Docs · Updated 2026-07-25

Debugging multi-step agents

Use Chainyze logs to reconstruct failed agent turns across tools and model calls.

Correlation

Put a shared `runId` or `sessionId` in metadata for every hop of an agent run so the dashboard list tells a story.

Empty tool outputs, truncated prompts, and slow models show up as latency + response content. Fix the step that broke first.

Copy the failing prompt from the log into a notebook. Chainyze is not a full replay engine — it is the audit trail.