crab ship

Ship files to cloud storage in a single command.

Synopsis

crab ship [OPTIONS] -m <MESSAGE> <PATTERNS>...

Description

crab ship combines crab add, git commit, and crab push into a single command. It's designed for ML/data workflows where you want files staged, committed, and uploaded without running three separate commands.

Under the hood, crab ship performs these steps:

1. Stage — Runs crab add with the given patterns (parallel chunking + dedup)
2. Commit — Creates a git commit with the provided message
3. Push — Runs the native crab push pipeline (concurrent xorb uploads, faster than git push)

If no tracking patterns exist in .gitattributes, crab ship auto-detects large file extensions and tracks them before staging — the same auto-track behavior as crab add.

Arguments

Options

Examples

Ship model files

crab ship '*.safetensors' -m "fine-tuned model v2"

Ship everything

crab ship . -m "weekly checkpoint"

Ship without pushing (local commit only)

crab ship '*.bin' -m "WIP: training in progress" --no-push

Ship with higher parallelism

crab ship -j 16 'data/**' -m "updated training data"

Performance

crab ship uses the native crab push pipeline rather than git push. This means xorb uploads happen concurrently (default 16 parallel uploads) and the push bypasses git's serial remote helper protocol. For repositories with many large files, this can be significantly faster than the equivalent crab add && git commit && git push sequence.

When to Use crab ship vs Individual Commands

Error Handling

• If crab add fails (no tracked patterns, no matching files), the command stops before committing
• If there's nothing to commit (files already up to date), the command reports this and exits cleanly
• If crab push fails (network error, auth issue), the local commit is preserved — re-run crab push to retry

Related Commands

• crab add — stage files without committing
• crab push — native concurrent push (standalone)
• crab init — initialize a repository