Share via

Agent observability with MLflow Tracing

  • Article

MLflow Tracing provides end-to-end observability for generative AI applications from development to deployment. Tracing is fully integrated with the Databricks gen AI toolset, tracking every step your AI agents takes.

MLflow Tracing overview

Key use cases include:

  • Streamlined debugging: See every step of your gen AI application to quickly diagnosing and resolve issues.

  • Offline evaluation: Tracing generates valuable data for agent evaluation so you can measure and improve agent quality over time.

  • Production monitoring: Monitor agent behavior and view detailed execution steps to optimize agent performance in production.

  • Audit logs: Tracing generates comprehensive audit logs of agent actions and decisions. This is vital for ensuring compliance.

Note

MLflow 3.0 offers enhanced capabilities for MLflow Tracing. For more information, see Get started with MLflow 3.0 (Beta).

Convenient and powerful instrumentation

Use automatic tracing to instrument your agent with a single line of code. Use manual tracing with flexible decorators, context managers and APIs for custom traces. Or, use a combination of both for convenient and comprehensive coverage.

See Add MLflow Tracing to AI agents.

In-line tracing captures detailed information for each step in a gen AI app