Current implementation heavily relies on OpenAI API due to vectorstore use. Todo: - use LLM agnostic vector db - make agent LLM provider agnostic