Agent runs

What Agent Runs shows

Agents do real, metered work on your behalf — setup builds and workspace chat runs. The Agent Runs page is the organization-wide record of all of it: every run across every project, streaming in live. For each run you see when it started, the project, its kind (Setup or Chat), live status (Running, Completed, Failed, or Canceled), duration, the model that conducted it, the tokens it consumed, and its cost.

It's the page to open when you want to know what your organization's agents are doing right now, why a build cost what it did, or which runs failed last week. Find it in the organization dashboard; like the audit log, it's included with the Enterprise plan.

Finding the run you're looking for

Filter by project, kind, status, and time range (the last 24 hours up to 90 days), or search across models, failure reasons, and what each run is currently doing. Runs that stop reporting progress are flagged as stalled, and runs that died with their machine are marked interrupted — so silence is visible, not mysterious.

Inside a run

Open Details on any run for the full anatomy: its phases (the main agent, checks, verification, typecheck, commit and push), where the time went (reading, editing, searching, running commands, waiting on the model), and a step-by-step timeline with the files each step touched. It's the same visibility an engineering lead would want over a human team — for your agents.

Runs and your credits

The cost shown on each run is what that run's model work cost in usage credits. Runs conducted on your own Cursor API key (a bring-your-own integration) show their cost but are never debited from your balance. On the Enterprise plan, the page also exports the full run history as CSV or JSON.