booboo-deploy

Deploy a brain, end to end

Read RULES.md in this pack first. The walls rule and the verify rule both bite during deployment.

The path

npx create-booboo my-brain     # scaffold: config + org + AGENTS.md + ONBOARDING.md
cd my-brain && npm install
npm run build                  # booboo.config.yaml → the snapshot
npm run serve                  # REST on :8787
npm run mcp                    # MCP over stdio

The scaffold ships a JSON starter with a commented Postgres upgrade path, so a brain exists before any database is involved. Prove the loop on the starter, then repoint it — debugging a config and a connection string at the same time is how deployments stall.

Shape the config

title: "My System"
root: { id: core, type: root, label: "MY SYSTEM" }

layers:                                   # order is Z-order, first is back
  - { name: agents,    color: "#c9a04a", label: "AGENTS" }
  - { name: memory,    color: "#a78bd0", label: "MEMORY" }
  - { name: knowledge, color: "#4ECDC4", label: "KNOWLEDGE" }

walls: [private, sealed]                  # filtered IN THE BUILDER, before anything exists

sources:
  - adapter: postgres
    url: ${DATABASE_URL}
    nodes:
      - { table: agents,       layer: agents,    id: slug, label: name, parent: core, weight: 0.6, icon: emoji }
      - { table: observations, layer: memory,    id: id,   label: title, cluster: project, where: "kind <> 'noise'" }
      - { table: kg_entities,  layer: knowledge, id: id,   label: name, weight_from: degree }
    links:
      - { table: edges, source: src, target: dst, type: rel }

  - adapter: json
    path: ./data/extra.booboo.json

output:
  snapshot: ./build/booboo.json

${VAR} reads from the environment or a .env. Nothing secret belongs in this file — it is committed, and a connection string with a password in it is a credential in git history forever.

Layers are yours to choose. The three above are a sensible default, not a schema. Pick planes that match how the system actually stacks.