Troubleshooting

If the service exits during startup, run it with the same configuration file and environment used by the deployment and inspect the redacted configuration error.

If /health/liveliness works but a chat request fails, check the model route, provider key injection, provider base URL, gateway bearer token, and upstream timeout. Do not paste provider keys or authorization headers into issue reports.

If a response stalls, first inspect /status and /metrics for active streams, aborts, cancellations, and upstream timing. For deeper local investigation, enable the loopback-only diagnostics listener and inspect the goroutine and goroutineleak profiles described in testing. Disable it again after the investigation.