-toggle [OPTIONS] [NAME]

Options

• --bin-dir : Override the target directory (default: ~/.local/bin on Unix, %LOCALAPPDATA%\AeroFTP\bin on Windows)
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Enable or disable an alias

aeroftp-cli alias-toggle deploy --offAlias 'deploy' is now DISABLED (kept in config)
Re-enable with: aeroftp-cli alias-toggle deploy --on

archive

Categories: advanced

Summary: Plaintext archive operations for `.aerozip` (application/x-aerozip).

Description: Plaintext archive operations for `.aerozip` (application/x-aerozip).

Syntax: aeroftp-cli archive [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `archive` (offline help)

aeroftp-cli help archivePlaintext archive operations for `.aerozip` (application/x-aerozip).

Creates, extracts, and lists passwordless archives backed by the aerovz lane: compression + integrity + Reed-Solomon recovery, not confidentiality.

Usage: aeroftp-cli archive [OPTIONS] <COMMAND>

Commands:
  create   Create an unencrypted `.aerozip` archive (application/x-aerozip)
  extract  Extract an unencrypted `.aerozip` archive
  list     List entries in an unencrypted `.aerozip` archive
  help     Print this message or the help of the given subcommand(s)

Options:
  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

audit

Categories: safe

Summary: Run autonomous server audits across saved profiles (M3)

Description: Run autonomous server audits across saved profiles (M3)

Syntax: aeroftp-cli audit [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Audit recent agent operations

aeroftp-cli audit --since 24hWHEN                 ACTOR        ACTION              TARGET
2026-05-27 09:14:02  agent:claude sync                /srv/site
2026-05-27 10:02:18  user:alex    profile-add         staging
2026-05-27 10:44:51  agent:claude transfer            /srv -> /backups
3 entries in last 24h

batch

Categories: advanced

Summary: Run batch automation scripts

Description: Execute commands from a batch script (.aeroftp-script file). The `.aeroftp` extension is reserved for profile-export backups and is rejected here with a rename hint (issue #225)

Syntax: aeroftp-cli batch file.aeroftp-script

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Run a batch automation script

aeroftp-cli batch ./deploy.aeroftpLoaded ./deploy.aeroftp (12 steps)
[1/12] connect ftps-prod        OK
[2/12] sync ./_site /srv/site   OK  (52 added, 14 updated)
[3/12] hashsum /srv/site/...    OK
...
[12/12] notify slack            OK
Batch finished in 24.1s

benchmark

Categories: safe

Summary: Run a community-driven benchmark across upload/download/list/stat operations. Output is sanitized: no hostnames, paths, credentials, or bucket names are recorded. Schema v1 contract in `docs/dev/roadm

Description: Run a community-driven benchmark across upload/download/list/stat operations. Output is sanitized: no hostnames, paths, credentials, or bucket names are recorded. Schema v1 contract in `docs/dev/roadmap/APPENDIX-BENCHMARK/01_JSON-Schema-v1.md`

Syntax: aeroftp-cli benchmark [OPTIONS] [LEVEL]

Options

• --sizes : Override single-file sweep sizes (comma-separated). A bare number is MEBIBYTES (e.g. "1,100,1024"); K/M/G suffixes are explicit (e.g. "1M,100M,1G")
• --runs : Override timed runs per (operation, size) tuple
• --operations : Comma-separated operations: upload,download,list,stat,delete
• --consent-publish: Append anonymized environment metadata and print a paste-ready Issue block
• --anonymize-extra: Hash provider hint as well (extra anonymization)
• --report : Write the JSON report to a file
• --profile-timeout : Total benchmark timeout in seconds (default 3600). Bumps the cap for slow cold storage (idrive S3, InfiniCloud) so 1G transfers can finish without aborting mid-sweep
• --test-root-prefix : Custom remote prefix for the scratch tree (default: provider initial path joined with `aeroftp-bench`). Use this to redirect the bench payloads under a writable sub-path on providers that refuse operations on `/` (kDrive, SeaFile WebDAV)
• --pre-delete: Delete the remote payload between successive upload runs of the same size. Workaround for strict providers (4shared and several WebDAV servers) that reject overwrite-on-PUT
• --file-count : Many-small-files workload: number of files to exercise (e.g. 100, 1000). When > 0, runs a dedicated small-files benchmark on a separate axis (upload-all / download-all / list-dir / stat-all / delete-all) reporting files/sec and mean per-file time. This is the workload where per-file overhead (handshake, signing, metadata round-trips) dominates and S3 typically pulls ahead of WebDAV and FTP. On its own it replaces the single-file size sweep; add explicit --sizes to run both axes together
• --file-size : Size of each file in the many-small-files workload. A bare number is MEBIBYTES (e.g. 1 = 1 MiB, matching how chunk sizes are discussed); K/M/G suffixes are explicit (e.g. 64K, 4M). Defaults to 64K. Total payload (file-count x file-size) is capped at 5 GiB (default: 64K)
• --compare : Compare several saved profiles in one run: comma or space separated selectors (names, indices or ids, e.g. "S3 Blackblaze,S3 Storj"). Each profile runs the same workload, then the CLI prints a combined comparison table. Without this flag, benchmark runs the single profile from the global --profile
• -i, --interactive: Interactively type the profiles to compare: prints a numbered list of saved servers, then reads names and/or index numbers (space separated) from a prompt. Reuses the current benchmark options (level, --file-count, ...). For a full-screen checklist use --tui instead. Requires an interactive terminal
• --tui: Pick the profiles to compare from a full-screen checklist (tick the servers with Space/v, navigate with arrows or Tab, run with Enter). Reuses the current benchmark options (level, --file-count, ...). For inline name/index typing use -i instead. Requires an interactive terminal
• --group : Benchmark the members of one or more groups (the My Servers group chips). Repeatable and comma-aware: `--group A --group B` or `--group A,B`. Without -i/--tui every member of the named groups is benchmarked; with -i/--tui the picker is restricted to those members. Group membership is the same `config_server_groups` blob the GUI uses
• --all: Benchmark every saved profile (Ehud #277): the opt-in "all available" selection, so you never have to list them with --compare or tick them with -i/--tui. Combine with --group to run all members of those groups, or with -i/--tui to pre-tick the whole list. Profiles that fail to connect are reported as skipped, not fatal. `--all-profiles` is an accepted spelling: it pairs read-for-read with `--all-protocols`, and the two combine to benchmark every profile over every transport mode its provider offers.
• --all-protocols: Expand every selected profile into one run per transport mode of its provider (Ehud #277): a Koofr profile is benchmarked over both its REST API and its WebDAV surface, giving one comparison row and one mini table per mode, exactly as if you had saved a profile for each. Combines with --all, --compare, --group and the pickers. Modes whose credentials or endpoint cannot be resolved are reported as skipped, never measured as zero. Multi-mode providers: Koofr, OpenDrive, FileLu, Filen
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Full benchmark suite

aeroftp-cli benchmark --profile ftps-prod --duration 60sPhase 1/4: tiny files (1 KB x 1000)   ...  87 ops/s
Phase 2/4: medium  (1 MB x 32)        ...   9.4 MB/s
Phase 3/4: large   (64 MB x 2)        ...  12.1 MB/s
Phase 4/4: list+stat sweep (5000 obj) ... 412 obj/s
Score: 6.8 / 10 (FTPS, residential uplink)

cat

Categories: safe

Summary: Print file to stdout

Description: Print remote file to stdout (for piping)

