Traces and evaluations

Trust in AI comes from being able to check its work. The AI group of the admin area gives you two oversight tools: traces for what happened, and evaluations for how good it was.

Traces

Traces is the flight recorder: every agent turn in your workspace, browsable and filterable, with a live tail for watching activity as it happens. Open any turn to see its full detail — what was asked, what steps the agent took, and what it produced.

Use traces when you want to understand how an answer came to be, investigate something unexpected, or simply see what your team’s agents are being used for.

Evaluations

Evaluations is the quality queue. Responses that deserve review land here — including ones your teammates flagged with thumbs-down feedback — for triage. Review each item’s detail and track it through to resolution.

This closes the loop on quality: users flag, admins review, and your workspace’s assistants get better for it.

Next steps

• Managing your team’s agents — the roster and per-agent controls.