Run

A run is a single execution of a mission — one specific instance with a start time, a sequence of agent steps, tool calls, and a final outcome.

Missions are templates; runs are instances. Every time the trigger fires, a new run is created with its own ID. The run records every prompt, every tool call, every model response, and the final state. This is the unit you debug. Good platforms make runs replayable: you can re-execute a failed run with the same inputs after fixing the prompt or the tool, and compare the outputs. Without replayability, debugging an agentic system is guesswork.

Example

Mission 'send weekly digest' has 12 runs over 12 weeks. Run #7 failed because the analytics API timed out. You inspect the run, see the failure, fix the retry logic, replay run #7, confirm success.

How OpenLabor uses it

Every OpenLabor mission produces inspectable, replayable runs. You can scrub the timeline like a video.

Are runs the same as 'executions' or 'sessions'?

Different products use different words for the same idea. We use 'run' because it matches the mental model of 'this run of the script'.

Can I cancel a run mid-execution?

Yes — and you should be able to. Cancellation should also undo any reversible side effects taken so far where possible.

Related: mission, agentic-workflow, audit-trail.

AI Labor Glossary