Verified project record

garrytan/gbrain

gbrain is an open-source memory application that gives AI agents persistent, source-cited facts with support for corrections, withdrawal, and cross-agent sharing. It can return a synthesized, well-cited answer along with an explicit note on what it does not yet know.

Project overview

Instead of returning a list of retrieved chunks, gbrain returns an actual cited answer with explicit gap analysis, addressing the persistent-memory gap agents face today.

Project type: MCP · AI Agent
Use cases: Knowledge Q&A · Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers and AI engineers who want to give an existing agent persistent, source-cited memory with corrections, withdrawal, and cross-agent sharing.
Users comfortable with terminal commands and some coding, since installation and use are CLI-based and setup is rated medium difficulty.

Key capabilities

Stores explicit facts with sources, supports corrections and withdrawal, and the memory can be shared across agents; keyword retrieval works keyless.
Returns a synthesized, well-cited answer plus an explicit note on what the brain does not know yet.

Limitations and risks

Markdown export from the brain is not a full database backup, so it should not be relied on as one.
When sharing a brain, local files and shared database credentials form a different trust boundary; sources alone do not isolate callers, so consider this before sharing memory across parties.

Getting started

Requires Bun 1.4.4+ and terminal commands. Install with `bun install -g github:garrytan/gbrain`, then run `gbrain init --pglite --no-embedding`, and verify the environment with `gbrain doctor`. The README states "Install in 30 minutes"; setup difficulty is medium.

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: 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.
README: A synthesis layer that gives you the actual answer. Synthesized, well-cited prose across people, companies, deals, and ideas.
README: Alongside keyword and semantic retrieval, GBrain offers two optional ways to use your stored knowledge
README: bun install -g github:garrytan/gbrain gbrain init --pglite --no-embedding # keyless; no server, no Docker gbrain doctor # verify health gbrain import ~/notes/ --no-embed # index y…

Source updated at: 2026-10-29T18:09:03Z

Open original GitHub project