Know what your API is doing.
Every request through the Gateway, traced: what was asked, what served it, how long it took, what it cost, and — when it failed — why.






Requests
Every call that reached the API, including the ones that failed before a token existed. Counts, error rate, true p50, p95 and p99 latency, and cost — over time, by model, or by key.
Failures
Errors are classified, not listed. Auth, quota, validation, provider, timeout, internal — each with whose problem it is. A provider outage and a malformed request need different people.
Cost
Priced when the request ran and frozen, so a later change to a model price never moves a historical figure. Managed and bring-your-own-key spend are counted apart, because only one of them is yours.
An API you cannot observe is a liability. Every request carries an id you can quote; every failure carries a class that says whose problem it is; every dollar was priced when it was spent.
The Console comes with every plan. Pair it with an API key and anything you build is on the same instruments.


