Clipper Documentation
Clipper is a container registry designed for large images. It deduplicates at the file level so transfers are fast, even for multi-gigabyte images.
The problem
Container images are transferred as whole layers. If you change one file in a 2GB layer, the entire layer is re-uploaded and re-downloaded.
Pulling an updated tag should be fast when only a few files changed. But traditional registries don’t know what changed - they re-transfer the entire layer.
For images with large dependencies, ML frameworks, or model weights, this adds up quickly.
How Clipper is different
Clipper breaks layers into files and tracks content by hash. Pushes and pulls only transfer files that are new or changed.
- Patch pulls: pulling an updated tag only downloads the files that differ from what you already have
- Fast pushes: updating a large image transfers only the changed files, not entire layers
- Deduplication: files shared across images are stored and transferred once
Clipper is a drop-in addition to your workflow. Your images stay compatible with Docker, containerd, and Podman.