Reference
crab download
Download selected files from a Crab repository without creating a Git clone.
crab get is an alias.
Synopsis
crab download [OPTIONS] <REPO> [PATHS]...
crab get [OPTIONS] <REPO> [PATHS]...<REPO> can be a Crab remote URL or local repository path. Positional paths
select exact repository-relative files or subtrees ending in /. Use
--include and --exclude for glob selection.
By default, files are written under Crab's revision-scoped download cache. Use
--local-dir to materialize them under a directory you choose.
Options
| Option | Description |
|---|---|
--revision <REVISION> | Read a branch, tag, ref, or full commit SHA |
--include <PATTERN> | Include a glob; repeat as needed |
--exclude <PATTERN> | Exclude a glob; repeat as needed |
--cache-dir <DIR> | Override the Crab cache root |
--local-dir <DIR> | Write selected files under this directory |
--force-download | Download even when a destination is fresh |
--dry-run | Preview the selection without writing files or metadata |
--max-workers <N> | Set file-level download concurrency |
--quiet | Suppress progress and summaries while printing final paths |
--json | Emit one structured JSON result |
--jsonl | Emit streaming JSONL output |
Examples
Download two exact files into a local directory:
crab download crab://bucket/models README.md config.json --local-dir ./exportSelect a subtree at a specific revision:
crab get crab://bucket/models models/ --revision refs/tags/v1 --local-dir ./modelsPreview a glob selection:
crab download crab://bucket/models --include '*.safetensors' --dry-run --jsonRelated commands
crab clone— create a Git checkout.crab hydrate— materialize pointers in an existing checkout.