Connect an MLflow registry
Connect your MLflow tracking server or Databricks workspace, and Drift & Bias reads its registered models, versions, training runs, dataset inputs and feature importance. The connection is read only: model files and datasets are never downloaded.
Connect the server
- Open Settings > Integrations and find MLflow in the Library. You can also select Open Integrations from the Add an ML model panel on the ML models tab.
- Add the connection for your tracking server. It can sign in with a token, a username and password, or no sign-in. Credentials are stored encrypted.
- Select Check access to confirm QuilrAI can reach it.
The connected server shows its address, authentication, the number of registered models and when it was Last refreshed, with a Connected badge.
Choose the models to monitor
Figure 1. The MLflow integration: connection and registered models.
Under Registered models, the MLflow registered models and monitoring table lists every registered model with its key, Latest version, Versions and Last seen.
Select Monitor for each model you want Drift & Bias to check. Versions and training runs are read only for monitored models.
When models are read
- Every 6 hours, automatically.
- At once, when you select Refresh models.
Version aliases such as champion show on each version, and training runs show their MLflow run names.
A model can come from MLflow and from a pipeline at the same time. Its Source on the ML models tab then shows Pipeline + MLflow.