A run failed or is stuck
Read the error code, identify the cause, and know when to retry, cancel, or escalate.
Symptoms
A run ends as failed with an error code; a run sits in running far longer than usual; a run seems to make no progress.
Likely causes
Runs fail explicitly and for documented reasons. The most common: the output violated the agent's declared schema (a hard failure by design — never a silent success); the run hit a budget ceiling (maximum duration, tool calls, or tokens); the tenant hit a rate limit or quota — the error code tells you which (RATE_LIMITED, QUOTA_EXCEEDED, BUDGET_EXCEEDED); a required credential was missing (an explicit failure, no anonymous fallback); or — from the V1 Beta — the run is paused awaiting a human approval, which times out after 30 minutes by default.
What to try
- Open the run in the Console and read the error — every failure carries a documented code and message.
- For RATE_LIMITED, retry after a pause. For QUOTA_EXCEEDED or BUDGET_EXCEEDED, ask your workspace admin to review tenant quotas or the agent's run budgets.
- Check your input against the agent's input schema — the pre-filled run form shows the expected shape.
- For a stuck-looking run, check whether it is awaiting an approval (Beta). Otherwise cancel it — cancellation is safe and recorded — and submit again.
Escalate
If a run keeps failing, contact support with the run ID (it starts with run_). Every run is fully traced — model, tool calls, timings — so we can reconstruct exactly what happened from the ID alone.
Still stuck?
Tell us what you were trying to do and where it failed — screenshots and run IDs help. A human follows up.