Skip to main content
An agent run is one accepted execution of a specialist workflow.

​
Run states

A run can be queued, running, awaiting review, complete, failed, or cancelled. Linked work and Project outputs remain available after the run finishes.

​
Usage

One accepted run consumes one included run. Continuation turns and infrastructure retries do not consume another run. A manual retry starts new work and may consume another run.

​
Concurrency

Your plan controls how many runs can execute at once across the organization. Additional accepted work remains queued until capacity is available.

​
Stop and retry

Stop requests cancellation of active work. Some external operations may already have completed before cancellation is observed. Retry starts another attempt using the available saved context.

​
Recovery

Lantern monitors queued and running work for stalled progress and can safely re-enqueue eligible work. Safeguards prevent the same accepted request from being charged or executed twice during infrastructure retries.