Usage

Per-request telemetry for calls to your deployed models.

The Usage screen shows per-request telemetry for calls to your deployed models. Filter by API key, deployment, model, status, and time range.

Usage: per-key rollups and a filterable request log.

Usage by key

Headline cards summarize total requests, total tokens, and average latency. Below them, a per-key table breaks down requests, error rate, total tokens, average and p95 latency, and when each key was last used. Calls made with a dashboard session (rather than an API key) are grouped as No key (JWT / anonymous).

Request log

The Requests table lists individual calls with their timestamp, API key, deployment, model, HTTP status, whether the response was streamed, latency, and token count. Filter by:

• API key, deployment, and model
• Status, All / 2xx / 4xx / 5xx
• Streamed, All / Streamed / Non-Streamed
• Date range, plus a Replayable only toggle

Use the status filter to find failing calls fast, filter to 4xx/5xx and check the deployment column to see which model is erroring. See Errors.