A model registry is a central repository for storing, versioning, and managing trained machine learning models. Core features: Versioning: Track model versions with metadata. Storage: Store model artifacts (weights, configs) reliably. Lineage: Record training data, code, parameters used. Lifecycle: Manage stages (development, staging, production). Access control: Permissions for teams and environments. Benefits: Reproducibility (recreate any model version), governance (track what is deployed), collaboration (team shares models), rollback capability. Common registries: MLflow Model Registry, Weights and Biases, Sagemaker Model Registry, Vertex AI Model Registry, custom solutions. Metadata stored: Model version, accuracy metrics, training config, data version, author, timestamp, stage. Integration: CI/CD pipelines pull from registry for deployment. Training pipelines push new versions. Comparison shopping: Compare versions on metrics before promoting. Governance: Approval workflows for production deployment. Audit trail for compliance. Best practices: Register all models (including experiments), include comprehensive metadata, automate promotion workflows.