GBrain - A personal production-grade AI memory system open-sourced by the Y Combinator president

What is GBrain

GBrain is a personal production-grade AI memory system open-sourced by Y Combinator president Garry Tan, designed specifically for OpenClaw, Hermes Agent, and other AI Agent platforms. The tool transforms scattered Markdown notes into a searchable, reasoning-capable, continuously growing knowledge graph, giving Agents long-term memory. Garry himself uses the tool to manage 10,000+ files, 3,000+ contacts, and 13 years of calendar data. The system already has 17,888 pages, 4,383 people, 723 companies, and 21 scheduled tasks running fully automatically, built in 12 days.

GBrain's main features

• Automatic signal capture: Each message triggers a cheap model to extract original opinions and entity mentions in parallel, automatically archived in the background.
• Brain-ops priority query: The Agent checks the brain before answering; if it cannot find the answer, it says so directly rather than hallucinating.
• Multi-source content ingestion: Supports automatic digestion of all formats including meetings, emails, Twitter, PDFs, videos, and GitHub repositories.
• Automatic entity enrichment: People/companies are automatically upgraded based on mention count (1 mention stub → 3 mentions online lookup → 8 mentions/meeting full profile).
• Compiled Truth + Timeline: The upper layer is a rewritable current best understanding, and the lower layer is an append-only evidence timeline.
• Hybrid search: Vector + keyword + RRF fusion + multi-query expansion + 4-layer deduplication, achieving 95% Recall@5.
• Knowledge graph self-linking: Every write automatically extracts entity relationships with zero LLM calls (attended/works_at/invested_in, etc.).
• Minions background queue: Deterministic tasks go through the Postgres-native job queue, completing in 753ms, with zero token cost and 100% success rate.
• Skillify skill management: Automatically generates SKILL.md + tests + evaluations + parser triggers, preventing the skill tree from becoming a black box.
• Voice calls: Integrates with Twilio + OpenAI Realtime; incoming calls automatically pull the caller's profile and generate a call brain page.
• Daily briefing: Automatically summarizes meeting context, active deals, citation tracking, and to-dos.

GBrain's technical principles

• Three-layer architecture separation: GBrain adopts a separated architecture of Brain Repo (Git-hosted Markdown files, the only human-readable and editable data source) → GBrain retrieval layer (Postgres + pgvector hybrid search engine) → AI Agent (26 Skill definitions for read/write logic). The Agent reads and writes through the BrainEngine interface, with underlying support for PGLite (embedded Postgres, zero configuration) and PostgresEngine (Supabase) with bidirectional hot migration.
• Compiled Truth + Timeline dual-layer knowledge model: Each Markdown page is divided by --- into two layers: above is Compiled Truth (the current best understanding, continuously rewritten as new evidence arrives), and below is Timeline (the raw, append-only evidence chain). This design allows cognition to evolve while preserving a complete audit trail, avoiding the loss of history from purely overwriting updates or query noise from purely appending.
• Zero-LLM-call knowledge graph self-linking: After each page write, a deterministic regex pipeline automatically extracts entity references (Markdown links + bare slugs), infers typed relations (attended/works_at/invested_in/founded/advises), and completes deduplication, dead link cleanup, and backlink updates. The entire process involves zero LLM calls, implemented purely with local regex + a priority inference cascade.

How to use GBrain

• Agent automatic installation (recommended): Paste the instruction into OpenClaw or Hermes Agent Retrieve and follow the instructions at: https://raw.githubusercontent.com/garrytan/gbrain/master/INSTALL_FOR_AGENTS.md, and the Agent automatically clones the repository, installs dependencies, loads 26 Skills, and configures scheduled tasks, completing in about 30 minutes.
• Standalone CLI：git clone https://github.com/garrytan/gbrain.git && cd gbrain && bun install && bun link, then gbrain init initialize the local brain (PGLite, ready in 2 seconds),gbrain import ~/notes/ import notes,gbrain query "..." query.
• MCP Server: Locally expose 30+ MCP tools via gbrain serve to connect to Claude Code/Cursor/Windsurf; after remote deployment to Supabase, connect to Claude Desktop/Cowork/Perplexity via a fixed ngrok domain.
• Data ingestion: Run gbrain integrations to view and configure integration recipes for Gmail, Calendar, Twitter, meeting transcripts, etc.

GBrain's key information and usage requirements

• Environment requirements: Node.js + Bun (recommended), PGLite runs locally with zero configuration, or Supabase Postgres ($25/month) for multi-device sync.
• API Key: Requires configuration of OpenAI (embeddings), Anthropic (optional, for sub-agents), Groq (voice transcription), etc.
• Storage: Markdown files are the single source of truth, directly editable by humans,gbrain sync automatically syncs.
• Relationship with GStack: GStack (80k+ Stars) teaches Agents to write code, GBrain teaches Agents to take notes and think, and they can be combined for use through the hosts/gbrain.ts bridge.
• License: MIT open source, available for commercial use.

GBrain's core advantages

• Production-grade validation: Used daily by the YC president personally, not a toy-level demo, but a workflow forged in real combat.
• Self-evolving knowledge graph: Entity relations are automatically extracted with zero LLM calls, and search accuracy improves by 28.8 percentage points over pure vectors.
• Deterministic background tasks: Minions separate deterministic work from judgment work, reducing cost from $0.03 per run to $0 and increasing success rate from 0% to 100%.
• Layered knowledge model: Compiled Truth ensures cognition evolves in real time, and Timeline ensures the evidence chain is never lost.
• Agent-native design: Installation, operations and maintenance, and extension are all designed for AI Agents rather than manual human operation.

GBrain's project address

• Project official website：https://github.com/garrytan/gbrain

Comparison of GBrain with similar competitors

GBrain application scenarios

• Investor/founder relationship management: Automatically archives meetings, emails, and social updates, and generates a complete profile and interaction history of the other party 30 seconds before a meeting.
• Personal knowledge base: Transforms years of notes, articles, and ideas into a reasoning-capable knowledge graph, supporting queries such as “my view on something six months ago.”
• Research automation: After ingesting papers, reports, and videos, automatically extracts entities, builds citations, and generates timelines, assisting academic or industry research.
• Content creation assistance: Tracks inspiration sources, automatically completes citations, and maintains topic tag consistency, preventing information omissions during the creation process.
• Smart schedules and tasks: Automatically prepares meeting briefs daily, inspects expired pages, fixes broken citations, and generates periodic reports.