@booboo-brain/build

@booboo-brain/build

Turn any database into a Booboo graph with a YAML config — no glue code.

Use

booboo build --config booboo.config.yaml

Each source is read through an adapter, then the engine merges them, applies privacy walls, auto-wires a spine edge for every parent, drops dangling links, validates against @booboo-brain/spec, and writes the snapshot.

Config

title: "My System"
root: { id: core, type: root, label: CORE, layer: agents }
layers:
  - { name: agents, color: "#c9a04a" }
  - { name: memory, color: "#a78bd0" }
walls: [secret, private]            # a node whose `cluster` is listed here is NEVER emitted
sources:
  - adapter: postgres
    url: ${DATABASE_URL}            # ${ENV_VAR} is substituted from the environment
    nodes:
      - { table: agents, layer: agents, id: slug, label: name, parent: core, weight: 0.6, icon: emoji, cluster: team }
      - { table: notes,  layer: memory, id: id,   label: title, cluster: project, where: "archived = false" }
    links:
      - { table: edges, source: src, target: dst, type: rel }
  - adapter: json                   # merge an existing Booboo file as-is
    path: ./extra.booboo.json
output:
  snapshot: ./my.booboo.json

Node mapping — id label cluster icon color weight_from are column names; weight tier parent type prefix are literals. weight_from is min-max normalised to 0..1. where is raw SQL (trusted config).

Adapters — postgres (read-only; Supabase / Neon / local; SSL auto-on for remote hosts) and json (passthrough merge). Add one by exporting (src) => { nodes, links } and wiring it into build.ts.

API

import { buildFromConfig, build, loadConfig } from "@booboo-brain/build";
const { graph, snapshotPath } = await buildFromConfig("booboo.config.yaml");

Layout is the viewer's job, so the emitted graph carries no positions — it stays small and portable. Feed the snapshot to @booboo-brain/viewer, serve it over the REST/MCP layer, or commit it as a fixture.