@vectojs/knowledge-graph

RDF-driven domain layer on @vectojs/graph3d. Rendering and force layout stay in graph3d; this package adds:

• KgDataSource — lazy getNodes / getNeighbors adapter contract
• MemoryDataSource — in-memory implementation for tests and small graphs
• KgEntity / KgFact — typed entities with multi-language labels + provenance
• FixedZLayout — planar (z = const) wrapper around VectoForceLayout for 2D
• parseRdfTurtle — N3-backed Turtle → KgGraphData
• KnowledgeGraphSession — wires source → layout → Graph3D → GraphCamera → GraphInteraction

Install

bun add @vectojs/knowledge-graph three

three is a peer; @vectojs/graph3d and n3 are direct dependencies.

Minimal 2D session

import * as THREE from 'three';
import { KnowledgeGraphSession, MemoryDataSource, parseRdfTurtle } from '@vectojs/knowledge-graph';

const data = parseRdfTurtle(myTurtle);
const source = new MemoryDataSource(data);
const canvas = document.querySelector('canvas')!;

const session = new KnowledgeGraphSession({
  domElement: canvas,
  source,
  mode: '2d',
  focusIds: [data.entities[0]!.id],
});

const renderer = new THREE.WebGLRenderer({ canvas });
const scene = new THREE.Scene();
session.attach(scene);

function frame() {
  session.tick();
  session.render(renderer);
  requestAnimationFrame(frame);
}
frame();

Design rules

• Acyclic deps: knowledge-graph → graph3d only (plus n3 / three).
• a11y is onDemand-only in the host app — do not project one DOM node per entity.
• No WASM force kernel here; layout performance lives in graph3d's JS Barnes-Hut.

See references/graph/RESEARCH.md in the workspace for the ecosystem baseline.