Find issues across traces
A single trace tells you what happened on one request. To find problems across many traces at once — where errors cluster, which requests are slow or expensive, and what failure patterns recur — use Detect Issues. MLflow analyzes your traces with Genie Code and groups what it finds into issues you can act on.
Detect issues
- On the Traces tab, click Detect Issues (the sparkle at the top right of the toolbar).
- MLflow runs a Genie Code analysis over your traces: it establishes baseline metrics and trace volume, triages and deep-inspects failing or outlier traces, computes per-agent cost baselines to flag anomalies, and then clusters what it finds.
- Review the results in the Issues tab. Each issue groups the related traces so you can open representative examples and root-cause them.
The analysis runs in the Genie Code panel, so you can read its reasoning as it works and send follow-up questions to dig into any finding.
Schedule detection
To keep detecting issues as new traces arrive, open the Issues tab and click Schedule detection. Set a Cadence for how often MLflow analyzes new traces, and optionally add trace filters to scope the analysis to a subset of traffic.
Additional resources
- Ask questions with Genie Code - Explore and debug traces in natural language, and ask ad-hoc questions beyond the detected issues.
- Programmatic access to traces - Search, read, and analyze trace data programmatically with the SDK and SQL.
- Enrich traces: tags, context, and feedback - Add metadata to traces so you can segment and group them.
- Build evaluation datasets - Convert traces into evaluation datasets.
Next step: Build a custom trace view