The problem
Standard app monitoring stops at the HTTP call. It does not tell you which model answered, how many tokens it burned, what it cost, whether a policy fired, or why a chain of tool and agent calls was slow. That detail lives inside the AI traffic you cannot see.
How it works
Every request through the gateway is recorded and traced. You get a searchable request log and OpenTelemetry spans and metrics you can export to the stack you already run, with no change to your app code.
What you get
- A request log with the identity, model, tokens, cost, latency, status, and policy outcome of every call.
- OpenTelemetry traces and metrics, exported to your own collector.
- Per-caller, per-team, and per-model views, so you can see where time and money go.
- Tool and agent calls traced as part of the same request, so a slow chain is visible end to end.
See the docs to get started.