Agent summaries now show how conversations went, and every metric opens the records behind it

An agent's Summary tab now shows total conversations, average and P95 duration, error rate, and average tokens per conversation, each compared with the previous period. Charts show the same metrics over time, and Segment by splits them by model, workflow or task.

Trace stats count individual turns, but users experience a whole conversation. One failed turn can ruin a session even when the trace error rate looks fine. Now you can see how many sessions hit an error and how long they took, at a glance.

Select any metric on the agent's Summary tab to see what's behind it. Each metric opens the Traces or Conversations tab, already filtered or sorted to the records behind that number:

• Errors opens traces with errors
• Avg latency sorts traces by duration
• Error rate opens conversations with errors
• Top tool opens the traces that called it

Hover over a card to see where it leads. Clear filters puts back the filters you had before.

Learn more here: Trace and conversation structure