Usage and logs
Read the usage trend for load, and the per-request logs for time, key, model, tokens, cost, latency, and status.
The console gives you two views of consumption: the usage trend on the account overview (macro — how load moves) and the per-request logs (micro — exact records).
Usage trend (account overview)
- A time-series chart of call volume, metered in tokens;
- Use it to gauge overall load and spot abnormal spikes (e.g. a day where usage suddenly doubles);
- An empty state simply means no calls in that period.
Logs (per-request records)
The logs page records every API request with fields you can audit:
| Field | Description |
|---|---|
| Request time | When the call happened — anchors reconciliation |
| API key | Which key made the call (distinguished by name) |
| Model | The model ID actually invoked |
| Tokens | Tokens consumed by the call (input + output) |
| Cost | The amount deducted |
| Latency | Request duration — for hunting slow calls |
| Status | Success / failure with the corresponding status code |
Typical uses
- Reconciliation: aggregate the logs over a time range and compare counts and amounts with your application's request records;
- Locating abnormal spend: filter by key to find the biggest spenders — the culprit is often a test key accidentally used in production;
- Debugging failures: pair with Error codes and look at status codes over time (failures clustered at one moment usually mean a platform-side blip);
- Performance watch: the latency column reveals slow models or long-context requests, informing model choice and timeout settings.
Logs and usage come from real calls — records are queryable immediately after a request, nothing to wait for.