Billing and usage

InferencePass reserves a conservative maximum before upstream dispatch. After the provider reports usage, the gateway normalizes token/media/tool counts, computes the customer charge from the price snapshot, settles the reservation, and exposes the final charge where available. If no usable usage is available, the reservation is released or marked for review according to the terminal state.

Successful responses include normalized usage and may include:

"inferencepass": {
  "charge": { "amount_minor": "46", "currency": "USD" }
}

amount_minor is an integer string in the published currency's minor unit. The charge is not a client-side estimate. Playground requests use the same reservation, settlement, budget, and ledger path as API traffic.

Usage records

GET /v1/usage returns settled usage records with request ID, transaction ID, charge, currency, settlement time, and available dimensions. Use the dashboard Usage view for workspace filtering and reconciliation workflows.

Safe retries

Only retry documented retryable failures. An idempotency key prevents duplicate settlement for a completed non-streaming request, but it does not erase a billed attempt that reached an upstream provider with uncertain usage. Keep the original request ID and inspect usage before replaying.