Model Registry
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A model registry is a central catalog of trained ML models: every version, its metrics, its approval status, and which one is running in production.
What is a Model Registry?
A model registry is a central catalog of an organization’s trained machine learning models, recording for each one the versions that exist, how they were trained and evaluated, who approved them, and which version serves in production. Without one, that knowledge lives in file names and chat threads. With one, “which model made this decision, and why did we trust it?” is a lookup, not an investigation.
How a Model Registry Works
When training produces a model worth keeping, the team registers it: the model artifact plus its metadata, such as the training code version, a pointer to the exact training dataset (which is where data versioning comes in), evaluation metrics, and the author.
Each registered model then moves through stages like staging, production, and archived. Promotion between stages is an explicit, recorded act, so the registry doubles as the approval trail for what runs in production. Tools such as MLflow build a registry into the training workflow, and the deployment system pulls exactly the artifact the registry names.
This record-keeping is also a governance asset. The EU AI Act requires providers of high-risk AI systems to maintain technical documentation on how their systems were built and tested; a registry holds much of that evidence in one queryable place, ready for an AI audit.
Example of a Model Registry
A bank’s credit-scoring model draws an audit question: which model version scored applications last March, and what data was it trained on?
The team opens the registry. Version 14 was in production from January to April; its entry links the training dataset snapshot, the evaluation metrics it had to clear, and the approval that promoted it out of staging. The answer takes minutes, and every claim in it rests on a recorded artifact rather than someone’s memory.
Related AI terms: MLflow · Deploy · Pipeline · Data Versioning · AI Audit
Did you like the Model Registry gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About