On the main branch, there is support for in-browser AI, such as Chrome AI (built-in) and WebLLM.

However, these models require modern hardware to run well (and WebGPU for better performance) and also a couple of flags in the case of Chrome AI, which makes them not very accessible to the vast majority of users.

So it can make sense to develop them separately in another extension and use the provider registry API to register them.