I am working on a project to enable an LLM to perform web searches using RAG (Retrieval-Augmented Generation). I’ve successfully implemented the RAG component, but now I need to integrate historical chat processing. I’m encountering challenges in creating a decision-making mechanism that determines whether the LLM should respond based on RAG or its internal knowledge. Do you have any suggestions for addressing the issues I’m facing with the historical processing and decision-making?

i have tried ConversationSummaryBufferMemory