ModelDB

What ModelDB is

ModelDB versions machine learning models and tracks their metadata across development, deployment, and live monitoring. It records code, data, configuration, and environment details to retain context for reproducing model work. Experiment management includes performance dashboards and report sharing, while flexible metadata logging captures metrics, artifacts, tags, and user information. The project provides Python and Scala clients and lists TensorFlow and PyTorch integrations. It supports pluggable storage and is configured and verified out of the box with PostgreSQL. Deployment documentation covers Docker Compose and Helm, and the project says ModelDB works on Docker and Kubernetes. It is available through an API and web interface, or as a self-hosted deployment. The open-source edition is free under Apache 2.0. The README says to contact [email protected] about a hosted version. For support, the project directs users to GitHub issues for bugs and feature requests, and Slack for questions and help.

Who it suits

ModelDB suits teams that need to manage, version, and audit machine learning models across their lifecycle. Its Python and Scala clients and listed framework integrations may fit teams working with TensorFlow or PyTorch.

Limits

The Open-source ModelDB plan is self-hosted and costs nothing. The project directs users to contact [email protected] about a hosted version.

Getting started

1. Open the ModelDB project website on GitHub
2. Choose a self-hosted deployment
3. Use the documented Docker Compose or Helm deployment option
4. Connect a supported storage system such as PostgreSQL
5. Use the Python or Scala client, API, or web interface

Verdict

ModelDB combines model versioning, lifecycle metadata, and experiment management with API and web access. Teams should account for self-hosting on the free plan or contact the project about its hosted version.