# Model Registry
A model registry is a centralized store for trained ML models — versioned, annotated with metadata (metrics, parameters, lineage), and tagged with lifecycle stages (Staging, Production, Archived). It's the equivalent of a package registry for models: one source of truth, audit trail per version, controlled promotion between environments.
The registry decouples training from deployment. CI/CD pipelines pull a specific model version by name + stage instead of reaching into a notebook author's filesystem.
## References
## Related
- [[MLflow]]
- [[MLOps]]
- [[ML Deployment Patterns]]