Documentation
Runs and activity
One execution of a published workflow, recorded step by step.
What a run is
A run is one execution, with the input it started from and the exact published version it pinned. Every step is created when the run is created, so the shape of the work is known before any of it happens.
How it advances
The engine takes the steps whose inputs are ready, runs them, and advances along the graph. Work is claimed one step at a time, so the same step is never run twice by two workers.
When something goes wrong
| Situation | What happens |
|---|---|
| A transient failure, such as a timeout | The step retries, with a growing delay. |
| A permanent failure | The step fails and the run stops there. |
| An approval is needed | The run pauses until somebody decides. |
| A guardrail blocks | The run pauses for an explicit override or rejection. |
| Somebody cancels | The run stops at the next safe point. |
In every case the work done so far is kept, so a paused or failed run shows exactly how far it got and what each step produced.
Test runs
A test run uses the draft rather than the published version, and is deliberately not the real thing: it runs without credentials, without dispatching to your systems, without pausing for approvals, and without writing to an agent's memory. Guardrails stay on.
Activity
Activity is the operator's view: every run, its steps in order, what each step received and returned, which agent acted, and who decided. It is a record rather than a dashboard, and it is what you show somebody who asks what happened on a given day.