garrytan/gbrain

Traditional RAG and personal-knowledge tools return lists of chunks or pages, forcing the user to read and synthesize themselves. GBrain solves this by generating direct, well-cited answers with gap analysis, and by building a typed entity graph (people, companies, deals) without LLM calls, enabling relational queries that vector search alone cannot answer. It also addresses memory persistence for coding agents, which otherwise forget non-code context.

Developers can build autonomous agents that maintain a long-term institutional memory, answer complex relational questions (e.g., 'who works at Acme?', 'what did Bob invest in?'), and prepare for meetings with synthesized briefs. It supports team-scoped access control for a company brain, and can be wired into Claude Code or Codex as a retrieval layer. The ceiling includes running hundreds of cron jobs for continuous ingestion, enrichment, and overnight memory consolidation, effectively creating a self-improving knowledge base.