offdata ai — agentic AI for data modelers and data engineers

Ontology builder

Turn your data model into a business ontology

OffDataAI derives concepts, relations and metrics from your tables, lets your team review every AI suggestion, and publishes the result as a standard OWL ontology that people, tools and AI agents can all use.

Build your ontology freeBook a demo
sales.ttlowl · turtle
@prefix : <https://app.offdataai.com/ontology/acme/sales#> .
@prefix owl:  <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .

:Customer a owl:Class ;
    skos:prefLabel "Customer" ;
    skos:definition "A person or company that has
        placed at least one order." .

:Order a owl:Class ;
    skos:prefLabel "Order" .

:placedBy a owl:ObjectProperty,
        owl:FunctionalProperty ;
    rdfs:domain :Order ;
    rdfs:range  :Customer .

Most warehouses are full of names that only their authors understand. An ontology is the layer that says what those tables mean: which business concepts exist, how they relate, and how metrics such as revenue are calculated. Without it, every analyst, dashboard and AI agent has to rediscover that meaning on its own, and they rarely agree.

OffDataAI builds the ontology from the model you already have, whether it came from the AI interview, a template, the Model Builder or a live database. Concepts stay bound to their tables, so when the model changes the ontology can be re-derived instead of rewritten by hand. When it is ready, publish it to give every term a resolvable IRI, export it as OWL, or let AI agents query it through the built-in MCP server.

What the ontology builder does

  • Derived from your model, not from scratch

    Concepts, relations and metrics come straight from the data model, with bridge tables folded into relations and every concept bound to its source tables.

  • Human review for every AI suggestion

    AI-proposed concepts, relations and renames wait in a review queue. “Improve names” cleans up labels like fct_ord_ln in one pass, for you to approve.

  • Published, resolvable IRIs

    Publish and each term gets a link that works: a readable page for people, Turtle, RDF/XML or JSON-LD for machines, via content negotiation.

  • Standard exports

    Download OWL as Turtle, JSON-LD or RDF/XML, or the raw JSON. Open it in Protégé, validate it with SHACL, or load it into a knowledge graph.

  • Several sources, one vocabulary

    Combine a warehouse model with a live operational database. The same Customer in both becomes one concept with two bindings.

  • Ready for AI agents

    Agents connected over MCP can search concepts, get the query context for a question and find join paths before they write any SQL.

Frequently asked questions

What is a data ontology, and how is it different from a schema?
A schema describes storage: tables, columns and keys such as fct_ord_ln.amt. An ontology describes meaning: the business concepts (Customer, Order Line), how they relate (an Order contains Products), and the metrics built on them (Revenue). OffDataAI keeps both in sync because the ontology is derived from, and bound to, the same data model.
How does OffDataAI build the ontology?
First a deterministic pass derives concepts, relations and metrics from the structure of your model, with no AI and no credits. It strips modeling prefixes like dim_ and fct_, turns bridge tables into many-to-many relations instead of fake concepts, and keeps a binding from every concept to its tables. An optional AI pass then improves names and writes business definitions.
Do AI suggestions change the ontology automatically?
No. Every AI-proposed concept, relation or rename goes into a review queue. A person accepts, edits or rejects it, so the published ontology only contains what your team has approved.
Can one ontology cover several databases?
Yes. You can add several sources, such as a project's data model and a live database read by the Database Modeler, to one ontology. The same business concept in two systems becomes one concept with two bindings.
What does publishing do?
Publishing gives every term a resolvable IRI. A person who opens the link sees a readable page with the term's definition; an RDF tool asking for Turtle, RDF/XML or JSON-LD gets the machine-readable form. Unpublished ontologies stay private and are only reachable with an API key.
Which export formats are supported?
OWL in Turtle, JSON-LD and RDF/XML, plus the raw JSON document. The RDF formats open in tools such as Protégé and can be validated with SHACL.
Is the ontology builder included in the free plan?
Yes. Creating an ontology from your model, publishing it and exporting it are available on every plan, including Free. The AI enrichment pass (better names and definitions) uses credits and is included from the Pro plan.

Explore OffDataAI

Your data warehouse is one conversation away.

Describe your domain, or open one of 170+ production-grade templates. ERDs, DDL, and a complete dbt project — generated in under a minute.

Get startedTalk to usBrowse templates