What is GBrain? My open source project is a knowledge system, not RAG in a box.
It gives agents 8 layers that work together to improve memory in a way that makes your already smart OpenClaw or Hermes Agent feel clairvoyant about who you are.
Personal AI becomes possible.
Many users praised Garry Tan's GBrain as a groundbreaking open-source memory system that gives AI agents real persistent knowledge and soul-like recall via Markdown, while others dismissed it as overhyped or unimpressive.
Damn，Garry Tan刚放出来的这个东西,直接把个人AI的天花板捅破了🤯
这意味着所有AI Agent一直以来都有的致命缺陷,现在终于被解决了。
Garry Tan昨天发布的GBrain,不是又一个RAG玩具，我们可以认为它是一个完整的个人知识操作系统。
给你的OpenClaw 或者Hermes 或者Claude Code装上它,就能一直记得你的人际关系,你的决策轨迹,你的长期认知演化。
大多数人以为RAG只有4层， GBrain直接把它干到了8层，
前4层是基础检索的全面升级， 真正的杀手级差异在后面4层，
相当于把个人AI从每次对话从零开始,升级成终身记忆加自我进化。
Garry自己的生产环境已经跑了17888页,4383人,723家公司。
而且现在就能用:
1. 把安装链接直接丢给你的Agent，它会自动搞定一切。
2. 命令行版按GitHub上的步骤走,30分钟搞定。
3. 永远先改Markdown再同步,这是最高优先级数据源。
#GBrain #AI代理 #个人AI
What is GBrain? My open source project is a knowledge system, not RAG in a box.
It gives agents 8 layers that work together to improve memory in a way that makes your already smart OpenClaw or Hermes Agent feel clairvoyant about who you are.
Personal AI becomes possible.
GBrain is free open source. It’s MIT license.
You can find it and install it in your agent with one command.
https://github.com/garrytan/gbrain
What is GBrain? My open source project is a knowledge system, not RAG in a box.
It gives agents 8 layers that work together to improve memory in a way that makes your already smart OpenClaw or Hermes Agent feel clairvoyant about who you are.
Personal AI becomes possible.
Garry Tan's GBrain makes the memory write path the reliability boundary for agent systems, because useful context must survive edits, sync, retrieval, and reuse as state.
The repo treats markdown as the source of truth, with Postgres and pgvector underneath the retrieval layer. The concrete problem is a 7,471-file, 2.3GB markdown wiki that becomes painful when git alone is the operating surface. After sync, a human edit can become queryable agent memory with ownership.
The reusable model is simple: agent memory should have a write path, a system of record, and drift tests. GBrain's CLI and MCP surface expose the same operations, while 30+ MCP tools turn the database into an action surface instead of a passive archive.
Serious AI infrastructure keeps moving toward this shape. Bigger prompts can carry more text for one run, but durable agents need state that can be written, audited, searched, and repaired between runs. Memory becomes production data.
Pretty insane that Garry built all this himself while also running the most impactful job on the planet. AI makes millionX engineers possible.
What is GBrain? My open source project is a knowledge system, not RAG in a box.
It gives agents 8 layers that work together to improve memory in a way that makes your already smart OpenClaw or Hermes Agent feel clairvoyant about who you are.
Personal AI becomes possible.
把INSTALL_FOR_AGENTS.md的链接直接丢给Claude Code,它真的会自己克隆安装配置,全程不用你动手。这才是AI应该有的样子。
GitHub链接👇 https://github.com/garrytan/gbrain
Y Combinator代表のGarry TanがオープンソースのAI記憶システム「GBrain」を公開。RAGとは違い、8層構造でエージェントの記憶を強化する設計。OpenClawやHermesエージェントに組み込むと「あなたのことを知り尽くした」パーソナルAIが作れる。MIT ライセンスで無料。
What is GBrain? My open source project is a knowledge system, not RAG in a box.
It gives agents 8 layers that work together to improve memory in a way that makes your already smart OpenClaw or Hermes Agent feel clairvoyant about who you are.
Personal AI becomes possible.
@mycomradio Yes we are using it multiplayer
You can use it over MCP or you can use multiple harness instances with GBrain that all point to the same GitHub knowledge base repo
Or any GBrain server also supports multiple federated repos too (personal + work repos)
Both work
@garrytan The epistemology layer is interesting, not just what's similar, but who believed what and when. That's where it stops being a search tool and starts being something else.
@AYi_AInotes @garrytan 非常谦逊专注的大佬，昨天刚参加他的gbrain活动
@jatingargiitk I started using it with a team of 7 and it’s amazing
Enterprise is probably a constellation of these with federation
@AYi_AInotes j推特，第一骚@acrmrve |
@AYi_AInotes f她好涩s我不行了👉@dadi2412 e
@AYi_AInotes f比她好看的没她骚%比她骚的没她好看@acrmrve (
@AYi_AInotes r线下sao货没人比她sao @dadi2412 \
@garrytan 8-layer memory beats flat RAG only if the routing survives the latency budget. Most production systems quietly collapse to single-layer retrieval. The layer count is the easy part.
i build agents with gbrain. what makes it different isnt just the memory. its that garry turned agent reliability from a prompt engineering problem into an architecture problem. 8 retrieval layers where the agent decides which one to use per task. flat pipelines cant give you that.
@garrytan Durable personal AI needs write paths, conflict resolution, and eviction policy. A memory graph without TTL and provenance becomes a hallucination cache after the third agent handoff.
@garrytan 8 layers of memory that make an agent feel like it actually knows you is what personal AI has been missing
@AYi_AInotes 记忆和能力做更多的分层,第二大脑就越来越像我们自己; 之前折腾的很多方案就是白费; ai时代真的是,很多东西才刚刚学会就过时了.
@garrytan Very similar approach like http://github.com/rohitg00/agentmemory for memory in this.