Skip to main content

​
Metering

  1. Reserve. Expected usage is reserved before the provider call.
  2. Settle. On success, the reservation settles for actual usage.
  3. Release. A request that fails before producing usage is not charged.

​
The usage Block

Every successful chat response includes one:
Reasoning models add reasoning_tokens. Server-side tools add server_tool_use with counts such as web_search_requests. A call settled against a prepaid balance may carry customer_cost, an exact amount in integer nanos of the named currency (1,000,000,000 per unit), so JavaScript clients keep precision. On a stream, ask for the usage chunk with stream_options.include_usage: true. See Streaming.

​
The Developer Console

The overview tab at app.trynito.ai/api shows spend for the selected window, requests, token totals, spend by model, and a daily breakdown. Owners and admins also see active keys and last use. Views are per organization, not per key, and need a browser session rather than an API key.

​
Failed Requests

  • An error before the provider returns usage: not charged.
  • A streaming failure after usage was recorded: settled for what was produced.
See Safe Retries.

Plans

Funding and allowances.

Safe Retries

Retry without paying twice.

Developer Console

Keys, usage, playground.

​
Read balance from code

Owners and admins can use an admin-enabled key with GET /api/v1/billing/balance. Amounts are integer USD nanos strings. This read returns available balance and top-up attribution; usage history remains in response usage blocks and the developer console.