Garry Tan has released GBrain as an open-source, structured knowledge system used to power his personal AI agents. The project repository is published at github.com/garrytan/gbrain.
GBrain is neither a notes app nor a conventional RAG (retrieval-augmented generation) pipeline. Its central idea is to avoid re-deriving knowledge on every query by maintaining a pre-computed, self-updating "compiled truth" for each entity. Agents therefore receive richer context on each request, and the system compounds over time as it ingests meetings, emails, and calls. According to Tan, he built the system in 12 days.
Key features
-
MECE knowledge structure: information is organized into mutually exclusive, collectively exhaustive categories — people, companies, deals, meetings, projects, concepts, originals, and media. Each page displays a compiled summary at the top and an append-only evidence trail beneath.
-
Self-wiring graph: every page write automatically extracts entity references and creates typed links (for example: attended, works_at, invested_in) without any LLM calls. That allows the system to answer relational queries — e.g., "who works at Acme AI?" — that vector search alone may miss.
-
34 built-in skills: capabilities range from signal detection to content ingestion to research synthesis. The intelligence lives in markdown skill files rather than the runtime; this defines how agents act, not just what they remember.
-
Auto-enrichment tiers: one mention creates a stub; three mentions trigger web enrichment; meeting someone in person or eight-plus mentions initiates the full research pipeline. The brain decides how much attention an entity deserves.
Usage and deployment
GBrain can run as a standalone CLI, as an MCP server for Claude Code and Cursor agents, or be deployed with one click on OpenClaw or Railway. The code and deployment instructions are available at github.com/garrytan/gbrain.
Why it matters
The approach aims to give personal assistants a persistent, explorable background knowledge that reduces the cost and uncertainty of regenerating context on each query. Structured, entity-centered storage plus automatic enrichment lets the system grow more useful over time.
Current snapshot
At the time of the open-source release, GBrain contains more than 17,000 pages, tracks over 4,000 people, and runs 21 autonomous cron jobs. The initial build was completed in 12 days and the codebase is now available for the community to inspect and extend.
Conclusion
GBrain represents a structured, self-maintaining knowledge base designed for personal AI agents. By open-sourcing the project, Garry Tan enables other developers and teams to adopt or adapt the model for their own assistant workflows.
