Comment by asixicle

That's what the embedding model is for. It's like a tack-on LLM that works out the relevancy and context to grab.