Why run a container registry when you already have a S3?

grue is a CLI that turns any S3-compatible bucket into a container registry. It pushes and pulls Docker images directly to and from object storage. You can use Cloudflare R2, AWS S3, MinIO, Backblaze B2, or another S3-compatible store. Authentication uses your S3 API keys.

The whole program is one file, grue.go, and under 1000 lines of code.

• grue connect - configure an S3-compatible storage backend
• grue push <name[:tag]> - push a local image to the bucket
• grue pull <name[:tag]> - pull an image from the bucket into Docker
• grue ls - list repositories and tags
• grue rm <name[:tag]> - delete a tag, or all tags of a repository

# Connect to S3
grue connect

# Push a local container image to your bucket
grue push my-image:1.2.3

# On another machine, pull it back
grue pull my-image:1.2.3

# The image appears in Docker
docker run my-image:1.2.3

Use env secrets:

GRUE_ACCESS_KEY=... grue push my-image:1.2.3

List what is in the bucket:

grue ls

Delete a tag:

grue rm my-image:old

Delete all tags of a repository:

grue rm my-image

Environment variables take priority over the config file.

The config file is at ~/.config/grue/config.json by default. Override the path with GRUE_CONFIG.

The grue action builds a Docker image and pushes it to your S3 bucket. You do not need a container registry.

Replace docker/build-push-action with the grue action. Then add your S3 credentials.

Before — push to a registry:

- uses: docker/build-push-action@v7
  with:
    context: .
    tags: myimage:latest
    push: true

After — push to S3:

- uses: Simple-Observability/grue@v1.0.0
  with:
    context: .
    tags: myimage:latest
    endpoint: ${{ secrets.GRUE_ENDPOINT }}
    bucket: ${{ secrets.GRUE_BUCKET }}
    access-key: ${{ secrets.GRUE_ACCESS_KEY }}
    secret-key: ${{ secrets.GRUE_SECRET_KEY }}

The grue action uses docker/build-push-action to build the image. All build inputs (build-args, target, cache-from, cache-to, platform, and more) are sent to it directly. See their inputs list for the full set.

These inputs control the S3 connection and the grue binary. Build inputs are the same as docker/build-push-action.

• No blob garbage collection. rm deletes tags only; the blobs they referenced are retained. A future gc/prune command is planned.
• One platform per tag. The action builds one platform at a time. grue stores one image per tag. There is no multi-arch index.