Visual ontology design
Drag-and-drop entities, relationships, and class hierarchies in Ontology Studio. Or import FIBO, CDISC, IOF and other OWL/RDFS standards.
Databricks Labs · Source-available
Turn your Databricks tables
Turn your Databricks tables
into a living knowledge graph
OntoBricks designs ontologies, maps them to Unity Catalog tables, materializes or virtualizes triples into a triple store, reasons over the graph, and serves it to your apps and AI agents — the full pipeline in four clicks.
OWL / RDFS
R2RML
Lakebase · Lakehouse · Neo4j
OWL 2 RL · SWRL · SHACL
GraphQL · SPARQL
MCP
01
Design
Ontologies in OWL — visually or LLM-generated
02
Map
Entities to Unity Catalog tables via R2RML
03
Reason
OWL 2 RL, SWRL rules & SHACL checks
04
Serve
GraphQL API & MCP for AI agents
What it does
Semantic web power, Databricks-native
OntoBricks brings ontologies, reasoning, and graph exploration to the Lakehouse — without leaving Unity Catalog or writing a single triple by hand.
AI-assisted mapping
An LLM wizard writes the SQL that binds ontology entities to your tables, column by column — then auto-maps everything else in a single batch.
Materialized or virtual triples
Materialize triples into the backend you choose — Lakebase (Postgres), Lakehouse (Delta) or Neo4j — incremental by default and no manual SQL, or keep them virtual with a live pass-through view over the Lakehouse.
Formal reasoning
Run OWL 2 RL deductive closure, SWRL business rules, transitive/symmetric expansion, and W3C SHACL data-quality validation over the graph.
Typed GraphQL API
A GraphQL schema is generated from your ontology at runtime — traverse nested relationships in a GraphiQL playground, per project.
MCP for AI agents
Expose the knowledge graph to LLM agents through the Model Context Protocol — connect from Cursor, Claude Desktop, or the Databricks Playground.
The pipeline
From metadata to knowledge graph in four clicks
LLM-powered automation handles the heavy lifting — you stay in control at every step.
1
Import metadata
Fetch table and column metadata straight from Unity Catalog. OntoBricks reads your schema so it understands what it's modeling.
2
Generate the ontology
A three-stage wizard detects entities from Data Assets and the Knowledge Store, you review the catalog, then relations and axioms merge in — append-only.
3
Auto-map to tables
Generate R2RML mappings for every entity and relationship, with multi-pass column matching and partial-mapping detection.
4
Synchronize & explore
Materialize triples into the store, then query with GraphQL or SPARQL, or explore the interactive, WebGL-powered knowledge graph.
Inside the app
See it in action
Watch & learn
Videos from the datacorner channel
Walkthroughs, release highlights, and deep dives. View the full playlist ↗
Get started
Up and running in minutes
bash
# Clone the repository
git clone https://github.com/databrickslabs/ontobricks
cd ontobricks
# Install dependencies with uv
uv sync --frozen --extra lakebase
# Configure your Databricks credentials
cp .env.example .env
# edit .env → host, token, SQL warehouse ID
# Start the app → http://localhost:8000
scripts/start.sh
bash
# Authenticate against your workspace
databricks auth login --host https://<workspace>
# Deploy as a native Databricks App
make deploy
# Then bind the sql-warehouse & volume resources
# in Compute → Apps → ontobricks → ResourcesPrerequisites
- Python 3.10+
- A Databricks workspace with Apps enabled
- A SQL Warehouse (classic or serverless for builds; Lakehouse//RT optional for reads)
- A Unity Catalog catalog/schema (Volume still used for Lakehouse graphs and export bundles)
- Databricks Lakebase Autoscaling (registry + Knowledge Store)
Under the hood
Built on open standards & the Lakehouse
OntoBricks stitches together semantic-web standards, graph databases, and formal reasoning on top of the Databricks Platform. Ontologies compile to R2RML, mappings become Spark SQL, and triples land in Lakebase, Delta or Neo4j — all governed by Unity Catalog.
- Multi-engine backend — Lakebase (Postgres), Lakehouse (Delta) or Neo4j, per domain. Split Build vs Query warehouses for Lakehouse//RT reads.
- Version lifecycle — domains move through DRAFT → IN-REVIEW → PUBLISHED with audit trails, optional GitHub mirror.
- Managed-synced mode — delegate bulk movement to Lakeflow snapshot pipelines.
Community
Built by contributors
OntoBricks is source-available under the Databricks License and welcomes contributions. See everyone on GitHub ↗
Latest release
What's new in OntoBricks 0.9.0
Released October 2026 — Lakehouse//RT reads, a staged Generate wizard, and a faster Explorer. Read the full release notes ↗
Lakehouse//RT graph queries
Point Knowledge Graph reads at a serverless Lakehouse//RT warehouse over the Statement Execution API. Builds stay on a classic or serverless warehouse that still accepts DDL.
Three-stage Ontology Generate
Detect entities, review and lock the catalog, then complete relations, attributes, and axioms with checkpoints. The merge appends to the existing ontology instead of replacing it.
Knowledge Store
Parsed documents live in Lakebase Postgres — not a Unity Catalog Volume. Generate and Mapping share the same corpus; no READ VOLUME grant is required.
GitHub domain mirror
Each instance can push domain configuration (OWL, R2RML, SHACL, layouts, rules) to one GitHub repository. Lakebase stays the source of truth; Load from GitHub restores domains through OBX import.
SPARQL playground & faster Explorer
Run SPARQL against the graph and inspect the generated SQL. Graph status no longer blocks first paint; adjacency indexes, paged BFS, and a canvas stopwatch cut preview and expansion time.
Studio, jobs, rules & entity DQ
Designer is now Studio. Bind Databricks Jobs as class actions, run SWRL on one entity, and check a node against inherited SHACL shapes — from Explorer or MCP.
Ready to build your knowledge graph?
Source-available, Databricks-native, and yours to explore.