gbrain provides source-backed memory that can be shared across connected AI agents. It supports correction and withdrawal of stored facts, hybrid retrieval, cited answer synthesis, and graph-based relationship queries.
Project overview
gbrain brings together explicit memory controls, cited synthesis with knowledge-gap notes, and knowledge-graph wiring that does not require LLM calls, making it relevant for developers who need inspectable knowledge across agents.
Project type
AI Agent · RAG · AI Search
Use cases
Knowledge Q&A · Automation
Deployment
Refer to project documentation
License
MIT
Best for
Developers, AI engineers, and teams that need source-backed memory shared across agents, with explicit controls for correcting or withdrawing stored facts.
Key capabilities
Stores explicit facts with source provenance, supports correction and withdrawal, and exposes the same memory to multiple agents.
Extracts entity references during page writes, creates typed edges without LLM calls, and supports multi-hop graph queries.
Produces synthesized prose answers with citations and notes gaps in the stored knowledge.
Limitations and risks
The keyless mode provides keyword memory without a model API key. Semantic search and automatic fact extraction require optional provider configuration, with OpenAI, Voyage, or Anthropic documented for capability upgrades.
The full always-on OpenClaw or Hermes path incurs server costs, while raw API-token usage scales with agent activity.
Memories saved by agents default to brain-wide visibility unless private visibility is explicitly selected, so visibility settings should be reviewed before use.
Getting started
Bun 1.3.11 or newer is required. A lighter local brain and MCP connection path is documented as two commands, but there is no single documented setup route covering every deployment.
Evidence and sources
README: GBrain stores explicit facts with their sources, supports corrections and withdrawal, and makes the same memory available across your agents.
README: **A synthesis layer that gives you the actual answer.** Synthesized, well-cited prose across people, companies, deals, and ideas. Not "here are 10 chunks that mention your query";…
README: Every `put_page` extracts entity refs from markdown/wikilinks/typed-link syntax and writes edges with zero LLM calls. Typed edges (`attended`, `works_at`, `invested_in`, `founded`…
README: It's the production brain behind my OpenClaw and Hermes deployments: **155,795 pages, 24,589 people, 5,340 companies**, 66 cron jobs running autonomously.