garrytan

gbrain

AI🌱 NEW PROJECT BOOST#AI Agents#Knowledge Graph#RAG#PostgreSQL#Automation

// summary

GBrain provides a persistent, self-wiring knowledge graph that enables AI agents to store and retrieve complex information across meetings, emails, and documents. The system automatically extracts entity relationships and maintains a structured timeline, allowing agents to answer queries that standard vector search cannot reach. By utilizing a durable job queue and modular skill system, it ensures that agents become smarter and more reliable over time.

// technical analysis

GBrain is a sophisticated knowledge management and agent-orchestration system designed to provide AI agents with a persistent, self-wiring 'brain' that compounds in intelligence over time. By utilizing a hybrid search architecture and a self-wiring knowledge graph, it solves the problem of agent forgetfulness and context loss by automatically extracting typed entity relationships from unstructured data without requiring constant LLM calls. The project prioritizes a 'thin harness, fat skills' philosophy, where complex workflows are encoded into modular, testable skill files, ensuring that the system remains maintainable, durable, and capable of handling deterministic tasks with high efficiency.

// key highlights

Implements a self-wiring knowledge graph that automatically creates typed links between entities like people and companies during page writes.

Features a 'Minions' job queue that provides a durable, Postgres-native way to execute deterministic tasks without the high latency and token costs of LLM sub-agents.

Supports a 'skillify' workflow that allows users to convert agent fixes into permanent, testable, and auditable skills to prevent recurring bugs.

Provides a multi-layer hybrid search engine that combines vector search with graph-based relationships to answer complex queries that traditional RAG systems miss.

Includes a robust health and self-healing suite, such as 'gbrain doctor' and 'skillpack-check', to ensure the system remains operational and consistent.

Offers broad integration capabilities, including an MCP server for tools like Claude Code, Cursor, and Windsurf, as well as support for remote MCP deployments.

// use cases

Automated entity extraction and knowledge graph construction from unstructured data

Durable background job execution for deterministic agent tasks using Minions

Modular skill development and management to prevent agent failure recurrence

// getting started

To begin, you can either deploy GBrain via an agent platform like OpenClaw or Hermes by following the instructions in the INSTALL_FOR_AGENTS.md file, or set it up as a standalone CLI tool by cloning the repository and running 'bun install' and 'bun link'. Once installed, use 'gbrain init' to initialize your local brain database and 'gbrain import' to index your existing markdown files. You can then start querying your knowledge base or exploring the 29 pre-built skills to automate your agent's workflows.