Review runs and usage statistics
Run records show what an agent did, whether it completed, and what it cost. Use them to investigate failures, check scheduled work, and understand usage.
Browse run history
Open Operations → Runs (/runs). The history lists manual, chat, scheduled,
and other recorded runs available to your workspace. Filter the list to narrow
down the agent, status, trigger, or time window you care about.
Open a run to inspect its timeline. Depending on the run, you can see model steps, tool calls, generated files, completion state, duration, and usage. Failed turns remain visible as failures rather than disappearing from the record.
Read a run safely
Start with the final status, then locate the first failed or refused step. A tool error is different from a model error, and a delivery failure can happen after a report or media file has already been generated. Check the timeline before rerunning expensive work.
Use the source and provenance details to confirm which metrics, datasets, files, or knowledge supported the result. A completed status shows that execution finished; it does not replace reviewing the result's evidence.
Use run statistics
Open Statistics (/runs/stats) to see aggregate activity for the selected
window. Use it to understand run volume, outcomes, usage, and which agents are
doing the work.
Statistics summarize recorded runs. Change the time range before comparing two periods, and remember that one user request may include several model or tool steps while still being one run.
Investigate common issues
- A scheduled run is missing: confirm the schedule is enabled and inspect its latest recorded outcome.
- A run failed: open the detail and start with the earliest failed step.
- Usage is higher than expected: identify the agents and run types in Statistics, then inspect representative run details.
- An output is missing: check whether generation completed but delivery failed, and look for any file artifact already attached to the run.
Run records may expose business data and generated files, so access follows the workspace's role and data boundaries.