When an AI request fails.
Read the evidence before deciding what to do next.
AI service status, errors and agents
Use the exact product, collection time and request error together. These guides explain the scope and limits of the reports shown here.
Is an AI service down? How to check the evidence
Compare official AI status, collection freshness and your failed request. Learn why a normal provider report cannot confirm every model or network.
Read guide Request troubleshootingAI API errors: 401, 429, 5xx and timeouts
Separate authentication, quota, rate limits and transient failures for OpenAI, Claude and Gemini APIs. Use error details before deciding to retry.
Read guide Agents & operationsHow agents should use AI service status
Use provider reports as one input to agent recovery. Preserve request evidence, check report scope and freshness, and avoid unsupported outage or failover decisions.
Read guide Request troubleshootingAI API error 429: rate limits, credits and spend caps
A 429 is a request limit response, not proof of a provider outage. Read the provider error code, message and retry headers before retrying.
Read guide Request troubleshootingAI API error 500: internal errors and bounded retries
A 500 indicates an internal server error on the responding endpoint. One failed request does not establish a global AI service outage.
Read guide Request troubleshootingAI API error 503: unavailable or overloaded endpoints
A 503 means the responding endpoint is unavailable for this request. Read its error details and the matching official report; it does not by itself confirm a global outage.
Read guide Request troubleshootingClaude API error 529: overload, evidence and retries
Claude documents 529 overloaded_error as temporary API overload. This is different from a 429 rate or spend limit, and a single response does not establish the reach of an incident.
Read guide