Tracing and troubleshooting

Record the response header X-Gateway-Trace-ID with your application request ID. Use it to correlate request metadata in the console without logging prompts, completions or credentials.

Investigate a request

Check model/provider, HTTP status, time to first token, total latency, retry count, token usage and cost where available. Distinguish upstream latency from client networking and application processing. A trace ID is a diagnostic identifier, not authorization.

Respond to an incident

For 401 check the key; for 403 check model access; for 402 review balance/budgets; for 429 reduce traffic and inspect the limit; for 5xx correlate traces and provider health. Use error codes for protocol-specific error fields.

Monitor failure rate, latency, retry rate and spend rate. These are recommended operational signals, not a promise that every threshold is configurable in the console. Configure available notifications and supplement them with application monitoring.