Turn documents into navigable knowledge graphs.
OntoSphere is an open-source tool that extracts structured ontologies from unstructured documents using LLMs. Upload a PDF, and it identifies classes, properties, and relationships -- then assembles them into an OWL/RDF knowledge graph you can explore, visually edit, and export. Built for anyone working with agentic systems, semantic layers, GraphRAG pipelines, or domain-specific ontologies.
- Auto-generate class URIs from labels -- typing a label auto-fills the URI with a slugified version; fully editable for manual override
- Relationship type picker -- after drag-to-connect, choose from SUBCLASS_OF, HAS_PROPERTY, RELATED_TO, EQUIVALENT_TO, DISJOINT_WITH, or enter a custom type
- Undo/redo for graph editing -- toolbar buttons and Ctrl+Z / Ctrl+Shift+Z keyboard shortcuts; supports add class, delete class, and add relationship
- Import existing ontology -- new option in the Create Ontology wizard to import .ttl, .owl, .rdf, .jsonld files directly using rdflib, as an alternative to generating from documents
Upload a PDF and let LLMs (Azure OpenAI, OpenAI, or Anthropic) extract entities, properties, and relationships with full provenance tracking. Processing runs asynchronously via Celery workers with real-time progress updates over WebSocket.
Toggle edit mode to modify your ontology directly on the graph canvas. Right-click nodes to edit, delete, or start a new relationship. Right-click empty space to add a new class. Drag between nodes to connect them.
Already have an ontology? Import it directly. The Create Ontology wizard offers an "Import existing ontology" mode that accepts Turtle (.ttl), OWL/RDF-XML (.owl, .rdf), and JSON-LD (.jsonld) files. The backend parses them with rdflib, extracting classes, properties, and relationships into the graph -- no LLM processing needed.
Browse and manage your ontologies from the dashboard. Each ontology tracks its processing status, document count, and version history.
Search for classes by name and inspect node properties, relationships, and provenance in the side panel.
Export to JSON, Turtle (TTL), JSON-LD, and RDF/XML for use in downstream tools and triple stores.
Every generation and edit creates a version snapshot. Compare versions and roll back when needed.
Validate generated ontologies against SHACL shape constraints. Violations are surfaced directly in the editor.
| Dependency | Version | Notes |
|---|---|---|
| Docker | 20.10+ | Required |
| Docker Compose | 2.0+ | Required |
| LLM API key | -- | Azure OpenAI, OpenAI, or Anthropic |
# Clone the repository
git clone https://github.com/boricles/ontosphere.git
cd ontosphere
# Create your environment file
cp .env.example .env
# Edit .env and set your LLM API key:
# ONTOSPHERE_LLM_API_KEY=your-actual-key
# ONTOSPHERE_LLM_API_BASE=https://YOUR-RESOURCE.openai.azure.com
# Start all services
docker compose up --build -d
# Open the application
# Frontend: http://localhost:5173
# API docs: http://localhost:8000/docsDatabase tables are created automatically on first startup.
docker compose down # stop, keep data
docker compose down -v # stop, wipe data| Layer | Technology |
|---|---|
| Frontend | React 18, TypeScript, Vite, Cytoscape.js, Tailwind CSS |
| Backend | Python 3.11, FastAPI, SQLAlchemy 2.0 (async), Pydantic v2 |
| Database | PostgreSQL 16 + Apache AGE (graph queries) |
| Queue | Redis + Celery |
| LLM | Azure OpenAI / OpenAI / Anthropic (pluggable) |
| Export | RDFLib (Turtle, JSON-LD, RDF/XML) |
| Validation | pyshacl (SHACL shapes) |
+---------------------+
| Frontend |
| React + Vite |
| Port 5173 |
+----------+----------+
|
| REST / WebSocket
v
+----------+----------+
| Backend |
| FastAPI + Uvicorn |
| Port 8000 |
+---+------------+----+
| |
+--------+--+ +----+--------+
| PostgreSQL | | Redis |
| + Apache | | (broker + |
| AGE | | pub/sub) |
| Port 5432 | | Port 6379 |
+------------+ +------+------+
|
+------+------+
| Celery |
| Worker |
+-------------+
All configuration is via environment variables. Copy .env.example to .env and adjust.
| Variable | Default | Description |
|---|---|---|
ONTOSPHERE_LLM_PROVIDER |
azure |
openai, azure, or anthropic |
ONTOSPHERE_LLM_API_BASE |
-- | Base URL for the LLM API |
ONTOSPHERE_LLM_API_KEY |
-- | API key for the LLM provider |
ONTOSPHERE_LLM_MODEL |
gpt-4o |
Model / deployment name |
DATABASE_URL |
(see .env) | Async SQLAlchemy connection string |
REDIS_URL |
redis://redis:6379/0 |
Redis for Celery and pub/sub |
CORS_ORIGINS |
localhost:* |
Allowed CORS origins |
See .env.example for the full list.
Interactive API docs are available when the backend is running:
- Swagger UI: http://localhost:8000/docs
- ReDoc: http://localhost:8000/redoc
cd backend
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000Requires a local PostgreSQL with Apache AGE extension and Redis.
cd frontend
npm install
npm run devcd backend && pytest -v- Document upload and text extraction (PDF)
- LLM-based entity/property extraction with provenance
- Automatic ontology assembly
- Apache AGE graph storage
- Interactive graph visualization (Cytoscape.js)
- Multi-format export (JSON, Turtle, JSON-LD, RDF/XML)
- SHACL validation
- Ontology versioning
- Real-time progress via WebSocket
- Docker Compose orchestration
- Robust WebSocket reconnect with dormant mode
- Visual graph editing (drag-to-connect, context menus)
- Auto-generate class URIs from labels
- Relationship type picker in drag-to-connect flow
- Undo/redo for graph editing operations
- Import existing ontology files (.ttl, .owl, .rdf, .jsonld)
- SHACL violation visualization in graph editor
- Ontology CI / agent-behavior dry-run (git-style diff, schema-registry compatibility checks, dry-run replay)
- Collaborative editing
- Authentication and authorization (OAuth 2.0 / OIDC)
Contributions are welcome. Please:
- Fork the repository
- Create a feature branch:
git checkout -b feature/my-feature - Make your changes and add tests
- Ensure tests pass:
cd backend && pytest -v - Open a Pull Request against
main
This project is licensed under the Apache License 2.0. See LICENSE for the full text.
Built by Boris Villazon-Terrazas, PhD (@boricles).