Routing
Fallbacks & retries
Keep requests flowing when a provider blips.
The gateway is built to keep requests flowing when an upstream provider has a problem.
- Retries — the completion path can retry on provider errors and fall back to an alternative model where configured.
- Health-aware key selection — platform keys are tracked as
healthy,degradedorunhealthy; the resolver prefers healthy keys and only falls back to degraded ones. - Deterministic routing — a concrete model id maps to its provider; auto-mode selects within a tier.
Define fallback models in your configuration so a request can continue on a second provider if the first is unavailable.
Honest scope
Routing is deterministic per model with retry and health-aware key selection. Round-robin load balancing across providers is not part of the current gateway.