← Routing

AIO · cite-ready

How fallback works before the first token

Answer: Cybertronic Router may try the next configured provider when an error is retryable and no output token has been sent yet. After streaming starts, errors surface to the client — another provider’s output is never stitched into the same response.

Why it matters

Users see coherent streams. Operators still get automatic failover for pre-token failures (rate limits, upstream 5xx, timeouts).

Verify in beta

Force rate limits on provider A; confirm provider B serves the completion with route metadata in headers.

Routing beta