GBrain: The Open-Source AI Memory System for Teams and Individuals

Overview

GBrain is an open-source AI memory system created by Garry Tan, the CEO of Y Combinator. It acts as a durable knowledge graph that allows AI agents to remember, learn, and operate with context over time. Unlike standard systems that forget past interactions, GBrain turns plain-text notes into a structured graph that every AI can access. The core project is free and open-source, but it also offers a paid commercial team workspace for organizations that need shared memory and centralized account management.

Benefits

GBrain solves the common problem of AI agents being smart but forgetful. By default, most agents start every session with a blank slate. GBrain ensures that information from meetings, emails, and voice notes persists and grows over time. The system works through a read-enrich-write-sync loop that runs on its own. It ingests various types of data, detects important ideas, checks its own memory before making external calls, and then provides the user with full context. A key feature is its ability to extract relationships like who works where or who invested in a company using simple rules without needing expensive AI calls every time. The system gets smarter on its own through three main mechanisms. It enriches information about people and topics as they are mentioned more often. It learns from its own mistakes to reduce reliance on external AI models. It also ranks information better by showing frequently referenced details more readily.

Use Cases

GBrain is designed for individuals and teams who want to own their data and control how their AI learns. It is ideal for operators running specific AI agents who value using Markdown files as the source of truth. Self-hosters who are comfortable managing their own infrastructure like Postgres will find it useful. It is also great for people who want to write their own workflows to define how their brain evolves. The system is built for long-term use, rewarding users who invest time over months rather than those looking for quick short-term trials. Users can run it locally on their own machines or set up a remote server to connect with popular AI clients like ChatGPT, Claude, and Perplexity. It even supports voice-to-text features where users can call a phone number, and the AI will pull context from the brain to answer questions and create new memory pages.

Pricing

The core GBrain project is open-source and free to use under the MIT license. Users can download the code and run it on their own hardware without any cost. For organizations, GBrain offers a commercial Team Workspace solution. This option is priced per workspace rather than per person. New team workspaces receive a promotional discount of $100 off their first month. This commercial version provides shared memory for the whole team, centralized account management, and permission levels for different AI tools.

Vibes

GBrain has gained attention in the AI community for its transparent and open architecture. Users appreciate that it uses plain-text Markdown files, making the data easy to read and edit. The system is known for its engineering depth, combining hybrid search methods with a typed-edge graph to deliver accurate results. Benchmarks show it outperforms many vector-only systems in finding relevant information. The community values its focus on data ownership and control over patterns that AI follows. While it requires more setup than managed cloud alternatives, serious AI users see it as a powerful tool for building a personal or team brain that truly compounds in value over time.

Additional Information

GBrain was created by Garry Tan, the President and CEO of Y Combinator, a well-known startup accelerator. The project is hosted on GitHub and is licensed under the MIT license, allowing anyone to use, modify, and distribute the code. The system includes production-grade infrastructure features like a job queue called Minions and durable agents that can recover from crashes. It connects to AI clients via the Model Context Protocol and supports OAuth 2.1 for secure authentication. The installation process is designed to be agent-driven and can take about 30 minutes to set up locally. The project emphasizes a philosophy of thin harness and fat skills, where intelligence lives in the workflow files rather than a complex runtime engine.