Usage & observability
Read request outcomes, latency, and provider-reported token measurements.
Open Usage to inspect requests made with your API keys. The dashboard supports 7-day and 30-day views, per-key filtering, daily outcomes, reported token charts, and paginated request details.
What each request records
- Request ID, API key name, endpoint, model, and start time.
- HTTP status, outcome, and error code when present.
- Total duration and time to first response when measured.
- Provider-reported prompt and completion tokens when supplied.
Prompts, answers, and API key secrets are excluded from usage records.
Outcomes and HTTP status
| Outcome | Meaning |
|---|---|
completed | The request reached its normal terminal response. |
failed | Validation, routing, dependency, or generation failed. |
cancelled | The client closed the request before completion. |
rate_limited | An account request or concurrency limit rejected the request. |
A streaming request can return HTTP 200 and still record a failed outcome if generation stops after response headers have been sent. Read the terminal stream event instead of treating the initial status as completion.
Correlate requests
Chat completions return an X-Request-Id header. Log that ID with your application trace and user-visible error reference. Do not log bearer tokens or full prompt content.
Understand token measurements
Token counts come from the participating provider. Missing usage stays missing rather than becoming zero. These values are useful for operational visibility, but they are not a verified billing ledger.
For request failures, pair the usage record with the error code reference.