Skip to documentation
Documentation

Operate

Runs and retries

Inspect production runs and recover safely from failures.

Updated Aug 15, 2026

Read the run timeline

The timeline shows inputs, step attempts, durations, and outputs in execution order. Secret values are redacted before storage. Select a failed attempt to see its error class and retry decision.

Retry a step

Automatic retries are intended for temporary failures. Configure the maximum attempts and delay on a step:

await step('send-report', sendReport, {
  retry: { attempts: 3, backoff: 'exponential', maxDelay: '30s' },
});

Resume a failed run

Manual resume starts at the failed step and reuses completed step results. Before resuming, confirm that the external operation is idempotent or verify its provider-side state.

Cancel a run

Cancellation prevents new steps from starting. An in-flight network request may still complete, so cancellation is not a rollback mechanism.