Runs

A run is one execution of a job, from start to finish. Runs appear in your Inbox as they happen, whatever started them.

What starts a run

Three things, and each run records which one it was:

• Schedule: the job’s schedule fired.
• Trigger: an event arrived from a connector. See triggers.
• Manual: you ran it yourself, from the app or by asking your main agent.

The lifecycle

1. Queued: the run is accepted and waiting for capacity. Time spent here does not count against the run’s time limit.
2. Running: a fresh, isolated machine starts and the worker agent carries out your prompt, using your connectors and your skills.
3. Reviewing: the overseer reads the result and records a verdict.
4. Completed, Failed, or Cancelled.

When the run ends, files left in the working directory are uploaded as artifacts and the result lands in your Inbox with the overseer’s verdict beside it.

Time limits

Every run has a time limit, set per job: 15 minutes by default, adjustable from 5 to 60 minutes. It covers the agent’s working time, not time spent queued.

A run that goes past the limit is stopped and marked failed with “Run exceeded its total time limit”. If runs regularly go long, raise the limit on the job.

When something goes wrong

If a run’s machine stops unexpectedly, the run is marked failed with “The run stopped unexpectedly before it finished.” Nothing relaunches it, and nothing relaunches a run the overseer fails either: that is a verdict, not an error. Either way the job waits for its next schedule, trigger, or manual start.

Cancelling

You can cancel a run while it is queued or running. The machine it was running on is torn down and the run is marked cancelled.