A cap is the most metered spend you allow in a billing period: overage plus hosted inference with its markup. Without a cap nothing stops, and usage past an allowance is billed. With one, work stops at it.
overlane billing caps set --scope team --usd 50
overlane billing caps set --scope project --id prj_1 --usd 20
overlane billing caps set --scope agent --id tok_1 --usd 5
overlane billing caps list
overlane billing caps remove --scope agent --id tok_1Dollars are read as text, so 12.50 is exactly 1,250 cents. Setting or removing a spending cap is a billing write, so it is human only: a signed-in owner does it, and an agent token is refused.
What a cap stops
At the cap, hosted inference, AI reviews and CI stop until the next period or until the cap is raised. Docs calls and ordinary writes are not spending and keep working. A project cap counts only that project’s spend, and an agent cap only that agent’s.
A stop is a policy refusal. It exits with code 6, names the cap, and says exactly what to run next:
overlane billing check --meter hostedInference --cost-micros 2000The same refusal comes back from every place the spend is checked: the command line, the API and the guardrail engine. It is enforced on the server, so an agent that ignores its instructions is still stopped.
How spend is counted
Spend comes from the usage ledger on the server, never from the caller. Within a period, the units beyond an allowance are attributed in the order they happened, so a project or an agent is charged for the overage its own work caused. The period is the one your plan uses, and a caller cannot move it to reset a cap.