Syntax: aeroftp-cli cat --profile NAME /path/file

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Print a remote text file to stdout

aeroftp-cli cat --profile ftps-prod /srv/site/robots.txtUser-agent: *
Allow: /
Sitemap: https://example.com/sitemap.xml

catalog

Categories: advanced

Summary: Browse the supported provider catalog (companies, protocols, free tier, regions)

Description: Browse the supported provider catalog (companies, protocols, free tier, regions)

Syntax: aeroftp-cli catalog [OPTIONS] [QUERY]

Options

• --category : Filter to one category: protocols, object-storage, webdav, cloud-storage, media-services, developer (aliases: s3, object, cloud, media, dev)
• --free: Show only companies with a free tier (at least one free method)
• --paid: Show only paid-only companies (no free connection method)
• --protocols: Expand to one row per connection method instead of per company
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `catalog` (offline help)

aeroftp-cli help catalogBrowse the supported provider catalog (companies, protocols, free tier, regions)

Mirrors the GUI "Add Service" list view, generated from the same `providerCatalog.ts` single source of truth (embedded as `cli_catalog.json`, regenerated by `npm run gen:cli-catalog`). Purely local: no vault, no network. Use `--json` for the full structured data.

Usage: aeroftp-cli catalog [OPTIONS] [QUERY]

Arguments:
  [QUERY]
          Case-insensitive search over company, country, regions, protocol labels, provider ids and free-tier notes

Options:
      --category <CATEGORY>
          Filter to one category: protocols, object-storage, webdav, cloud-storage, media-services, developer (aliases: s3, object, cloud, media, dev)

      --free
          Show only companies with a free tier (at least one free method)

      --paid
          Show only paid-only companies (no free connection method)

      --protocols
          Expand to one row per connection method instead of per company

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

check

Categories: advanced

Summary: Compare local and remote trees

Description: Verify local and remote directories are identical

Syntax: aeroftp-cli check --profile NAME ./local /remote

Options

• --checksum: Use checksums instead of size/mtime
• --one-way: Only check files present locally
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Verify local and remote trees are identical

aeroftp-cli check --profile ftps-prod ./_site /srv/siteScanning local: ./_site  (1248 files, 47 MB)
Scanning remote: /srv/site  (1248 files, 47 MB)
Hash mode: size+mtime (use --hash sha256 for cryptographic)
OK - trees identical

checkpoints

Categories: advanced

Summary: Inspect and prune the multipart resume records (transfer checkpoints).

Description: Inspect and prune the multipart resume records (transfer checkpoints).

Syntax: aeroftp-cli checkpoints [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `checkpoints` (offline help)

aeroftp-cli help checkpointsInspect and prune the multipart resume records (transfer checkpoints).

A resumable upload leaves a record so it can restart where it stopped. The TTL scavenger only prunes a record when the SAME endpoint is used again, so a server you stop using keeps its records until the store hits its cap, and the cap evicts terminal residue before resumable transfers. This is the explicit escape for a decommissioned server.

Usage: aeroftp-cli checkpoints [OPTIONS] <COMMAND>

Commands:
  list    List the endpoints that currently hold resume records
  forget  Drop every resume record bound to one endpoint
  help    Print this message or the help of the given subcommand(s)

Options:
  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

cleanup

Categories: destructive

Summary: Remove orphaned .aerotmp files from interrupted downloads

Description: Remove orphaned .aerotmp files from interrupted downloads

Syntax: aeroftp-cli cleanup [OPTIONS] [URL] [PATH]

Options

• --force: Actually delete orphaned files (default: dry-run listing)
• --min-age : Only delete .aerotmp files older than this age (e.g. 30m, 2h, 1d). Recent temp files may belong to an in-progress transfer and are kept. Files without a known modification time are never deleted. Default: 1h (default: 1h)
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Reclaim trashed / temp data

aeroftp-cli cleanup --profile filen-personal --older-than 30dScanning trash on filen-personal...
42 items eligible (older than 30d, 312 MB)
Purging...
OK - 312 MB reclaimed in 1.8s

completions

Categories: modify

Summary: Generate shell completions (bash, zsh, fish, elvish, powershell)

Description: Generate shell completions (bash, zsh, fish, elvish, powershell)

Syntax: aeroftp-cli completions [OPTIONS] <SHELL>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Generate bash completions

aeroftp-cli completions bash > ~/.local/share/bash-completion/completions/aeroftp-cli# Wrote 18 KB to /home/user/.local/share/bash-completion/completions/aeroftp-cli
# Reload your shell or run: source ~/.bashrc

compress

Categories: advanced

Summary: Compress files/folders into an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2).

Description: Compress files/folders into an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2).

Syntax: aeroftp-cli compress [OPTIONS] <OUTPUT> <PATHS>...

Options

• --archive-format : Archive format override (default: infer from the output extension) [possible values: zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar]
• -l, --level : Compression level 0-9 (zip, 7z, tar.gz, tar.xz, tar.bz2; default 5 = 7-Zip "Normal"; 0 = store for zip). 7-Zip presets: 1 Fastest, 3 Fast, 5 Normal, 7 Maximum, 9 Ultra. For 7z it maps onto the LZMA2 preset
• -p, --password : AES-256 password for zip / 7z (explicit intent to encrypt). For an ambient default applied only to encryptable formats, set AEROFTP_ARCHIVE_PASSWORD
• --encrypt-names: 7z only: also encrypt the archive header so the filenames are hidden, not just the file contents (7-Zip's "Encrypt file names", -mhe=on). Requires --password and is ignored for other formats. Off by default
• --method : 7z only: content compression method (default lzma2). lzma is the classic single-stream LZMA; ppmd suits text; bzip2 is block sorting. All are readable by the extractor. Ignored for other formats [possible values: lzma2, lzma, ppmd, bzip2]
• --dictionary : 7z LZMA2 only: dictionary size in bytes (encoder clamps to 4096..=4 GiB), e.g. 16777216 for 16 MiB. Ignored for other methods/formats
• --solid: 7z only: pack every file into one solid block (better ratio on many small files, slower random extraction). Off by default
• --threads : 7z LZMA2 only: compression thread count (1 = single-threaded). Ignored for other methods/formats
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `compress` (offline help)

aeroftp-cli help compressCompress files/folders into an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2).

The format is inferred from the output extension unless `--archive-format` is given. `zip` and `7z` accept an AES-256 `--password`; the tar family is unencrypted. The standalone `gz` / `xz` / `bz2` compress exactly ONE file (single-stream, no tar wrapper): pass a single file, or use tar.gz/tar.xz/tar.bz2 to bundle many. 7z exposes advanced knobs: `--method` (lzma2/lzma/ppmd/bzip2), `--dictionary`, `--threads` (LZMA2), and `--solid`; all are ignored for the other formats. Local-only, no connection. Honors `--json` for scriptable output.

Usage: aeroftp-cli compress [OPTIONS] <OUTPUT> <PATHS>...

