GBrain

GBrain is an open-source personal knowledge base engine designed to be installed and operated by AI agents. It indexes markdown files into a local or cloud Postgres database with pgvector, enabling hybrid search (vector + keyword + RRF fusion) so agents can read context before every response and write new knowledge after every conversation. The system supports a "compiled truth + timeline" knowledge model, 30+ MCP tools, and integration recipes for Gmail, Google Calendar, Twitter, Twilio voice calls, and meeting transcripts. Setup takes approximately 30 minutes with a frontier model like Claude or GPT handling the installation automatically.

• PGLite local brain — Run gbrain init for an embedded Postgres 17.5 brain via WASM with no server, no accounts, and no connection strings required.
• Hybrid search with RRF fusion — Combines vector (HNSW cosine) and keyword (tsvector) search with Reciprocal Rank Fusion and multi-query expansion via Claude Haiku for best-in-class recall.
• 30 MCP tools — Expose get_page, put_page, search, query, traverse_graph, sync_brain, and more to any MCP client including Claude Code, Cursor, Windsurf, and Claude Desktop.
• Integration recipes — Agent-readable markdown recipes for Gmail, Google Calendar, Twitter/X, Twilio voice-to-brain, and Circleback meeting transcripts; the agent reads the recipe, asks for API keys, and wires everything up.
• Compiled truth + timeline model — Every brain page stores a rewritable "compiled truth" section above a separator and an append-only evidence timeline below, keeping facts current while preserving provenance.
• Pluggable engine architecture — Swap between PGLiteEngine (local, zero-config) and PostgresEngine (Supabase Pro) with a single gbrain migrate --to supabase command.
• File storage migration — Three-stage lifecycle (mirror → redirect → clean) moves binary attachments to S3-compatible or Supabase Storage while preserving all markdown references.
• TypeScript library — Import createEngine from gbrain to embed the brain directly in any TypeScript/Bun project with the same operations as the CLI and MCP server.
• GBRAIN_SKILLPACK — A reference playbook telling agents when to read, when to write, how to detect entities, run enrichment pipelines, and maintain the brain autonomously via cron jobs.