AI Knowledge Repository

A Git and Markdown knowledge base where raw notes stay untouched, and AI agents suggest observations, principles and playbooks for a person to review.

Why I started

AI assistants forget everything between conversations. The usual fix is to hand them all your notes. And that lets them rewrite your notes without anyone checking. I wanted a structure where knowledge builds up slowly and stays traceable, and where the AI acts as an editorial assistant rather than an author.

What it does

• Keeps raw material such as conversations, documents and notes as untouched evidence.
• Has AI agents extract small, sourced pieces from that evidence: observations, insights, decisions, principles, mental models and playbooks.
• Holds every proposal in a dated review until a person accepts it.
• Answers questions from accepted knowledge, citing the files it used.
• Opens as an Obsidian vault, so the whole thing can be browsed by hand.

Approach

Plain Markdown in Git, with no special database. Everything can be traced, every change can be reviewed, and agents read summaries and maps before loading anything in full.