It's already partly exist, need to test it a bit more and implement a proper LLM class that can handle both openai and llama models.