Arguments:
  <OUTPUT>
          Output archive path (format inferred from its extension unless --format)

  <PATHS>...
          Files and/or folders to add to the archive (one or more)

Options:
      --archive-format <ARCHIVE_FORMAT>
          Archive format override (default: infer from the output extension)
          
          [possible values: zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar]

  -l, --level <LEVEL>
          Compression level 0-9 (zip, 7z, tar.gz, tar.xz, tar.bz2; default 5 = 7-Zip "Normal"; 0 = store for zip). 7-Zip presets: 1 Fastest, 3 Fast, 5 Normal, 7 Maximum, 9 Ultra. For 7z it maps onto the LZMA2 preset

  -p, --password <PASSWORD>
          AES-256 password for zip / 7z (explicit intent to encrypt). For an ambient default applied only to encryptable formats, set AEROFTP_ARCHIVE_PASSWORD

      --encrypt-names
          7z only: also encrypt the archive header so the filenames are hidden, not just the file contents (7-Zip's "Encrypt file names", -mhe=on). Requires --password and is ignored for other formats. Off by default

      --method <METHOD>
          7z only: content compression method (default lzma2). lzma is the classic single-stream LZMA; ppmd suits text; bzip2 is block sorting. All are readable by the extractor. Ignored for other formats
          
          [possible values: lzma2, lzma, ppmd, bzip2]

      --dictionary <DICTIONARY>
          7z LZMA2 only: dictionary size in bytes (encoder clamps to 4096..=4 GiB), e.g. 16777216 for 16 MiB. Ignored for other methods/formats

      --solid
          7z only: pack every file into one solid block (better ratio on many small files, slower random extraction). Off by default

      --threads <THREADS>
          7z LZMA2 only: compression thread count (1 = single-threaded). Ignored for other methods/formats

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

connect

Categories: safe

Summary: Test connection

Description: Test connection to a remote server

Syntax: aeroftp-cli connect --profile NAME

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Test connection to a profile

aeroftp-cli connect --profile ftps-prodResolving profile 'ftps-prod'... OK
Connecting to ftps.example.com:21 (FTPS, explicit)...
TLS handshake: TLS 1.3, AES-256-GCM
Auth: aeroftp@aeroftp.app -> OK
Connection healthy (RTT 42 ms)

correct

Categories: advanced

Summary: Error-correct any standalone file with a detached `.aerocorrect` sidecar (par2-style). Generate parity, verify integrity, or repair corruption. The same format protects vault containers and synced fil

Description: Error-correct any standalone file with a detached `.aerocorrect` sidecar (par2-style). Generate parity, verify integrity, or repair corruption. The same format protects vault containers and synced files; binding is by content SHA-256. Windowed streaming keeps memory bounded on large files

Syntax: aeroftp-cli correct [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `correct` (offline help)

aeroftp-cli help correctError-correct any standalone file with a detached `.aerocorrect` sidecar (par2-style). Generate parity, verify integrity, or repair corruption. The same format protects vault containers and synced files; binding is by content SHA-256. Windowed streaming keeps memory bounded on large files

Usage: aeroftp-cli correct [OPTIONS] <COMMAND>

Commands:
  gen     Generate a `.aerocorrect` recovery sidecar for a file. Default output is `<file>.aerocorrect`. The level sets the storage overhead (low ~7%, medium ~15%, quartile ~25%, high ~30%, or a percentage 5-50; default medium)
  verify  Verify a file against its `.aerocorrect` sidecar (read-only, never modifies the file). Exit 0 = intact, 1 = corruption detected (repairable with `correct repair`)
  repair  Repair a corrupted file in place from its `.aerocorrect` sidecar. Atomic and all-or-nothing: the original is replaced only if the repaired stream hashes back to the sidecar's recorded content; otherwise it is left byte-for-byte untouched
  help    Print this message or the help of the given subcommand(s)

Options:
  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

cp

Categories: modify

Summary: Server-side copy when supported

Description: Copy a remote file on the server side when supported

Syntax: aeroftp-cli cp --profile NAME /old /new

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Server-side copy (no local hop)

aeroftp-cli cp --profile ftps-prod /docs/report.pdf /archive/2026/report.pdfServer-side COPY supported by ftps-prod
Copying /docs/report.pdf -> /archive/2026/report.pdf (245 KB)
OK - 0.18s (no local hop)

crypt

Categories: advanced

Summary: Use encrypted overlay storage (v4 keyslots: migrate-v4 / list-slots / add-slot / remove-slot / rotate-slot)

Description: Encrypted overlay - zero-knowledge storage on any provider

Syntax: aeroftp-cli crypt <init|bind|unbind|to-headed|to-headerless|migrate-marker|migrate-v4|list-slots|add-slot|remove-slot|rotate-slot|kit-verify|ls|put|get> --profile NAME /path

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Initialize an encrypted overlay

aeroftp-cli crypt init --profile filen-personal /vaultGenerating per-vault keys (XChaCha20-Poly1305)...
Encrypting metadata blob and uploading to /vault/.aeroftp-crypt
OK - overlay ready. Use `aeroftp-cli crypt put/get/ls --profile filen-personal /vault`.

cryptcheck

Categories: safe

Summary: Cryptcheck integrity against encrypted rclone remote

Description: Cryptcheck integrity against encrypted rclone remote

Syntax: aeroftp-cli cryptcheck [OPTIONS] [URL] [LOCAL] [REMOTE]

Options

• --password : Password for rclone crypt (can also be passed via AEROFTP_RCLONE_CRYPT_PASSWORD)
• --password2 : Salt for rclone crypt (can also be passed via AEROFTP_RCLONE_CRYPT_PASSWORD2)
• --filename-encryption : Filename encryption mode (default: standard)
• --suffix : Suffix on objects when --filename-encryption=off (rclone default ".bin"; use "none" to disable). Stripped before comparing names
• --one-way: Only check files present locally
• --checkfile : Optional checkfile mode (not fully implemented)
• -a, --algorithm : Hash algorithm to use (sha256 or md5) (default: sha256)
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Verify rclone-crypt remote integrity

aeroftp-cli cryptcheck --profile rclone-vault ./local /encryptedDecrypting filenames in /encrypted...
Comparing 842 objects (4.1 GB)
OK - all 842 files match plaintext on local

daemon

Categories: advanced

Summary: Manage the background jobs daemon

Description: Background daemon for persistent mounts, jobs, and watch

Syntax: aeroftp-cli daemon <start|stop|status>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Start the background jobs daemon

aeroftp-cli daemon startSpawning aeroftp daemon in /run/user/1000/aeroftp/
PID 41827  - state file: /run/user/1000/aeroftp/daemon.state
OK - daemon listening on unix:///run/user/1000/aeroftp/sock

dedupe

Categories: advanced

Summary: Resolve duplicate remote files

Description: Find duplicate files on a remote by content hash and optionally remove them

Syntax: aeroftp-cli dedupe --profile NAME /path [--mode MODE]

