Monitor agent performance overview

The Monitor tab gives makers an in-context view of an agent's production health. Use it to understand usage, outcomes, user feedback, reliability, and Copilot credit consumption so you can identify trends and decide where to improve the agent.

Note

Features in this article are used by agents or workflows powered by the GitHub Copilot harness.

Usage-based billing applies to using, building, testing, and evaluating agents. These actions might consume Copilot Credits. Learn more in Manage costs for agents powered by the GitHub Copilot harness.

Note

You can open the Monitor tab before you publish your agent. Production usage, feedback, and credit data appear after you publish the agent and people start using it.

What can you monitor?

Monitor provides a summary-level view of an individual agent. It includes:

  • Overview: Total sessions, users, autonomous runs, average duration, success rate, and fail rate.
  • Reactions: Thumbs-up and thumbs-down feedback on agent responses.
  • Total estimated credits used: The credits the agent consumed during the selected period.
  • Sessions: Individual interactions, where they occurred, and the credits they consumed.

Use the date filter to focus on a specific reporting period. Percentage indicators compare each metric with the previous period.

Monitor and agent operations

Monitor is designed for makers who need a quick view of an individual agent while they build and improve it. For fleet-wide monitoring, governance, detailed investigation, and operational optimization across agents and environments, use the agent operations experience.

Overview metrics

The Overview section shows:

  • Total sessions: The number of conversational sessions during the selected period.
  • Total users: The number of users who interacted with the agent during the selected period.
  • Average duration: The average duration of sessions and runs.
  • Success rate: How successfully the agent completed sessions and runs.

Total estimated credits used

The credits card shows the total Copilot credits the agent consumed during the selected period and the change from the previous period. Unlike the near-real-time estimates you see in Preview and Evaluate, the value shown in Monitor reflects all events processed up to that point and is the final, source-of-truth number used for billing.

When a breakdown is available, it shows an estimated split of consumption by experience, including production usage, preview testing, evaluations, and other supported surfaces. Use it to understand which activities or outcomes consumed the credits.

Learn more in Understand estimated credit usage and Overview of usage-based billing.

Reactions

The Reactions section shows user feedback on agent responses. The chart counts how often users select thumbs up (positive) or thumbs down (negative) on each response they receive from your agent.

Agent capabilities

The Capabilities section helps you understand how users interact with your agent and where answer reliability can improve. It includes:

  • Tool use: Which tools the agent invoked during the session.

Session tracking

The session list shows individual conversation sessions with your agent. Each session includes:

  • Status: Whether the session is completed, failed, or in progress.
  • Duration: How long the session lasted.
  • User messages: The number of messages exchanged.
  • Tools used: Which tools the agent invoked during the session.

Filter sessions by status, such as failed, completed, or in progress, to quickly find sessions that need attention.

Select a session to view its full transcript, including the user's messages, the agent's responses, and details about tool invocations and knowledge source usage.

Error monitoring

The Monitor tab helps you identify and diagnose problems:

  • Filter sessions by Failed or Rejected status to find problematic interactions.
  • Review session transcripts to understand what went wrong.
  • Check tool connection status and authentication errors.
  • Look for patterns in failed sessions that might indicate a configuration problem.

For step-by-step guidance on using the Monitor tab, see Monitor an agent.

Download session transcripts

You can download session transcripts a few minutes after the session times out. You can download any time period within the last 29 days. You can download them in Dataverse using the Power Apps portal and as session chat transcripts using the Copilot Studio app. It can take up to an hour after the analytic session ends before the related data appears on the Monitor tab.

Review sessions

The sessions list shows recent interactions with your agent.

Note

The sessions list shows sessions from the past 28 days only.

Depending on the interaction, the list can show:

  • Name: The user associated with the interaction.
  • Date: When the interaction occurred.
  • Channel: The channel or Copilot Studio surface where the interaction occurred.
  • Number of messages: The number of messages in the interaction.
  • Duration: The length of the interaction.
  • Credits used: The Copilot credits consumed by the interaction.

Open the session details page to review aggregate session metrics, filter the list, search for sessions, and download session data.

For other transcript export options, learn more in Download agent session transcripts.

For step-by-step guidance, learn more in Monitor and analyze agent performance.

  • Last updated on