Troubleshoot a failed or slow request

Start with the observed error

Record the error code, request time, environment, and trace or run ID. A slow request alone does not prove that the model or database is responsible. The trace is the starting point for distinguishing a gateway rejection from a provider or runtime failure.

Check configuration first

• Confirm the organisation, project, gateway URL, and requested model.
• Check the matching BYOK configuration or the managed inference model available to your workspace.
• Check request size. Requests beyond the confirmed model context window return context_length_exceeded; reduce the input before retrying.

Keep the next attempt small

Use a short request with a known available model and record whether it succeeds. Avoid repeatedly launching the same long-running workflow: preserve its run ID so support can inspect the existing attempt.

Ask for support

Sign in to Console to ask a question or talk to the team. State what you expected, what happened, and the steps you already tried. Review any diagnostic context before sharing it. Never include API keys, Vault values, or full private customer payloads. The Inference API reference documents request shape and routing. The changelog records published product changes; a staging merge alone does not establish production availability.