Skip to main content
Response times is currently in beta. Use the filter bar, as on the Conversations page, to select a period, agents and a model. Add permission, organization or source filters with More filters. Remove a filter badge to clear that selection. Open Developer → Monitoring → Response times to track how quickly your AI agents answer. Use the header filters for period, agents, organizations, user type and conversation source. The Model filter selects the model used for the final answer, even if the agent’s configuration has changed since then.
  • Average complete response measures server processing from the start of a turn until its answer finishes.
  • Average time to first text measures server processing until the first visible text of the concluding answer segment.
  • Measured answers shows the number of completed answers included. Each average uses only answers with that measurement.
The chart groups measurements by hour for Last 24 hours, and by day for other periods, using your browser’s timezone. Missing measurements appear as gaps, not zero-second responses. The tables compare the selected answers by agent and final answer model. These timings exclude client pauses (such as waiting for a browser tool result) and processing after the answer. They measure server time, not network delivery to the user. Simulations, failed or interrupted answers, and answers without timing data are excluded. Older answers appear only when timing data was recorded. Unknown agent or model names are shown as Unknown.