Y Combinator CEO Garry Tan Open Sources Personal AI Second Brain GBrain
Y Combinator CEO Garry Tan shared his personal compounding system GBrain, built with AI over the past 5 months, which is now fully open source.
The system includes approximately 100,000 pages of a structured knowledge base (people, companies, meetings, books, articles) and implements automated workflows through over 100 "fat skills" (skill files), including Book-Mirror (mapping books chapter by chapter to real life), automatic meeting capture and dissemination, Meeting-Prep, etc.
GBrain adopts a "Thin Harness + Fat Skills + Fat Data" architecture, triggered solely through Slack/chat, enforcing public or searchable interactions. All skills are reusable and self-iterative, significantly enhancing personal knowledge compounding and work efficiency.
Source: Public Information
ABAB AI Insight
Garry Tan observed senior engineers at Siemens learning programming in his early years. The design of GBrain directly continues the "Lehrwerkstatt" (teaching workshop) concept, structuring personal knowledge graphs, meeting notes, and reading history, and achieving "skill self-generation" through Skillify meta-skills—each manual workflow can be automatically distilled into reusable skills.
In terms of capital, Tan invested personal time and coding effort into building GBrain (OpenClaw/Hermes Agent + GBrain), allowing the system to self-enhance 24/7 through cron tasks, entity dissemination, and cross-model evaluation (Opus/GPT-5.5/DeepSeek), forming a personal neural system far exceeding a single ChatGPT, attracting other founders to replicate similar compounding infrastructures.
Similar to Andrej Karpathy's LLM Wiki concept and Shopify River's open apprenticeship model, personal AI is currently in the early stage of transforming from a chat window tool to a persistent, structured, self-evolving second brain. Garry Tan's open-source stack provides YC founders with a directly deployable template.
Structural judgment: Essentially a technological substitution. GBrain replaces traditional personal notes and memory through fat skills and a persistent knowledge graph, with the mechanism of Skillify meta-skills allowing AI to shift from passive execution to active distillation and reuse, driving personal productivity from "using AI" to "co-constructing a compounding system with AI," concentrating capital and time on a few high-execution individuals capable of building personal OSs.
ABAB News · Cognitive Law
The model is the engine; what you build is the car.
A private window enhances one person, while structured compounding amplifies a lifetime.
Whoever turns AI into their neural system will possess a decade of unseen advantage.