One binary. Zero servers.
Git for cloud object storage.
A single Rust binary that acts as both a Git remote helper and a filter driver. Push and pull repositories backed by S3, GCS, or Azure with standard Git commands — no servers, no LFS endpoints, no infrastructure to manage.
Interactive Shell Sandbox
Experience the Crab CLI Workflow
Select a common developer command to see how the single Rust binary chunk-deduplicates, parallel-uploads, and lazy-mounts files straight to your cloud store.
Connect your local Git repository to a serverless S3, GCS, or Azure bucket. Auto-detects large files and configures the filter driver.
Five Stages, From Working Tree to Cloud
Hover or focus a step to see what happens under the hood.
- Gearhash content-defined chunking splits files into variable-size chunks at SIMD speed (~500 MB/s), resilient to insertions and deletions.
- Each chunk is classified A (already in DB), B (in shard), or C (new). Only C-class chunks travel to the cloud — saving bandwidth and storage.
- New chunks are concatenated and compressed into ~64 MiB xorb objects so a few large PUTs replace thousands of tiny ones.
- Xorbs and shards stream to S3, GCS, or Azure over a parallel transfer pool with checkpoint journaling — interrupted pushes resume cleanly.
- After all data lands, the ref is advanced via a compare-and-swap on the manifest object so concurrent pushes never race past each other.
Tab through the steps with your keyboard for the same details.
Built for Scale, Tuned for Speed
Numbers from the shipping CLI on commodity hardware.
- 0+ MB/sChunking throughput
Gearhash CDC on a single core, SIMD-accelerated.
- 0×Upload parallelism
Xorbs stream concurrently to the object store.
- 0%Typical dedup ratio
Bytes saved on incremental pushes of large repos.
- 0 MiBXorb pack size
Many tiny PUTs collapsed into a few large ones.
Technical Differentiators
Purpose-built for large-file workflows at scale.
Push Pipeline
From local files to cloud storage in five stages.
How Crab Compares to Git LFS and DVC
Eight feature dimensions, side by side.
| Feature | Crab | Git LFS | DVC |
|---|---|---|---|
| No server required | |||
| Deduplication method | Content-defined chunking (gearhash) | None (whole-file) | Whole-file content hashing |
| Maximum file size | TB-scale (chunked) | 5 GB (GitHub), server-dependent | Bucket-limited |
| Lazy checkout / partial materialization | Include/exclude patterns only | ||
| Resumable uploads | |||
| Cloud provider support | S3, GCS, Azure | Self-hosted or SaaS LFS server | S3, GCS, Azure, SSH, HDFS, local |
| Standard Git UX | Parallel CLI (dvc push/pull) | ||
| Chunk-level delta transfer |
Supported Cloud Providers
Works with the major cloud object storage providers out of the box.
Install Crab on Your Platform
One binary, copy-paste install. Pick your platform below.
# Install via Homebrew$ brew install crab-project/tap/crab # Confirm the install$ crab --versioncrab 0.6.0 (macos-arm64)crab binary and the git-remote-crab symlink in one step.Ready to ditch your LFS server?
Start pushing large files to cloud storage with standard Git commands. No infrastructure required.