Brain.Md

BRAIN.md is a sharp, opinionated answer to a problem every developer using Claude Code or Codex hits: the agent forgets everything between sessions. The core insight is to treat memory as a folder of Markdown files, not a service. That gives you three big advantages. First, zero lock-in—the brain is just files, so you can move it with your repo, read it with any Markdown tool, and switch agents freely. Second, provenance—every decision is timestamped, every change is in Git, so you can diff and audit why things were built this way. Third, simplicity—no daemon, no MCP server, no cloud bill; you just need Node.js. The atomic write guarantee via the CLI is a nice safety net: you can't accidentally rewrite a truth without leaving a timeline entry, which keeps the evidence chain intact. The setup script that installs skills for Claude Code and Codex lowers the barrier to entry for the two biggest coding agents. The main weakness is the manual discipline it demands. BRAIN.md won't work if you don't remember to run 'brain update-truth' after each meaningful decision. It's not an automatic memory system; it's a structured note-taking tool that's optimized for AI consumption. If you're the kind of developer who writes extensive documentation for your future self, you'll love it. If you want set-and-forget memory, you'll be frustrated. It's also not a replacement for a knowledge base with semantic search; the timeline is append-only and flat, so if you need to query across pages, you'll rely on grep and lint-links. For solo developers and small teams who are already disciplined, this is a great fit. For larger orgs that need centralized memory and access control, a cloud service like Mem0 might be more appropriate, though it costs money and adds a dependency.