gbrain
gbrain is Garry Tan’s opinionated pattern for an LLM-maintained personal knowledge base. We rate it trial: the ideas are worth stealing—especially two-layer pages and MECE directory resolvers—but the full package is prescriptive (agent-maintains-everything, enrichment on every signal). Cherry-pick patterns and adapt them to your vault and workflows; do not treat gbrain as a religion.
Blurb
A personal intelligence system where your AI agent builds and maintains an interlinked wiki of everything you know — people, companies, deals, projects, meetings, ideas — as structured, cross-referenced markdown files. The agent writes and maintains all of it. You direct, curate, and think.
Summary
Worth applying: pre-computed synthesis above a --- separator with an append-only Timeline below; directory README.md resolvers that say what belongs where; one primary home per entity with wiki-links for adjacency.
Take with skepticism: blanket “enrichment fires on every signal,” full automation of maintenance, and any rule that does not match how you file work (e.g. this vault uses PARA zones plus garden notes, not a single gbrain tree).
This vault: we adopt the two-layer pattern for research and garden notes (Agent Skills, tech-garden items); we assess a root RESOLVER.md and always-on enrichment pipelines before copying them wholesale.
Portable patterns (not dogma)
Two-layer pages (compiled truth + timeline)
Above ---: current synthesis—rewrite when facts change. Below ---: append-only edit log (**YYYY-MM-DD** | Author — …). Subject-matter history belongs above the line; the timeline records document changes only.
MECE directories (light touch)
Resolvers in zone README.md files help agents and humans file consistently. MECE applies to directories, not reality—use cross-links when entities span domains.
Enrichment on every signal (optional)
Powerful for high-touch CRM-style brains; optional here. Prefer explicit jobs (Dagu, morning briefing) over implicit always-on writes.
Page template (adapt as needed)
| |