fallback model

[GlauberF]

I’m not sure if this is possible, but I believe a very important feature when deploying an agent to production is the ability to define a primary model, for example, GPT-5 (OpenAI), and a fallback model, in case the main one fails, from another provider, such as Gemini 2.5 Pro.
Production systems need to be resilient to failures, and if OpenAI experiences instability, Gemini could automatically take over.

[cpfiffer]

It's on the roadmap, but we don't have a fixed timeline right now. You should be able to handle this client side by catching errors when sending agents to messages and changing the provider manually.

[rarisenpai]

Hello @cpfiffer , this is something I would be willing to work on

[GlauberF]

@cpfiffer I don’t think the ideal place would be on the client side. It would be simpler to have an additional fallback block and, if that fails, fall back to it on the server side.
Because when handling it on the client, you face the request/response time of the failure, and then additional request/response time for a new call, not to mention other implications and logic. In my view, it’s not an appropriate solution to do this on the client side.