Options

• --mode : Resolution mode: skip, delete, newest, oldest, largest, smallest, rename, interactive, list (default: skip)
• --dry-run: Preview only (don't delete)
• --force: Confirm destructive modes (delete/newest/oldest/largest/smallest/rename). Required in non-interactive (non-TTY) runs; in a TTY it skips the prompt
• --max-delete : Cap on the number of files a destructive mode may delete/rename in one run (absolute N or N%). Defaults to 100 when omitted; raise it to act on more
• --similarity : Similarity: exact (byte-identical, default, back-compat) or non-identical (perceptual for images, simhash for text/SVG) (default: exact)
• --distance : Hamming distance threshold override for non-identical mode (raster default 10, text 3)
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Resolve duplicate remote files

aeroftp-cli dedupe --profile filen-personal /Photos --mode newestHashing /Photos recursively (sha256)...
Found 14 duplicate groups (412 MB redundant)
Mode 'newest': keeping the most recently modified copy
Removed 23 duplicate files - 412 MB freed

df

Categories: safe

Summary: Storage quota

Description: Show storage quota/usage

Syntax: aeroftp-cli df --profile NAME [--json]

Options

• --scan: Compute "used" by recursively scanning the tree (item 4b). Needed for backends with no quota API (FTP/S3/WebDAV): combined with an API or manual total it yields used/total/%. Never run automatically; this is the explicit opt-in
• --full: With --scan, scan from the account root instead of the profile's initial path. Slower; may hit the entry cap on large accounts
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Show storage quota

aeroftp-cli df --profile ftps-prodProvider:      Aruba Hosting (FTPS)
Quota:         5.0 GB
Used:          1.8 GB  (36%)
Free:          3.2 GB
Objects:       12 481

edit

Categories: modify

Summary: Replace text in a remote UTF-8 file

Description: Find and replace text in a remote UTF-8 file

Syntax: aeroftp-cli edit --profile NAME /path/file "find" "replace" [--first]

Options

• --first: Replace only the first occurrence
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Find & replace inside a remote text file

aeroftp-cli edit --profile ftps-prod /srv/site/version.txt "v3.8.4" "v3.8.5"Fetching /srv/site/version.txt (38 bytes)...
Match: 1 occurrence
Patched and re-uploaded in 0.22s
OK

export

Categories: modify

Summary: Export saved server profiles to external configuration formats

Description: Export saved server profiles to external configuration formats

Syntax: aeroftp-cli export [OPTIONS] <COMMAND>

Options

• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Export the vault as an encrypted bundle

aeroftp-cli export --out ./aeroftp-vault.aeroftp-keystoreSealing 7 profiles + 3 alias entries...
Wrote ./aeroftp-vault.aeroftp-keystore  (8.4 KB, AES-256-GCM)
Restore on another machine: `aeroftp-cli keystore restore ./aeroftp-vault.aeroftp-keystore`

extract

Categories: advanced

Summary: Extract an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar).

Description: Extract an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar).

Syntax: aeroftp-cli extract [OPTIONS] <ARCHIVE> [OUTDIR]

Options

• --archive-format : Archive format override (default: infer from the archive extension) [possible values: zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar]
• -p, --password : Password for encrypted zip / 7z. Also read from AEROFTP_ARCHIVE_PASSWORD
• --subfolder: Extract into a subfolder named after the archive
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `extract` (offline help)

aeroftp-cli help extractExtract an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar).

The format is inferred from the archive extension unless `--archive-format` is given. `--password` unlocks encrypted zip / 7z (also AEROFTP_ARCHIVE_PASSWORD). Local-only, no connection. Honors `--json`.

Usage: aeroftp-cli extract [OPTIONS] <ARCHIVE> [OUTDIR]

Arguments:
  <ARCHIVE>
          Archive path (format inferred from its extension unless --format)

  [OUTDIR]
          Output directory (default: current directory)
          
          [default: .]

Options:
      --archive-format <ARCHIVE_FORMAT>
          Archive format override (default: infer from the archive extension)
          
          [possible values: zip, 7z, tar, tar.gz, tar.xz, tar.bz2, gz, xz, bz2, rar]

  -p, --password <PASSWORD>
          Password for encrypted zip / 7z. Also read from AEROFTP_ARCHIVE_PASSWORD
          
          [env: AEROFTP_ARCHIVE_PASSWORD]

      --subfolder
          Extract into a subfolder named after the archive

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

find

Categories: safe

Summary: Search files

Description: Search for files by pattern

Syntax: aeroftp-cli find --profile NAME /path/ "*.ext" [--json]

Options

• --name : Glob pattern alias for the positional argument. When both the positional pattern and `--name` are set, `--name` wins
• --files-only: Match only files (skip directories)
• --dirs-only: Match only directories (skip files)
• --limit : Cap matches returned. JSON output sets `summary.truncated: true` when this trims results
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Find files by pattern

aeroftp-cli find --profile ftps-prod /srv/ --name "*.log"/srv/logs/access.log
/srv/logs/error.log
/srv/cache/build.log
3 matches in 0.81s

flatpak-import

Categories: advanced

Summary: Import a native install's configuration into this Flatpak sandbox.

Description: Import a native install's configuration into this Flatpak sandbox.

Syntax: aeroftp-cli flatpak-import [OPTIONS]

Options

• --status: Only report whether an import is available; do not apply it
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `flatpak-import` (offline help)

aeroftp-cli help flatpak-importImport a native install's configuration into this Flatpak sandbox.

A Flatpak install keeps its own sandboxed config, so it does not see the servers and encrypted vault of a native (.deb/.rpm/AppImage) install. Run inside a Flatpak with a native `~/.config/aeroftp` present, this copies it into the sandbox (copy-only, never overwriting). Outside a Flatpak it is a no-op. Restart AeroFTP afterwards to load the import.

Usage: aeroftp-cli flatpak-import [OPTIONS]

Options:
      --status
          Only report whether an import is available; do not apply it

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

get

Categories: safe

Summary: Download file

Description: Download file(s) from remote server

Syntax: aeroftp-cli get --profile NAME /remote/file [./local]

Options

• -r, --recursive: Recursive download (directories)
• --segments : Segmented parallel download: split file into N chunks (2-16, default: 1 = off) (default: 1)
• --delta: Z.4.5 R1: route the download through `AerorsyncDeltaTransport` when the provider exposes one (SFTP today). Auto-falls back to the classic transfer when the transport is not delta-eligible (file too small, missing host-key pin, password auth without dispatch wire-up, etc.). No-op for non-SFTP providers and for recursive / glob downloads
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Download a file

aeroftp-cli get --profile ftps-prod /docs/report.pdf ./Downloading /docs/report.pdf (245 KB)
[#####################] 100%  245 KB / 245 KB  412 KB/s
OK - saved to ./report.pdf in 0.6s

groups

Categories: advanced

Summary: List and manage server-profile groups (the My Servers group chips)

Description: List and manage server-profile groups (the My Servers group chips)

Syntax: aeroftp-cli groups [OPTIONS]

Options

• -i, --interactive: Drop into an interactive prompt after the table (rclone-config-style): re-index(#), Rename(R), Copy(C), Delete(D), List(L). Requires a TTY; ignored in JSON / non-interactive mode
• --tui: Open the inline action menu (TUI) directly, skipping the `-i` line prompt (Ehud #311). Implies interactive and needs a TTY; equivalent to running `-i` and immediately typing `tui`. Quitting the menu exits
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Inspect `groups` (offline help)

aeroftp-cli help groupsList and manage server-profile groups (the My Servers group chips)

Groups are flat, named labels on saved server profiles, the CLI analogue of the GUI group chips. Membership lives in the vault under `config_server_groups`, shared with the GUI, so changes round-trip both ways. Mirrors `aeroftp-cli profiles`: a plain listing, or `-i` for an interactive prompt (rename / copy / delete / re-index). Per Ehud #311.

