Build ontologies and knowledge graphs from documents.
OntoLab is an open-source workspace for defining a domain’s vocabulary and turning documents into facts you can trace to their sources. Build and align ontologies on their own, or use a saved ontology to guide knowledge graph extraction.
How it works · Two workflows · Quick start · Architecture · Documentation · Contributing
- Every fact is traceable. Extracted facts keep the source passage they came from, so reviewers can check evidence before accepting it.
- Requirements first. Competency questions define what the ontology must answer and are reused to test the ontology and the graph.
- People stay in control. Models propose classes, mappings, and facts; you accept, reject, or edit them and save versions you can compare.
- Standards in, standards out. Import Turtle, RDF/XML, or JSON-LD, validate with SHACL and SPARQL, and export RDF or publish to a graph database.
- Runs where you work. Use it in the browser or as a desktop app, with local CLIs, Ollama, or API providers, and local or Postgres storage.
The two workflows share one project. A saved ontology guides knowledge graph extraction, and new documents can feed back into the ontology. Each step that uses a model produces proposals for review, and nothing is saved as a version until you choose to save it.
An ontology defines the kinds of things your graph can describe and how they relate.
- Build from documents. Upload sources, define the questions your ontology should answer, and generate a draft of classes and properties. Use new documents to refine an existing ontology. Build an ontology from documents
- Edit directly. Start from a template or import Turtle, RDF/XML, or JSON-LD. Use the editor to add or change classes, properties, and relationships. Manual editing
- Align and merge. Compare ontologies or standards, review proposed matches, and save the mappings you accept. Ontology alignment
A knowledge graph records the specific entities and facts found in your sources.
- Extract from documents. Choose a saved ontology and extract entities, attributes, and relationships from PDFs, Word documents, spreadsheets, and other supported files. Build a knowledge graph
- Review and edit facts. Inspect source passages, resolve duplicates, and accept, reject, or edit proposed facts. You can also edit graph entities and relationships directly. Review evidence · Edit graph facts
- Validate and use the graph. Save a reviewed version, ask questions with citations, or export RDF as Turtle. Ask questions · Export a graph
Install Node.js 22.22 or newer and choose a model provider. If you already use Codex CLI or Claude Code, install and sign in through the official CLI on the same machine and account where you’ll run OntoLab. Local CLI setup
git clone https://github.com/IndustriAgents/OntoLab.git
cd OntoLab
npm ci
npm --prefix website ci
STORE_MODE=local npm run devOpen localhost:5173 and select Codex or Claude Code under Local CLIs in the model picker when setting up a run.
- Try graph extraction first. Choose Open the sample project to start with a synthetic technical manual and a saved ontology. Sample walkthrough
- Build an ontology first. Create a project, open Ontology → Build & sources, and choose From documents. First project guide
For Ollama, start Ollama and pull a model such as llama3.1. Set OLLAMA_BASE_URL=http://127.0.0.1:11434 and OLLAMA_MODELS=llama3.1 in .env. For an API provider, set its key in .env, such as OPENAI_API_KEY or ANTHROPIC_API_KEY. Restart OntoLab and select the provider in the model picker. Environment template · Supported providers
| Layer | What it does |
|---|---|
| Clients | React/Vite web app, Electron desktop app, and AI clients through the local MCP bridge. Docs are bundled at /docs/. |
| HTTP API | Node server with loopback Host and Origin checks. Provider credentials stay on the server. |
| Services | Documents, requirements, induction, alignment, extraction, review, revisions, graph chat, and validation (server/services/). |
| RDF engine | N3.js for Turtle, rdf-validate-shacl for shapes, Oxigraph (WASM) for SPARQL, and built-in OWL RL checks. |
| Models | Codex or Claude Code CLIs, Claude, OpenAI, Gemini, DeepSeek, or Ollama. |
| Storage | Local filesystem or Postgres/Neon. Optional connections to GraphDB, Fuseki, RDF4J, generic SPARQL, or Neo4j. |
| Sidecars | Optional Docling parser and DL reasoner. |
The diagram sources are in docs/diagrams/. Each diagram has a light and a -dark version, and GitHub shows the one that matches your theme. Open the .drawio files in draw.io to edit them, then export a PNG with the same name. See the architecture guide for details.
| I want to… | Start here |
|---|---|
| Start with my own documents | Create your first project |
| Configure models or storage | Local setup · CLI providers · Postgres |
| Run the desktop app or connect an AI client | Desktop setup · Local MCP |
| Review facts or query a graph | Review · Ask · Connect a graph database |
| Inspect how a run was produced | Research records · Evaluation |
| Work on the code | Architecture · API |
Bug reports, documentation improvements, and code contributions are welcome. For a bug, include a minimal example, the steps to reproduce it, and what you expected to happen. Run these checks before opening a pull request.
npm run check
npm run docs:checkSee the contributor guide for project conventions and additional checks.
OntoLab’s original code, including work adapted from IndustriAgents GraphMind and its knowledge-graph chatbot, is MIT licensed. Modified Helix Foundry components retain their Apache-2.0 license and upstream notice. See third-party notices for bundled dependencies.