If you build a RAG vector database with your 2000 documents. Future queries will be compared to the content of that database and the closest match will be given

If you build a RAG vector database with your 2000 documents. Future queries will be compared to the content of that database and the closest match will be given to an LLM as source content for its reply to you. If you were trying to build a game element or simple responder with a very narrow purpose you might get by with one of the small 7b open source llm’s and host the whole thing on your personal computer