Usage: aeroftp-cli groups [OPTIONS]

Options:
  -i, --interactive
          Drop into an interactive prompt after the table (rclone-config-style): re-index(#), Rename(R), Copy(C), Delete(D), List(L). Requires a TTY; ignored in JSON / non-interactive mode

      --tui
          Open the inline action menu (TUI) directly, skipping the `-i` line prompt (Ehud #311). Implies interactive and needs a TTY; equivalent to running `-i` and immediately typing `tui`. Quitting the menu exits

  -P, --profile <PROFILE>
          Use a saved server profile instead of URL (name or ID)

      --user <USER>
          Select an AeroFTP local user partition (or set AEROFTP_USER)
          
          [env: AEROFTP_USER=]

      --user-passphrase <USER_PASSPHRASE>
          Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
          
          [env: AEROFTP_USER_PASSPHRASE]

      --passphrase-file <PASSPHRASE_FILE>
          Read the AeroFTP user passphrase from the first line of a file

      --master-password <MASTER_PASSWORD>
          Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
          
          [env: AEROFTP_MASTER_PASSWORD]

      --limit-rate <LIMIT_RATE>
          Speed limit (e.g., "1M", "500K")

      --bwlimit <BWLIMIT>
          Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")

      --parallel <PARALLEL>
          Number of parallel transfer workers (default: 4, max: 32)
          
          [default: 4]

      --partial
          Resume interrupted transfers using partial files or remote offsets when supported

      --include <INCLUDE>
          Include only files matching glob pattern (repeatable)

      --exclude-global <EXCLUDE_GLOBAL>
          Exclude files matching glob pattern (repeatable)

      --include-from <INCLUDE_FROM>
          Read include patterns from file (one per line, # comments)

      --exclude-from <EXCLUDE_FROM>
          Read exclude patterns from file (one per line, # comments)

      --files-from <FILES_FROM>
          Read file list from file (one path per line). Only listed files are transferred

      --files-from-raw <FILES_FROM_RAW>
          Like --files-from but don't skip empty lines or strip whitespace

      --immutable
          Never overwrite existing files on destination (append-only / immutable mode)

      --no-check-dest
          Skip listing destination before transfer (assume dest is empty)

      --max-depth <MAX_DEPTH>
          Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r

      --min-size <MIN_SIZE>
          Minimum file size (e.g., "100k", "1M", "1G")

      --max-size <MAX_SIZE>
          Maximum file size (e.g., "100k", "1M", "1G")

      --min-age <MIN_AGE>
          Skip files newer than duration (e.g., "7d", "24h", "2w")

      --max-age <MAX_AGE>
          Skip files older than duration (e.g., "7d", "24h", "2w")

      --max-transfer <MAX_TRANSFER>
          Abort after transferring this many bytes total (e.g., "10G", "500M")

      --max-backlog <MAX_BACKLOG>
          Maximum number of queued transfer tasks (default: 10000)
          
          [default: 10000]

      --schedule <SCHEDULE>
          Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed)
          
          [default: fifo]
          [possible values: fifo, size]

      --retries <RETRIES>
          Number of retries for failed operations (default: 3, 0 = no retry)
          
          [default: 3]

      --retries-sleep <RETRIES_SLEEP>
          Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s")
          
          [default: 1s]

      --dump <DUMP>
          Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)

      --chunk-size <CHUNK_SIZE>
          Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3

      --buffer-size <BUFFER_SIZE>
          Override download buffer size (e.g., "256K", "1M")

      --multi-thread-streams <MULTI_THREAD_STREAMS>
          Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set
          
          [env: AEROFTP_MULTI_THREAD_STREAMS=]
          [default: 1]

      --multi-thread-cutoff <MULTI_THREAD_CUTOFF>
          Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set
          
          [env: AEROFTP_MULTI_THREAD_CUTOFF=]
          [default: 250M]

      --sftp-download-preset <SFTP_DOWNLOAD_PRESET>
          SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers

      --sftp-readahead <SFTP_READAHEAD>
          Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active

      --default-time <DEFAULT_TIME>
          Default mtime when backend returns None (ISO 8601 or "now")

      --transfer-engine <TRANSFER_ENGINE>
          Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE`
          
          [env: AEROFTP_TRANSFER_ENGINE=]
          [default: auto]
          [possible values: auto, dag, legacy]

      --fast-list
          Use recursive listing in a single API call (S3 only, faster for large datasets)

      --inplace
          Write downloads directly to final path (no .aerotmp temp file)

      --order-by <ORDER_BY>
          KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order)
          
          [default: none]
          [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc]

      --no-traverse
          KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)

      --tpslimit <TPSLIMIT>
          KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
          
          [env: AEROFTP_TPSLIMIT=]
          [default: 0]

      --tpslimit-burst <TPSLIMIT_BURST>
          KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
          
          [env: AEROFTP_TPSLIMIT_BURST=]
          [default: 0]

      --checkers <CHECKERS>
          KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool
          
          [default: 8]

      --sftp-concurrency <SFTP_CONCURRENCY>
          KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16.
          
          IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
          
          [env: AEROFTP_SFTP_CONCURRENCY=]
          [default: 0]

      --s3-upload-concurrency <S3_UPLOAD_CONCURRENCY>
          KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_UPLOAD_CONCURRENCY=]
          [default: 0]

      --s3-no-check-bucket
          KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_NO_CHECK_BUCKET=]

      --s3-disable-checksum
          KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
          
          [env: AEROFTP_S3_DISABLE_CHECKSUM=]

      --s3-acl <S3_ACL>
          KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
          
          [env: AEROFTP_S3_ACL=]

      --s3-storage-class <S3_STORAGE_CLASS>
          KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
          
          [env: AEROFTP_S3_STORAGE_CLASS=]

      --onedrive-no-versions
          KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
          
          [env: AEROFTP_ONEDRIVE_NO_VERSIONS=]

      --onedrive-list-chunk <ONEDRIVE_LIST_CHUNK>
          KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
          
          [env: AEROFTP_ONEDRIVE_LIST_CHUNK=]
          [default: 0]

      --onedrive-link-scope <ONEDRIVE_LINK_SCOPE>
          KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
          
          [env: AEROFTP_ONEDRIVE_LINK_SCOPE=]

      --azure-upload-concurrency <AZURE_UPLOAD_CONCURRENCY>
          KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_UPLOAD_CONCURRENCY=]
          [default: 0]

      --azure-disable-checksum
          KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_DISABLE_CHECKSUM=]

      --azure-access-tier <AZURE_ACCESS_TIER>
          KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ACCESS_TIER=]

      --azure-archive-tier-delete
          KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
          
          [env: AEROFTP_AZURE_ARCHIVE_TIER_DELETE=]

      --drive-cross-account-copy
          KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_CROSS_ACCOUNT_COPY=]

      --drive-pacer-burst <DRIVE_PACER_BURST>
          KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
          
