This is simply automatic toggling of the “Show AI Search results” for people who have AI search enabled on their instances.
The AI plugin can still be disabled.
Worth mentioning that AI search doesn’t use traditional text generation LLMs, but embeddings models. Not sure if they fall under the same blanket for your specific community, but may be something to discuss.
Well, so you be happy to know that we are so compatible with Mistral AI that it is what is used on this forum for most user facing features, like summaries, proofreading, etc.
Discourse AI is LLM provider agnostic and is compatible with virtually all models, both closed and open weights, and you can host models like Mistral on your own hardware and connect it to your Discourse instance. It’s what we do on this instance here.