Set budgets and know when a run will stop

Agents cost money to run because each step uses your AI provider. Budgets make sure that cost has a ceiling.

Set a budget

1. Open the agent.
2. In the Budget panel on the right, set Per run (USD). This is the most one run may spend. It must be above zero, and the page tells you the highest amount allowed.
3. Set Per month (USD, 0 = no limit).
4. Choose Save changes.

The panel also shows how much the agent has used this month. A new agent starts with suggested budget amounts, and packages from Marketplace may suggest different ones. Check the Budget panel and set amounts you are comfortable with.

What happens at the limit

• When a run reaches its per-run budget, the run stops and is marked Failed with a plain message.
• When the month's spend reaches the monthly budget, new runs are refused until the month turns over or you raise the limit.
• Budgets apply even when you use your own AI key.

The other limits

• A step limit per run. A run that needs more model turns than that stops with a message saying it reached its step limit.
• A time limit for work in one go. Time spent waiting for an approval does not count.
• Limits on active runs per agent and per account, and on how many agents an account can have.
• Schedules cannot fire more often than a minimum gap.
• Approvals expire if you do not answer in time.

Your account may also have daily limits on runs and on how many run at once. If you hit one, the message names the limit.

Cancel a run

Open the run and choose Cancel run. It stops and shows Cancelled.

Watch spending

The Usage page shows cost per day and, per agent, This month / budget with a bar that turns red at 90 percent. See Usage and cost.

Good to know

• The cost figures are estimates from token counts. Your provider's invoice is what you are charged.
• Set a low per-run budget while testing and raise it once you trust the agent.
• A failed run still shows every step it took before it stopped.