AI service status

Why does an upstream outage look like a broken relay?

A relay doesn't generate answers itself — it forwards your request to the official endpoints of OpenAI, Anthropic, Google or xAI. The moment an upstream provider rate-limits, degrades or goes down, all the relay sees is timeouts, 429s and 5xx. Its retries, queueing and multi-channel failover then amplify that blip into longer latency, truncated responses, or even a silent switch to a different model — which looks almost exactly like the relay diluting your traffic.

Telling the two apart takes two steps. First check whether the provider on this page is having an incident right now: if the time window lines up, it is almost certainly upstream and will clear on its own. If it doesn't line up, run a fidelity check with the same key — when tokenizer fingerprints, thinking signatures or the identity echo don't match, this isn't upstream turbulence, it's the relay swapping models on you.