          [env: AEROFTP_DRIVE_PACER_BURST=]
          [default: 0]

      --drive-pacer-min-sleep <DRIVE_PACER_MIN_SLEEP>
          KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
          
          [env: AEROFTP_DRIVE_PACER_MIN_SLEEP=]

      --drive-acknowledge-abuse
          KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
          
          [env: AEROFTP_DRIVE_ACKNOWLEDGE_ABUSE=]

      --aimd-hint <AIMD_HINT>
          KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
          
          [env: AEROFTP_AIMD_HINT=]

      --aimd-disable
          KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
          
          [env: AEROFTP_AIMD_DISABLE=]

      --aimd-min-window <AIMD_MIN_WINDOW>
          KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_MIN_WINDOW=]

      --aimd-max-window <AIMD_MAX_WINDOW>
          KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
          
          [env: AEROFTP_AIMD_MAX_WINDOW=]

      --aimd-step-window <AIMD_STEP_WINDOW>
          KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
          
          [env: AEROFTP_AIMD_STEP_WINDOW=]

      --aimd-config <PATH>
          KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class.
          
          File schema: [aimd.file]   min = 1   max = 8   step = 2 [aimd.chunk]  min = 4   max = 32 [aimd.http]   ... [aimd.api]    ... Every section and every key is optional.
          
          [env: AEROFTP_AIMD_CONFIG=]

  -h, --help
          Print help (see a summary with '-h')

Output options:
      --format <FORMAT>
          Output format
          
          [default: text]
          [possible values: text, json]

      --json
          Shorthand for --format json

      --machine
          Machine / agent mode (implies --json --no-banner --quiet). Guarantees that stdout contains *only* structured result data. All human diagnostics, "Using profile", path notes, connection chatter, "Connected successfully", "Next: ..." etc. go to stderr (or are suppressed). This is the recommended flag for robust agent/LLM/tool-calling use. (also via AEROFTP_MACHINE env var)

      --no-banner
          Suppress the startup banner (also via AEROFTP_NO_BANNER env var)

      --json-fields <JSON_FIELDS>
          Restrict JSON output fields (comma-separated, e.g. name,size,modified)

  -v, --verbose...
          Verbose output (-v debug, -vv trace)

  -q, --quiet
          Quiet mode (errors only)

Connection options:
      --password-stdin
          Read password from stdin (pipe: echo "pass" | aeroftp ...)

      --key <KEY>
          SSH private key path for SFTP

      --key-passphrase <KEY_PASSPHRASE>
          SSH key passphrase

      --bucket <BUCKET>
          S3 bucket name

      --region <REGION>
          S3/Azure region

      --container <CONTAINER>
          Azure container name

      --token <TOKEN>
          Bearer/API token (kDrive, Jottacloud, FileLu)
          
          [env: AEROFTP_TOKEN]

      --tls <TLS>
          FTP TLS mode: none, explicit, implicit, explicit_if_available

      --insecure
          Skip TLS certificate verification

      --trust-host-key
          Trust unknown SSH host keys (skip TOFU verification)

      --two-factor <TWO_FACTOR>
          2FA code (Filen, Internxt)
          
          [env: AEROFTP_2FA]

      --no-crypt
          Connect to a crypt-overlay profile RAW (skip the bound Crypt overlay), mirroring the GUI's Crypt On/Off toggle. On-wire names and content stay encrypted; no decrypt/encrypt is applied. Ignored for non-crypt profiles

Safety options:
      --strict
          Refuse safety-relaxing flags (also via AEROFTP_STRICT=1).
          
          Hardens unattended / agent automation: any flag that disables a safety check (TLS or SSH host-key verification, the abuse / cross-account / archive-tier acknowledgements, AIMD backpressure) or auto-approves write/destructive agent tools becomes a hard error (exit 5) instead of silently relaxing safety. A generated command that smuggles in `--insecure` or `--auto-approve all` then fails loudly rather than proceeding.

hashsum

Categories: advanced

Summary: Compute remote checksum

Description: Compute hash of remote file(s)

Syntax: aeroftp-cli hashsum --algorithm ALGO --profile NAME /path/file

Options

• -a, --algorithm : Hash algorithm (md5/sha1/sha256/sha512/blake3). Defaults to sha256. Accepts `-a sha256` or `--algorithm sha256`. Omitting the flag yields a sha256 checksum Possible values: - md5 - sha1 - sha256 - sha512 - blake3 - quickxor: Microsoft OneDrive proprietary QuickXorHash. Server-side only: it cannot be computed locally, so there is no download fallback - dropbox: Dropbox content_hash (SHA-256 of per-4MiB-block SHA-256). Server-side only: not a plain digest, no local fallback (default: sha256)
• --download: Download and hash locally
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')

Examples

Compute SHA-256 of a remote file

aeroftp-cli hashsum --algorithm sha256 --profile ftps-prod /docs/report.pdfComputing sha256 of /docs/report.pdf...
9f7c8a3b1d5e2f4a6c8b0d2e4f6a8c0b2d4e6f8a0c2b4d6e8f0a2c4b6d8e0f2c  /docs/report.pdf

head

Categories: advanced

Summary: Read first lines of a remote text file

Description: Print first N lines of a remote file

Syntax: aeroftp-cli head --profile NAME /path/file [-n N]

