One option is that you could use gpt4all. You Install it, select a LLM model from the list, then take it offline. Then put a copy of all the qubes docs in that VM and configure it to index the qubes docs.
A installation procedure that worked at one point in time is here: A procedure that worked for getting gpt4all working