Clipper

1 min read Original article ↗

The world's fastest
container registry

Chunk-level deduplication across layers, tags, and repos.
Reuse the data you already have, only transfer what's new.

Blazing fast patch pulls

Clipper reuses the data your device already has on it. Adding a single dependency or changing a few lines of Python is no problem. Ever tried to pull a large image over cell service on a farm? DSL in a warehouse? It could take hours. Clipper does what Docker can't.

90% less data transferred

29m 4s faster over 3G

Dial-up 48 Kbps3G 4 MbpsLTE 48 MbpsBroadband 200 Mbps

Drop-in compatible

Switching is easy.

Pulled images can be run by docker, podman, and containerd with no modification. Clipper handles chunking and dedup transparently.

terminal

$ docker push myimage:latest^C

$ clipper push myimage:latest acme/myimage:latest

uploaded 125 blobs (823.9 MB) in 1m1.348s, skipped 0

pushing manifest to clipper.dev/acme/myimage:latest ...

done

$ docker run clipper.dev/acme/myimage:latest

Built for

Robotics

OTA updates that only transfer what changed.

Local chunk cache on every device. Pulls skip bytes already on disk.

Large images

A registry built for multi-gigabyte images.

CUDA runtimes, model weights, datasets. Purpose-built for the images Docker wasn't designed for.

Inference

Deploy new model versions to a warm fleet.

Most chunks are already on disk from the last version.