GBrain — персональная база знаний с синтезом ответов

vibe-coding

gbrain is a Claude Code skill that adds a synthesis-and-memory layer to AI agents, combining knowledge-graph traversal, vector search, and gap analysis in a single self-hosted tool. Built by Garry Tan, President and CEO of Y Combinator, it powers his OpenClaw and Hermes deployments — 155,795 pages, 24,589 people, 5,340 companies, and 66 autonomous cron jobs ingesting meetings, emails, tweets, voice calls, and notes overnight. Instead of returning a list of matching chunks, GBrain produces cited prose answers and explicitly flags what the brain does not yet know, so agents and users can act on the gaps rather than be surprised by them. The self-wiring knowledge graph extracts typed edges (attended, works_at, invested_in, founded, advises) with zero LLM calls, benchmarked at P@5 49.1% and R@5 97.9% on a 240-page corpus — +31.4 points over its graph-disabled variant. Setup takes about 15 minutes on the Claude Code path; the database starts in 2 seconds on PGLite with no server required.