Spend caps
A cap is a ceiling in dollars on one key. An agent that loops stops at the figure you chose, not at your whole balance.
| Cap | What it limits | Resets |
|---|---|---|
| Total | Everything the key has ever spent | Never; raise it or make a new key |
| Monthly | What the key spends in a calendar month | At the start of each month, UTC |
What happens at the cap
The request is refused with a 402 before anything is sent to a model, so nothing is billed. The message says which cap was reached and how much the key has spent, and your account credit is untouched. Other keys carry on.
How a request is checked
Before a request runs, the most it could cost is held: the prompt plus the output limit, at the model's prices. If that would take the key past its cap, the output limit is lowered to fit. Only when even a small answer would not fit is the request refused. When the request finishes, the hold settles to what it really cost.
Set caps in the dashboard under Limits, or when you create a key.