Runs and error investigation
The Runs screen collects execution records from the integration and from runs started in the panel. Filter by app, agent, status, and time, then open the input, output, model, token counts, and latency that were actually reported.
Recorded run data
| SDK field | What it records |
|---|---|
| status | success or error for wrapAgentRun; reportRun accepts a status string. |
| inputSummary / outputSummary | Short descriptions for scanning runs. |
| fullInput / fullOutput | Optional captured content. Supply these explicitly when reporting a run. |
| errorMessage | The error captured when the wrapped function throws. |
| tokensInput / tokensOutput | Usage extracted from the provider result, or supplied through reportRun. |
| executionTimeMs / modelUsed | Elapsed time and the model recorded for the run. |
The wrapper does not automatically capture full prompts, outputs, or individual tool-call traces. A missing field means it was not reported. wrapAgentRun rethrows your application’s error after recording it.
Investigate a failure
| Method and route | Purpose |
|---|---|
| GET /api/runs | List runs with filters. |
| GET /api/runs/:id | Read one run. |
| POST /api/runs/:id/explain-error | Request an AI explanation of a failure. |
| POST /api/runs/:id/solve-error | Request suggested remediation. |
Review suggested fixes before changing your application. Explanations and fixes are generated on request; they are not evidence that the underlying problem has been resolved.