Gbrain: Open-Source Second Brain for AI Agent Persistent Memory
Generated: 2026-07-08 · API: Gemini 2.5 Flash · Modes: Summary
Gbrain: Open-Source Second Brain for AI Agent Persistent Memory
Clip title: Give Your AI Agent a Second Brain (Gbrain + Hermes Agent) Author / channel: Tonbi’s AI Garage URL: https://www.youtube.com/watch?v=-fSjdYzrFvA
Summary
This video introduces Gbrain, an open-source “second brain” tool designed to augment AI agents like Hermes Agent with a persistent, searchable knowledge base. The presenter highlights a key limitation of traditional agent memory: it only retains information explicitly discussed in chat sessions. External data, such as meeting notes or private documents, remains invisible to the agent, leading to an “amnesia problem” that Gbrain aims to solve by making this broader context accessible.
Gbrain establishes a knowledge base for “your world,” rather than just conversations or a specific domain. Its core data resides in Markdown files within a Git repository, serving as the system of record. This content is then synchronized to a local, in-process PostgreSQL database (PGlite), eliminating the need for external servers or Docker. For retrieval, Gbrain employs a hybrid search mechanism that combines vector search, keyword matching, and a self-wiring knowledge graph capable of discerning relationships between entities like people, companies, projects, meetings, and decisions, all without requiring explicit LLM calls for graph creation.
The architecture of Gbrain is notably decentralized and private, operating with “no server, no tunnel.” Hermes Agent interacts with gbrain serve as a local MCP (Message Control Protocol) sub-process, enabling seamless integration and exposing over 100 type-safe tools for querying and writing. This setup ensures that every layer, from Markdown files to the local database, is inspectable. Installation is streamlined, achievable either manually with bun and an embedding API key (OpenRouter is recommended for cost-effectiveness) or via the Hermes Agent’s automated installation process, including optional “skill packs” for various data ingestion methods.
In practical application, after seeding Gbrain with initial Markdown-based information, the Hermes Agent can leverage Gbrain to answer complex questions with high accuracy and full source citations. The knowledge graph is particularly powerful, automatically identifying and displaying connections between entities, which significantly boosts retrieval precision compared to vector-only methods. The use of an MCP server further optimizes performance, making interactions much faster and smoother by avoiding repeated CLI calls. Ultimately, Gbrain acts as a complementary system to Hermes Memory; while Hermes Memory remembers conversations, Gbrain offers a comprehensive “world catalog” for broader, durable knowledge, providing a cleaner and more effective integration for personal or small-team knowledge management.
Video Description & Links
Description
Hermes can only remember what you actually talked about with it — everything else in your world is invisible to it. Gbrain, a popular open-source second brain, fixes that, and I set it up with Hermes Agent.
Sign up for my FREE weekly newsletter, where I spill my unfiltered thoughts on the latest AI news, cool research, and projects I’m building: https://www.onchainaigarage.com/
🐦 Follow Tonbi on X for real-time AI x blockchain updates! https://x.com/tonbistudio
Gbrain (open-source, 25k+ stars) gives your agent a knowledge base for your whole world — typed pages for people, companies, meetings, and decisions — including everything that never passed through your chat, like Obsidian vaults and meeting notes. I explain how it differs from Hermes memory (which remembers your conversations) and an LLM wiki like my Agent Wikis (which knows a domain), then install it with Hermes: markdown in a git repo synced to a local PGlite database, exposed as 30+ tools over MCP with no server or Docker. Along the way I show the parts that make it click — cited answers, the zero-LLM knowledge graph that maps how everything connects, and why wiring it up as an MCP server beats shelling out to the CLI. I also cover the habit that makes it worth it: search the brain first, write decisions back, keep world-facts out of chat memory. This is a first look — I’ll follow up after using it daily for a while.
Resources: 🔗 Gbrain: https://github.com/garrytan/gbrain
Timestamps: 0:00 - What is Gbrain? (and why memory isn’t enough) 2:01 - Gbrain vs Hermes memory vs an LLM wiki 4:39 - The architecture: markdown + PGlite + MCP 5:18 - Installing Gbrain with Hermes 7:46 - Seeding it + asking cited questions 8:37 - The knowledge graph (what most retrieval can’t do) 9:44 - Adding the MCP server (and why it’s better) 12:04 - The five rules for using it well 13:06 - Writing decisions back to the brain
Have you used Gbrain, or another second-brain setup with your agent? Drop your tips in the comments — I’ll try them. If this was useful, please like, subscribe, and hit the bell! 🦐✨
Want to sponsor Tonbi’s AI Garage? Reach out to us at sponsors@tonbistudio.com for more information.
Gbrain HermesAgent SecondBrain AIAgents MCP KnowledgeGraph AITools Obsidian Productivity
Tags
ai, hermes, gbrain, hermes agent