A serverless Docker image registry backed by Databricks Unity Catalog Volumes. No registry server, and no extra infrastructure.

If you already run Databricks, you already have a governed, access-controlled, versioned storage layer sitting in your workspace. This turns that into a private container registry with three commands: push, pull, run.

This is an idea that started out as a Proof of Concept.

Also, if you're already a Databricks shop, Unity Catalog Volumes give you something a purpose-built registry would charge you extra for:

• Unified governance - The same Unity Catalog permissions (Catalog/Schema/Volume), that guard your tables and files now guard your container images. No separate registry IAM to maintain.
• Zero new infrastructure - No registry server to deploy, patch, or scale. Volumes are already highly available, durable object storage under the hood.
• One audit trail - Image pushes and pulls show up alongside other Unity Catalog access logs.
• Cost simplicity - You pay for storage you likely already provision, not a second storage bill for a dedicated registry.
• Team-native - Data and ML engineers already have Databricks credentials, there's no separate registry login to distribute or rotate.

This project is intentionally simple, it doesn't try to reimplement the OCI Distribution Spec. It saves images as compressed tarballs (docker save -> gzip) and streams them to and from a Volume path using the Databricks Files API. Think of it as docker save/docker load, automated, with your Volume as the shared drop point.

• Push: docker save -> gzip-compress with a live progress bar and SHA-256 checksum → stream directly to the Volume via PUT.
• Pull: stream the object back with GET -> decompress -> docker load into your local daemon.
• List: reads the Volume directory listing and reconstructs image name/tag pairs from the stored filenames.
• Run: pulls automatically if the image isn't cached locally, then hands off to docker run.

Image references are stored as <name>__<tag>.tar.gz, so nginx:alpine becomes nginx__alpine.tar.gz inside your configured Volume path.

• Python 3.9+
• Docker installed and running locally
• A Databricks workspace with a Unity Catalog Volume you have read/write access to
• A Databricks personal access token (or other bearer token) with Files API permissions on that Volume

Please refer to the INSTALL.md file for detailed setup instructions and environment variable configuration.

dcip push nginx:alpine

Saves, compresses, and uploads the image to your configured Volume.

dcip pull nginx:alpine

Downloads, decompresses, and loads the image into your local Docker daemon.

dcip list

Prints a table of every image name, tag, size, and last-modified time stored in the Volume.

dcip run nginx:alpine --name my-nginx -p 8080:80

⚠️ Image reference must come immediately after run, before any Docker flags. Since DBRX Registry forwards arbitrary docker run flags transparently, the image has to be bound first so flags like -p, -e, --name, and -d are correctly passed through to Docker, not misread as the image name.

# -------------------------------------------------
# Detached, with env vars and a container name
# -------------------------------------------------
dcip run my-app:v1.2.0 -d -p 8080:80 --name production-app -e NODE_ENV=production

• Not OCI-compliant. This is a tarball shuttle, not a Distribution Spec-compatible registry, you can't docker pull from it directly, only through this CLI.
• No layer deduplication. Every push uploads a full compressed tarball of the image, there's no shared-layer storage like a real registry.
• 5 GiB per-file cap, per the Databricks Files API's direct-upload limit, fine for most application images, worth knowing for very large ones.
• Single reader/writer model. There's no locking, concurrent pushes to the same tag can race.

If you need full OCI compliance, layer caching, or multi-writer safety at scale, treat this as a lightweight bridge rather than a Harbor/ECR replacement, the sweet spot is small teams or CI pipelines that want zero extra infrastructure and already trust Unity Catalog for governance.

To do