Actually there are a few LLM wrappers around that use the openai API spec (localai is a good one)... so you could just allow a configurable openai endpoint URI and technically users can swap in any model.