Every step, source, tool call and decision — recorded and searchable.
Every run produces a trace: an ordered record of everything the agent did.
| Step type | Recorded |
|---|---|
trigger | Trigger type and input |
retrieval | Query, knowledge bases, passages retrieved with scores and permissions applied |
reasoning | The agent's intermediate plan and conclusions (summarised) |
tool_call | Tool, parameters, scope, response, duration; simulated flag in test mode |
approval | Policy triggered, approvers notified, decision, decider, timestamp |
handoff | Recipient and case summary |
result | Final status and output |
Agents → (agent) → Runs → (run) or GET /runs/{id} and GET /runs/{id}/trace. Traces are searchable by input fields, tool, status and time.
Default 90 days; configurable from 30 days to 7 years per workspace. Traces can be exported continuously to S3-compatible storage or your SIEM.
Traces contain the data the agent processed. Access to traces follows workspace roles; admins can restrict trace access per agent to a group (for example, finance traces to finance).