I tried to do RAG on my laptop just by setting it all up myself, but the actual LLM gave poor results (I have a small thin-and-light fwiw, so I could only run weak models). The vector search itself, actually, ended up being a little more useful.