How it works
Why brain?
Git as storage
Entries are markdown in a git repo. No server to run. No infrastructure to manage. Version history and access control come from git.
FTS5 search
SQLite FTS5 with BM25 ranking, prefix matching, and contextual snippets. Sub-millisecond search. Works offline.
MCP server
5 tools and 2 resources exposed via Model Context Protocol. AI agents can search, read, and publish entries. brain serve starts the server.
Repo ingest
Seed your brain from existing repos. brain ingest <url> scans for markdown, auto-detects titles and tags, and scores freshness at import.
Freshness scoring
Entries scored as Fresh, Aging, or Stale based on recency and read frequency. brain prune archives what's stale. brain restore brings it back.
Knowledge trails
Auto-computed links between entries. brain trail <topic> follows connections across your team's knowledge.
brain connect https://github.com/acme/brain-hub.git
Works with your AI agent
Add to your MCP client config:
5 tools: push_knowledge · search_knowledge · whats_new · get_entry · brain_stats
2 resources: brain://digest · brain://stats
Your agent can search team knowledge, publish findings, and check what's new.
Claude · Copilot · Cursor · Windsurf
Quick start
Requires Node.js 20+ and git.