I’ve been developing a local retrieval-augmented generation (RAG) system to maintain conversational character and long-term memory across separate ChatGPT sessions. The central design principle is straightforward: the model handles the conversation; an external…