agent-memory

A file-based memory system for AI coding agents. Plain markdown, four rules, one scheduled job.

Every AI session starts from zero. You spend the first twenty minutes explaining who you are, what you're working on, and what you decided last week. Then the window closes and it all evaporates. The model is capable and has amnesia, and you are its short-term memory.

This is the structure I built so I stopped being the memory. It has been running my daily work for months. Most of what makes it work isn't AI — it's the discipline a knowledge base or a well-run codebase needs: decide what's true, write it down once, and guard it against drift.

The repo is the system with the content removed. Clone it, fill in your own files, point your agent at CLAUDE.md.

Why not just a bigger prompt

The instinct when a model doesn't know your context is to paste more into the prompt. That works for one session and fails across many:

• You can't remember what to paste.
• The paste grows every week.
• Half of it is stale by the time you send it.
• Worst of all, you don't know which version of a fact the model is using, because you've pasted three slightly different versions over three sessions.

The problem isn't context window size. It's that there's no durable place for context to live between sessions, and no rule for which copy of a fact wins when two disagree. Fix those two and the amnesia mostly goes away.