elitsecorp/mnimosyne ? reverse-engineered prompt

Build me a Python web app that works like a persistent memory layer for an LLM.

I want a local FastAPI app with a simple chat page where I can talk to an OpenAI compatible model, and every message gets saved so the app can remember things across sessions. It should also pull out people, places, facts, and relationships from the conversation, store them in SQLite, and use both semantic search and a knowledge graph to bring back relevant memories before each reply.

Please include pages where I can browse saved memories, explore the graph visually, import text or transcript files, and review suggested cleanup like merging duplicates or normalizing relationships before applying changes. The chat responses should also show a small details view that explains what memory was retrieved and why.

Use local embeddings if possible, keep it simple to run with uv, and if you need current docs for anything, look them up online.