Monitoring
How much work the agents did and what it cost: tokens, an estimated cost, run time, and task counts by day and by week.
Monitoring answers the question a manager asks first: what did the agents do this period, and what did it cost. The section gathers usage and activity across the whole workspace — for a day, a week, a month, or half a year.
What the counters show
Four figures for the chosen period:
- Cost — the estimated spend in money.
- Tokens — how many were used, split into input and output.
- Run time — the agents' total working time, and how many tasks it came from.
- Tasks — how many runs there were, and how many of them failed.
Cost is an estimate, not the provider's invoice. It is computed from token counts against a built-in model price list, so a model missing from that list is counted at zero. Reconcile with your provider's billing when you need the exact figure.
Periods and slices
- By day — 1, 7, 30, or 90 days.
- By week — 30, 90, or 180 days; a partial week is marked as such.
- Filter by project — keep only the spend from a given project's issues.
Each figure has its own chart, so you see the shape and not just the total: steady load, a release-day spike, or a nightly autopilot.
The agent leaderboard
The table under the charts shows each agent's share: tokens, cost, time, and task count, sorted by spend. It is the fastest way to see which automations account for most of the spend.
What you can see
Monitoring respects agent visibility: private agents you have no access to appear neither in the leaderboard nor in the totals. When such agents exist in the workspace, the section says so and hides the trends — otherwise their usage could be worked out by subtraction.
Admins and owners see the whole picture.
Next
- Agent tasks — what the time and usage of a single run are made of.
- Agents — where to cap how many runs go at once.
- Autopilots — the recurring work that usually drives the spend.