Concept | Agent Run Inspector#

The Agent Run Inspector provides live visual feedback on an agent’s activities, giving you real-time insight into an agent’s logic and enabling you to debug it as needed.

As an AI builder, you can track a test run live, and after the fact, see each step the agent took, what happened during that step, and exactly where the agent failed.

The inspector provides feedback in three areas of the visual agent editor interface:

  • An event timeline in the chat

  • The agent diagram

  • The event inspection panel

Timeline#

As you chat with an agent in the editor, a timeline in the chat interface shows the real-time sequence of steps as they process. You can see whether the steps have successfully completed or failed.

The timeline includes visibility into LLM reasoning, tool calls, and guardrails. For Structured Visual Agents, the timeline also includes nested tool calls.

The event timeline in the chat response shows each step an agent takes.

Agent diagram#

As the agent continues through its steps, the active reasoning step, tool call, or block showing in the timeline is also highlighted in the agent diagram.

The diagram is a graph in the lower right panel of the agent editor showing the components that make up the agent.

The agent diagram shows whether the agent executed tools and blocks.

Inspection panel#

After a test run completes, you can drill into the details in the event inspection panel. Open it by clicking on a step, tool call, or block in the timeline.

You can view input/output data, error messages, and metadata such as time to complete the step.

For Structured Visual Agents, the panel also shows state and scratchpad snapshots at specific steps.

The inspection panel lets you drill into the details of an agent run.

See also

For more information, see Agent Run Inspector in the reference documentation.

Next steps#

Get hands-on experience with the Run Inspector in Tutorial | Agent Run Inspector.