Usage analytics

Each request is logged with everything you need to understand spend and behavior. The usage log records, per request:

• user_id, organization_id, workspace_id
• model_id, provider
• input_tokens, output_tokens, cached_tokens, cache_hit
• cost_usd, plus platform_cost and margin_multiplier for platform keys
• response_status, response_time_ms, stream_mode
• auto_mode_tier — the tier auto-mode selected
• key_source — tenant or platform
• client_source — e.g. Cursor, Zed, Qwen or your app
• correlation_id (ULID) and timestamp

In the dashboard, analytics break down tokens, spend and model mix per user, team and provider, under Analytics.

Logging is written asynchronously through a durable queue keyed by the correlation id, so the request hot-path stays fast.