Options

• -n, --lines : Number of lines to print (default: 20) (default: 20)
• -c, --bytes : Print the first N bytes instead of N lines. Mutually exclusive with --lines (--bytes wins when both present). Useful for "first 4KB" previews on binary or huge files
• -P, --profile : Use a saved server profile instead of URL (name or ID)
• --user : Select an AeroFTP local user partition (or set AEROFTP_USER)
• --user-passphrase : Account passphrase for the selected AeroFTP user (prefer env/file over CLI args)
• --passphrase-file : Read the AeroFTP user passphrase from the first line of a file
• --master-password : Master password for encrypted vault (or set AEROFTP_MASTER_PASSWORD)
• --limit-rate : Speed limit (e.g., "1M", "500K")
• --bwlimit : Bandwidth schedule (e.g., "08:00,512k 12:00,10M 18:00,off")
• --parallel : Number of parallel transfer workers (default: 4, max: 32) (default: 4)
• --partial: Resume interrupted transfers using partial files or remote offsets when supported
• --include : Include only files matching glob pattern (repeatable)
• --exclude-global : Exclude files matching glob pattern (repeatable)
• --include-from : Read include patterns from file (one per line, # comments)
• --exclude-from : Read exclude patterns from file (one per line, # comments)
• --files-from : Read file list from file (one path per line). Only listed files are transferred
• --files-from-raw : Like --files-from but don't skip empty lines or strip whitespace
• --immutable: Never overwrite existing files on destination (append-only / immutable mode)
• --no-check-dest: Skip listing destination before transfer (assume dest is empty)
• --max-depth : Maximum directory recursion depth (default: unlimited). Applies to ls -R, find, sync, get -r
• --min-size : Minimum file size (e.g., "100k", "1M", "1G")
• --max-size : Maximum file size (e.g., "100k", "1M", "1G")
• --min-age : Skip files newer than duration (e.g., "7d", "24h", "2w")
• --max-age : Skip files older than duration (e.g., "7d", "24h", "2w")
• --max-transfer : Abort after transferring this many bytes total (e.g., "10G", "500M")
• --max-backlog : Maximum number of queued transfer tasks (default: 10000) (default: 10000)
• --schedule : Multi-file admission schedule for the streaming transfer frontier (DAG-P2-08). `fifo` (default) preserves arrival order; `size` prefers small files for low latency with aging so large files never starve (rclone-style size,mixed) [possible values: fifo, size] (default: fifo)
• --retries : Number of retries for failed operations (default: 3, 0 = no retry) (default: 3)
• --retries-sleep : Delay between retries (e.g., "5s", "1m", "500ms"; default: "1s") (default: 1s)
• --dump : Dump HTTP debug info to stderr (comma-separated: headers,bodies,auth)
• --chunk-size : Override upload chunk size (e.g., "64M", "16M"). Min 5M for S3
• --buffer-size : Override download buffer size (e.g., "256K", "1M")
• --multi-thread-streams : Number of concurrent Range streams per single-file download (rclone `--multi-thread-streams`). Default 1 = disabled. Range 1-16. Honored by S3/Azure (native) and, after a live strict 206 probe, by WebDAV (Basic/anonymous auth; Digest excluded) and Koofr. Any backend or server that does not prove Range honesty falls back to single-stream transparently. Reads default from `AEROFTP_MULTI_THREAD_STREAMS` if set (default: 1)
• --multi-thread-cutoff : Minimum file size for multi-thread download (rclone `--multi-thread-cutoff`). Default `250M`. Accepts size suffixes `K`/`M`/`G`. Files smaller than this always use the single-stream path, regardless of `--multi-thread-streams`. Reads default from `AEROFTP_MULTI_THREAD_CUTOFF` if set (default: 250M)
• --sftp-download-preset : SFTP single-file download tuning preset. Presets configure independent SSH connections, read-ahead, and the multi-connection cutoff together. `efficient` matches the GUI default. The option is ignored for non-SFTP providers
• --sftp-readahead : Explicit SFTP read-ahead window. Overrides `--sftp-download-preset`. Range 2-1024. When neither CLI option is supplied, the legacy `AEROFTP_SFTP_READAHEAD` provider fallback remains active
• --default-time : Default mtime when backend returns None (ISO 8601 or "now")
• --transfer-engine : Single-file transfer engine selection. `auto` consults the data-driven router (default), `dag` forces every transfer through the shaped-graph DAG engine, `legacy` forces the provider-direct path. The router is populated from Phase A benchmark measurements (see `docs/dev/benchmarks/2026-05-2{4,5}_phaseA-*/`) and revalidated by T-DEBT-RTR-04. The default is DAG everywhere; `legacy` remains an operator override. Reads default from `AEROFTP_TRANSFER_ENGINE` [possible values: auto, dag, legacy] (default: auto)
• --fast-list: Use recursive listing in a single API call (S3 only, faster for large datasets)
• --inplace: Write downloads directly to final path (no .aerotmp temp file)
• --order-by : KE-A5: Order the transfer queue by the given key (rclone `--order-by`). Applies to multi-file operations: sync, recursive get/put, dedupe. Single-file operations and `ls` are unaffected (`ls` keeps its own `--sort`). Defaults to `none` (FIFO from the input order) [possible values: none, name, name-desc, size-asc, size-desc, modtime-asc, modtime-desc] (default: none)
• --no-traverse: KE-A4: Skip the full remote listing during sync but still stat every candidate upload to avoid overwriting files that already match by size (rclone `--no-traverse`). Use when the source has a small number of files and the destination has millions (typical S3 bucket with a long-running history). Implies `--direction upload`; combine with `--no-check-dest` for the fully-trust-me variant (skip the stat too)
• --tpslimit : KE-A3: Hard cap on HTTP transactions per second across all providers (rclone `--tpslimit`). Layered ABOVE AIMD: AIMD may shrink the effective parallelism well below this cap on 429s, but never exceeds it. Default `0` = unlimited (AIMD remains the sole throttle). Honored by every backend that goes through the shared `send_with_retry` wrapper (S3, B2, Azure, all OAuth clouds, Filen, WebDAV, ImageKit, Drime, Uploadcare, Cloudinary). SFTP / FTP session protocols are unaffected. Reads default from `AEROFTP_TPSLIMIT`
• --tpslimit-burst : KE-A3: Burst capacity for `--tpslimit`. Lets the limiter absorb short spikes without queueing: e.g. `--tpslimit 10 --tpslimit-burst 50` allows up to 50 requests back-to-back before throttling kicks in at 10/s. Default `0` = same as `--tpslimit` (no burst). Reads default from `AEROFTP_TPSLIMIT_BURST`
• --checkers : KE-A2: Concurrency cap for metadata-only operations during sync (rclone `--checkers`). Default `8`, range 1-64. NOTE: today this only bounds and labels the `--no-traverse` stat sweep, which still runs sequentially: `StorageProvider::stat` takes `&mut self`, so a single provider handle serialises the probes. The value is honoured semantically (reported in the sweep's summary note) but does not yet parallelise anything; it becomes a real concurrency gate with the provider pool refactor (Sprint K2). Distinct from `--transfers`, which caps the actual data-transfer pool (default: 8)
• --sftp-concurrency : KE-A1: SFTP-specific concurrency override for single-file downloads (rclone `--sftp-concurrency`). When set and the remote is SFTP, takes precedence over `--multi-thread-streams` for SFTP only. Default `0` = follow `--multi-thread-streams`. Range 1-16. IMPORTANT: AeroFTP implements SFTP "concurrency" as N independent SSH connections (PD-SFTP-1 / PD-SFTP-2 pattern), NOT as N pipelined SSH_FXP_READ on a single channel like rclone. The semantics differ: AeroFTP pays N handshakes once per transfer and gets fully independent streams (incl. on servers that ignore SSH window scaling); rclone uses a single channel with deeper read queue depth (lower handshake cost, bounded by server window-scaling honesty). On 1 Gbit links to modern OpenSSH the two converge within ~5%; on links with degraded window scaling AeroFTP's N-connections wins. Reads default from `AEROFTP_SFTP_CONCURRENCY`.
