Adapter definitions are too strict

[olokobayusuf]

Hey! My team is considering building a community adapter for running AI models with TanStack AI. For context on us, we're building a platform that allows developers to run arbitrary AI models by "compiling" them into self-contained binaries that can run anywhere.

The adapters guide states that we have to define model metadata ahead of time, listing things like model names, capabilities, options and so on. I totally get the design goals here (strong typing), but it them becomes impossible for devs to run a wider variety of models (especially if they are bringing these models themselves).