Skip to main content
Go to Developer → Monitoring → Response times to see how quickly your AI agents answer. Use the filters at the top to choose a period, agents, a model, a user type, organizations and a conversation source. Remove a filter badge to clear it. The Model filter uses the model that wrote the final answer, even if the agent’s configuration has changed since.

Metrics

  • Average complete response: time from the start of a turn until its answer is complete.
  • Average time to first text: time until the first visible text of the final answer.
  • Measured answers: number of completed answers included. Each average uses only the answers that have that measurement.
  • P05 and P95 under each average show the spread: 5% of answers are faster than P05, and 5% are slower than P95.

Chart and tables

The chart shows both averages and the complete response P95 (dashed line). It groups measurements by hour for Last 24 hours and by day for other periods, in your browser’s timezone. Periods without measurements appear as gaps, not as zero-second responses. The By agent and By model tables compare the selected answers, with the same averages and percentiles for the complete response. Slowest answers lists the 10 longest complete responses for the current filters; click a date to open the conversation.

What is measured

Times are measured on AGO’s side. They don’t include network delivery to the user, pauses while AGO waits for the user’s app (for example a browser tool result), or work done after the answer. Simulations, failed or interrupted answers, and answers without timing data are left out, so older answers may not appear. Agents or models that can’t be identified are shown as Unknown.