Routing and fallback

Control deployment selection

Add usable deployments for a public model before selecting a routing strategy or template. Choose weights, load, latency or cost according to business needs. Confirm candidate deployments support the request.

Steps

Routing templates can inherit through key, team and organization scopes. Clearing a binding restores inheritance. Templates choose deployments within existing permissions and do not expand model access.

Checks and limits

Cross-model fallback is available only before response bytes are emitted. Set disable_fallbacks:true to disable it. Pinned continuations, native asynchronous creation and Responses background requests are not replayed on another model. For unknown task outcomes, preserve the task ID and x-litellm-call-id instead of blindly resending.