Requests, usage & costs

Request → receipt → usage

Capture completion ID and X-Request-Id. Inspect request state and its receipt, then correlate recorded inference usage by request ID. Token measurements may be unknown. Absence is not zero; SIMULATED is not live execution.

Report project consumption

GET /v1/usage/report supports scoped inference keys with telemetry:read or an authorized project management credential. GET /v1/inference/usage provides raw management records. An inference-only inference:use key is insufficient for those reads.

Telemetry and charges

Authorized telemetry queries and latency distributions describe measured observations. Inspect sample coverage and freshness. Usage, estimated cost, recorded charges and final settlement are distinct. Unknown provider liability must not be represented as a zero bill.

Usage & Account reference → · Errors, limits and retry handling →