What gets recorded
Every billable call writes a usage event carrying what the service was, how many units it used, and what produced it. Model spend is priced per model and per route — the same model reached a different way can carry different rates — with fresh input, cached input, cache-write input, and output metered as separate line items. Everything else meters on its own unit: per request, per second, per query, per hour, per GB-day. Usage and limits has the detail.Where you read it
Usage & Billing in the Portal, a month at a time: balance, free credits, current-month spend, spending cap, workspace and knowledge storage, top services, top capabilities, daily spend, and recent monthly statements.When the balance runs out
The balance and the spending cap are enforced at the moment a run or an MCP request allocates its budget. An agent run with no budget available does not start; an MCP request with none is rejected with 402 Payment Required. That is worth planning around if what you sell runs unattended: the stop lands on the next call, not at the end of a billing period.What you charge
The meter measures your account’s consumption. The price of the thing you sell is a separate decision, made by you and set outside the platform.Packaging what you run
Back to the overview.