Model registry
Centrally manage ML models and versions and link them to AI systems.
The model registry is a central directory of all ML models used in your AI systems. It enables version tracking, metadata management, and direct linking to the AI inventory. Model versions and training metrics can also be pulled directly from MLflow instead of being retyped by hand.
Information
The model registry is an Enterprise feature.
What is captured?
Model name and version
Unique identification of each model deployed, including version number.
Provider / framework
Where the model comes from – own training, open source, or a purchased service.
Training dataset
Reference to the training data used, for auditability.
Performance metrics
Accuracy, F1, AUC, or other relevant metrics as evidence of quality.
Fairness checklist
Guided bias/fairness assessment (Demographic Parity, Equal Opportunity, and others) with a value, judgment, and justification per metric – instead of a free text field.
Linked AI system
Which inventory system uses this model.
Creating an entry – step by step
- 1Open Model Registry in the left navigation
- 2Select the system in the selector (if multiple systems exist)
- 3Click "Create new model entry" or import models from MLflow
- 4Enter model name, version, and provider
- 5Add framework, training dataset, metrics, and the fairness checklist
- 6Save – the entry appears in the registry
Relevance for audits
Auditors frequently request proof of which model version was in use at which point in time. The model registry delivers this traceability directly – without manual research in internal wikis.