This paper shows a new way to build AI assistants without using a complex database. Instead of hiding data in a database, we store the AI’s memory, personality, and skills in simple Markdown files that anyone can read. For systems like personal assistants or those with fewer users, a heavy database is often overkill. working_memory.md episodic_memory.md semantic_memory.md personality.md habits.md self_reflection.md skills.md skill_context.md To help the AI find information quickly, we use a tool called FAISS to search through these files, but the files themselves always remain the main source of truth. By using simple files instead of a database, the system is much easier to fix, track, and move. It’s a perfect 'middle ground' for personal AI projects because it’s simple to manage but still powerful enough to handle complex tasks.
No takes yet. Share an insight, caveat, or question.
Hrishikesh Maluskar (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: