Rolepod Brain — AI skill for Claude Code

rolepod-brain

Persistent memory for AI coding agents. One binary, one SQLite index, a git-versioned markdown wiki — and **nothing running between events**.

Your coding CLI forgets everything when the session ends. rolepod-brain captures what happened, consolidates it into readable pages, and hands the next session a short list of pointers it can pull from. Install it once and stop thinking about it.

What makes it different

**Your brain never leaves your machine.** No cloud, no remote, no telemetry, no account. There is no sync command that uploads anything, because there is nowhere for it to upload to — see below for why that is a design decision rather than a missing feature.

**No resident process, ever.** No daemon, no server, no supervised worker, no port. Hooks spawn a short-lived process that exits; the MCP server lives exactly as long as one session; consolidation runs when a session boundary fires. After a reboot there is nothing to start.

$ brain doctor
ok   no resident process  nothing of ours is running

**No API keys. No tokens. No provider configuration.** Summaries are written by whichever CLI you are already signed into, through that vendor's own supported headless entry point (`claude -p --model haiku`, `codex exec`). This project never holds a credential, because it never has one to hold. If no model is reachable it writes rule-based summaries instead — a permanent, first-class mode, not a broken one.

**Pointers in, content pulled.** Automatic injection carries titles and ids against a hard byte budget. Full content is never pushed into your context; the agent calls `brain_search` / `brain_get` when the task actually needs a body. `brain doctor` reports the real bytes spent, not the configured limit.

**Found by meaning, not only by words.** A session that recorded `login token expiry` is one someone later searching `auth` n