Brainery

The keyline

“You can outsource your thinking, but you cannot outsource your understanding.”

— @yacineMTB, cited by Karpathy, 2026-04-30

A /skills folder for your company.

Connect Slack, Notion, Zendesk, GitHub. Brainery ingests the knowledge, audits it for contradictions, and compiles executable skill files — versioned, sourced markdown that any AI agent reads before doing work.

Today’s chatbots outsource thinking. Brainery encodes understanding — the rules, authorities, sources, exceptions — so agents make sourced decisions instead of confident guesses.

See the demo →Read the code

The 7 layers

  1. Connectors — Slack, Notion, Zendesk, GitHub via Maton + MCP
  2. Ingestion + entity extraction — reused from GBrain
  3. Wiki + knowledge graph — PGLite or Postgres+pgvector
  4. Coherence engine ⭐ — catches contradictions, missing sources, staleness
  5. Skill compiler ⭐ — emits versioned /skills/*.md
  6. Brain server (MCP + HTTP) — agents call brain.get_skill
  7. Approval workflow — human-in-the-loop for policy changes

Coherent brains. Auditable agents.