## Problem faced - Current state of backends (Groq, Together.AI, etc) have aggressive rate-limits - Its important to have model fallbacks in case things we hit something like this ## Proposed solution - Add support for _"Model Presets"_. - These are nothing but a primary model backed by fallback models that can be configured centrally - These get exposed to clients as a value that can be passed in place of a model. - Let's make sure all preset models have a `preset:` prefix so folks know that the underlying model can be changed from beneath them.
Problem faced
Proposed solution
preset:prefix so folks know that the underlying model can be changed from beneath them.