Model Registry

A model registry is a centralised system for managing the lifecycle of machine learning models. It serves as a single source of truth for all models within an organisation, storing model artefacts (the trained model files), metadata (training parameters, performance metrics, data lineage), version history, and deployment status. Model registries support key MLOps workflows. Developers register new model versions after training, reviewers approve models for production deployment, and operations teams track which model versions are serving in which environments. This provides traceability and governance across the entire model lifecycle. Popular model registry solutions include MLflow Model Registry, AWS SageMaker Model Registry, Azure ML Model Registry, Weights & Biases, and Neptune. Some organisations build custom registries to integrate with their specific workflows and compliance requirements.

As organisations deploy more AI models, managing them becomes a significant challenge. Without a registry, teams lose track of which models are in production, what data they were trained on, and who approved their deployment. This creates risk, particularly in regulated industries where model governance and auditability are required. A model registry enables critical practices: rollback to previous model versions when issues are detected, comparison of model performance across versions, compliance documentation and audit trails, and coordination between data scientists and operations teams. The registry also reduces duplication of effort. When teams can discover and reuse existing models, they avoid re-training models that already exist. This accelerates time-to-deployment and ensures that the best-performing models are consistently used across the organisation.