You can let users override the model in their request payload by configuring the APISIX ai-proxy plugin so that the "model" field in the JSON body is forwarded to the AI backend. There are no built-in restrictions on which model names can be set by the user—any string value for "model" will be accepted and proxied, as long as your backend supports it source.

If you want to enforce a default model or restrict overrides, use the plugin's "options" field to always set a specific model (which will override user input), or use the "defaults" field to provide fallback values that only apply when the user doesn't specify them source.

For more control, you can create multiple routes with the same URI and use the route "vars" to match on "post_arg.model"—this way, requests with different model names are routed to different backends source.

If you're using ai-proxy-multi, dynamic model selection from the request body isn't supported; you need separate routes for each model source.

Let me know if you need example route configs or want to discuss best practices for restricting allowed models.

To reply, just mention @dosu.

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}