• --s3-upload-concurrency : KE-B1.1: Override S3 multipart upload parallelism (rclone `--s3-upload-concurrency`). Caps how many parts are uploaded concurrently per file. Default `0` = use the built-in 4-in-flight ceiling. Range 1-64. Memory usage scales with `--s3-upload-concurrency * <part-size>`; raising both at once on small VMs can OOM. Silently ignored when the remote is not S3
• --s3-no-check-bucket: KE-B1.2: Skip the S3 bucket-existence probe on `connect()` (rclone `--s3-no-check-bucket`). Use with IAM credentials that grant `PutObject` but deny `ListBucket`: the default probe would 403 even though uploads work. Silently ignored when the remote is not S3
• --s3-disable-checksum: KE-B1.3: Replace the per-payload SHA-256 in S3 SigV4 with `UNSIGNED-PAYLOAD` (rclone `--s3-disable-checksum`). Reduces client CPU on large multipart uploads (~5-15% on a CPU-bound producer) at the cost of SigV4 in-flight tamper protection. Empty-body requests still hash normally (constant cost, some gateways require it). Silently ignored when the remote is not S3
• --s3-acl : KE-B1.4: Canned ACL applied to S3 uploads (rclone `--s3-acl`). Common values: `private`, `public-read`, `public-read-write`, `authenticated-read`, `bucket-owner-read`, `bucket-owner-full-control`. Emitted as `x-amz-acl`. Validation is permissive (vendor extensions accepted; AWS rejects unknown values at the API level). Silently ignored when the remote is not S3, or when the bucket is governed by a bucket policy that forbids ACLs (AWS default since April 2023). Reads default from `AEROFTP_S3_ACL`
• --s3-storage-class : KE-B1.5: Storage class for S3 uploads (rclone `--s3-storage-class`). Common AWS values: `STANDARD`, `STANDARD_IA`, `ONEZONE_IA`, `INTELLIGENT_TIERING`, `GLACIER_IR`, `GLACIER`, `DEEP_ARCHIVE`, `REDUCED_REDUNDANCY`, `EXPRESS_ONEZONE`. Vendor-specific tiers (B2 / Wasabi / R2) accepted. Takes precedence over the profile-level `storage_class` setting. Emitted as `x-amz-storage-class`. Silently ignored when the remote is not S3. Reads default from `AEROFTP_S3_STORAGE_CLASS`
• --onedrive-no-versions: KE-B3.1: After every successful OneDrive upload, sweep all previous versions of the resulting drive item (rclone `--onedrive-no-versions`). Mitigates the OneDrive Personal "quota inflation from auto-versioning" issue where every modify spawns a new version that counts against the user's storage quota. The sweep is best-effort: failures are logged but never propagated, so the upload itself still reports success. Silently ignored when the remote is not OneDrive
• --onedrive-list-chunk : KE-B3.2: Override the Microsoft Graph `$top` paging size for `/children` listings (rclone `--onedrive-list-chunk`). Server default is ~200. Bigger pages reduce round trips on large folders; smaller pages help under throttling. Clamped to `[1, 999]`. `0` = leave server default. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LIST_CHUNK`
• --onedrive-link-scope : KE-B3.3: Override the share-link `scope` on `createLink` calls (rclone `--onedrive-link-scope`). Typical values: `anonymous` (default), `organization`, `users`. Validation is permissive; the Graph API rejects unknown scopes at request time. Silently ignored when the remote is not OneDrive. Reads default from `AEROFTP_ONEDRIVE_LINK_SCOPE`
• --azure-upload-concurrency : KE-B4.1: Override the number of `Put Block` requests in flight during Azure block uploads (rclone `--azureblob-upload-concurrency`). Default `0` = serial (1 block at a time, historical behaviour). Range 1-32. Memory usage scales with `concurrency * 4 MiB` per upload; raising on small VMs can OOM. Silently ignored when the remote is not Azure Blob
• --azure-disable-checksum: KE-B4.2: Reserved for future Content-MD5 metadata gating on Azure uploads (rclone `--azureblob-disable-checksum`). AeroFTP does not currently send `x-ms-blob-content-md5` on uploads, so this flag is structurally wired but has no observable effect today. It will gate the future MD5-on-upload code path once that lands. Silently ignored when the remote is not Azure Blob
• --azure-access-tier : KE-B4.3: Apply this Azure access tier to every successful upload as a post-PUT `Set Blob Tier` call (rclone `--azureblob-access-tier`). Values: `Hot`, `Cool`, `Cold`, `Archive`. Tier failures are logged at WARN but do not invalidate the upload itself. Vendor / future tiers pass through; Azure rejects unknown values at the API level. Silently ignored when the remote is not Azure Blob
• --azure-archive-tier-delete: KE-B4.4: Before overwriting an existing blob, check if it is in Archive access tier; if so, DELETE the blob before the new PUT (rclone `--azureblob-archive-tier-delete`). Without this flag, PUT against an Archive blob fails with `BlobArchived`. Adds one pre-upload HEAD round trip per file. Silently ignored when the remote is not Azure Blob
• --drive-cross-account-copy: KE-B2.3: Append `supportsAllDrives=true` to Google Drive `files.copy` so the server-side copy succeeds across Shared Drives and between My Drive and a Shared Drive. Without this flag, copies that cross drive boundaries can fail with 404 / 403 even when the OAuth scopes are sufficient. Aligned with rclone's `--drive-server-side-across-configs`. Silently ignored when the remote is not Google Drive
• --drive-pacer-burst : KE-B2.1: Allow this many Google Drive API dispatches inside one AIMD pacing window before `--drive-pacer-min-sleep` is applied. `0` preserves the controller default. Ignored for non-Drive providers
• --drive-pacer-min-sleep : KE-B2.2: Minimum delay between Google Drive AIMD API pacing windows. Accepts `500ms`, `2s`, `1m`, `1h`, or `0` to disable the overlay and preserve the controller default
• --drive-acknowledge-abuse: KE-B2.4: Append `acknowledgeAbuse=true` to Google Drive binary download URLs (`alt=media`) so files flagged by Google as abusive still download instead of failing with `cannotDownloadAbusiveFile` (rclone `--drive-acknowledge-abuse`). Workspace export endpoints are not affected. By enabling this flag the user acknowledges the risk associated with downloading content Google's classifier flagged. Silently ignored when the remote is not Google Drive
• --aimd-hint : KE-D3: Static Retry-After fallback overlay for the AIMD controller. Repeat as `--aimd-hint=provider:secs`, for example `--aimd-hint=drive:30 --aimd-hint=s3:5`
• --aimd-disable: KE-D1: Disable the AIMD backpressure controller for the whole run. When set, dispatch concurrency stays pinned at the honest per-class ceiling: a 429 / 503 / timeout response no longer halves the window and a healthy completion no longer regrows it. Intended for dedicated/lab links where the operator has out-of-band guarantees the backend will not throttle and AIMD would only mask a genuine throughput cap diagnostic
• --aimd-min-window : KE-D2: Floor for the AIMD `target` after multiplicative decrease. Broadcast to every controlled class (file / chunk / http / api). Use the TOML file (`--aimd-config`) for per-class overrides. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-max-window : KE-D2: Ceiling cap for the AIMD `target`. Broadcast to every controlled class. The value clamps the budget-derived ceiling from below: it can only shrink the honest cap, never raise it (AIMD is decrease-biased). Default: `ceiling` (no extra clamp)
• --aimd-step-window : KE-D2: Additive step applied to the AIMD `target` once per quiet `healthy_window`. Broadcast to every controlled class. A value of `0` is sanitized to `1`. Default: `1`
• --aimd-config : KE-D2: Path to a TOML file with per-class AIMD window overrides. When unset, AeroFTP looks at `$XDG_CONFIG_HOME/aeroftp/aimd.toml` (or `~/.config/aeroftp/aimd.toml`) and loads it silently if it exists. CLI broadcast flags (`--aimd-min-window`, etc.) take precedence over file values per class. File schema: [aimd.file] min = 1 max = 8 step = 2 [aimd.chunk] min = 4 max = 32 [aimd.http] ... [aimd.api] ... Every section and every key is optional.
• -h, --help: Print help (see a summary with '-h')