Tracking events

Use noryen.track() when you are not using a built-in wrapper, or when you need full control over the payload (custom providers, proxies, or non-HTTP flows).

Validation rules

• model is required.
• For successful calls prompt and response must be non-empty. Otherwise the event is dropped (a debug warning will appear if debug is on).
• For failures, set success: false and prefer including error.

Example

noryen.track({
  model: "gpt-4o-mini",
  provider: "openai",
  prompt: "Summarize this text in 3 bullets.",
  response: "1) ... 2) ... 3) ...",
  latency: 820,
  cost: 0.0012,
  inputTokens: 120,
  outputTokens: 80,
  metadata: { env: "prod", feature: "summary" },
});

Common fields

Optional but useful fields include provider, latency (ms), cost, token counts (inputTokens, outputTokens, totalTokens), requestId, and metadata (see Metadata).

Failed requests

noryen.track({
  model: "gpt-4o-mini",
  success: false,
  error: "Rate limited",
  prompt: "…",
  // response can be omitted or minimal when success is false
});