The Problem
Claude Code is smart, but its memory is a joke. You can have a 200K context window and still lose every decision, bugfix, or dumb Slack argument between sessions. It’s like paying for an AI intern who forgets everything by morning. If you want persistent memory without duct-taping a database, you’re out of luck.
What This Does
claude-brain fixes this with one file: .claude/mind.mv2. No databases, no cloud, no API keys, no “just spin up a Postgres instance.” The core is native Rust for speed—sub-millisecond searches, even with thousands of memories. The guts live in src/core/mind.ts and the CLI hooks are in src/scripts/. Want to see what gets injected? Check src/hooks/session-start.ts and src/hooks/post-tool-use.ts. Every memory, decision, and bug is auto-captured and versionable. You can literally git commit your AI’s brain and hand it to a new teammate.
Real-World Use
You’re halfway through a project, and the auth bug comes back. Instead of slamming your head on the desk, you run /mind search "auth" in Claude Code. It pulls up every relevant fix, decision, and Slack rant, instantly. Or, if you’re a CLI junkie, memvid find .claude/mind.mv2 "JWT" spits out the timeline in seconds. The memory file stays under 5MB for a year of use, so you can toss it around with scp or drop it in a repo without worrying about bloat.
The Bottom Line
If you want persistent AI memory without the usual database nonsense, claude-brain is the simplest solution I’ve seen. It’s fast, dead simple, and actually practical. If you’re running a one-person project, maybe overkill—but for teams and serious workflows, it’s a no-brainer. Just don’t expect fancy dashboards or analytics. It’s a brain in a file, not an ERP suite.