Meter

Every request accounted for as it happens, failures included.

The Meter records every request through the Gateway: route, model, tokens, cost, time to first token, and how the request finished. Failures are metered too, with finish=error, because a request that failed is still a request that happened.

Where to read it

• The Meter page in your dashboard: spend and traffic by route, model, key, and workspace.
• The Billing page: the month's token volume against your plan.

Metering adds zero cost to the hot path; the numbers are written off the request path and are typically queryable within seconds.

Usage alerts

You get an email at 80%, 90%, and 95% of your monthly plan volume, each at most once per month. Plan volume is a soft entitlement: work never pauses at a threshold. The wallet balance is the only hard gate, and it returns a 402 only for new requests, never by cutting a stream.

What a request costs

Provider-reported usage is the billing truth, priced against the catalog. The gateway's own overhead is reported on every response in x-allocate-overhead-ms, so you can hold the platform to its 30ms p50 budget yourself.