Debug a run

Use the request, approximate time, agent and environment to find the run behind an unexpected result.

If an agent cannot find an order, its run history shows whether the lookup returned no result, failed or never ran.

Find the affected execution

Open Runs and find the execution by status and time. Open it to inspect the input, output and events.

Confirm the input and agent match the interaction you are investigating.

Follow the evidence

For the order lookup example, inspect whether the order tool was called. If it ran, check the supplied order reference and the response it returned.

EvidenceNext check
No tool call appears.Review the agent's instructions and enabled tools.
The tool reports an authentication failure.Check the integration connection and environment credentials.
The tool returns no matching order.Check the supplied reference and the external account or service being queried.
The tool returns the order but the answer is wrong.Review how the agent uses tool output in its response.

Inspect runtime logs

Open Logs when the issue involves a build, deployment, unavailable service, or runtime exception. Find the matching time window and check the deployment, server, or route involved.

For a browser-only preview problem, use Console. Browser output and server logs describe different parts of the request.

Retest the fix

Apply the relevant change in Agents, Tools, configuration, or source. Redeploy if the affected runtime needs the updated version or settings.

Repeat the original request. Compare the new run's events and result with the failure.

Next