Ontology for AI Agents, Grounded in Your Business.
Connect business concepts and relationships to the metrics and data behind them. With Datus and DOSI, agents can work through explicit business meaning—and your team can inspect the path to an answer.
The business expert should not have to explain it every time.
A retail finance expert knows why booked orders differ from recognized revenue. Making the concepts, relationships, and calculation rules explicit gives agents a path to that knowledge—and gives the team something it can review.
Why is recognized revenue lower than booked sales?
Without ontology
One expert. Repeated explanations.
With ontology + semantic model
Reviewed once. Reused across workflows.
Illustrative workflow. Teams define and review the policy and mappings before agents use them.
Business meaning, made explicit
More context than a column name.
An ontology for AI agents describes the concepts, properties, and relationships in a business domain. It gives an agent explicit vocabulary for interpreting a question. In Datus, that vocabulary connects to semantic models and source mappings, so supported questions can reach data through DOSI.
01
Ontology
Which objects are involved, and how are they related?
Customer → places → Order
02
Semantic model
Which metrics, dimensions, and calculations answer the question?
Revenue · order count · region
03
Mapping
Which source fields and relationships give those concepts meaning?
Explore business concepts and their data mappings.
In this Northstar demo, select a concept to explore its relationships and mapped properties. Switch to lineage to trace source data, or execution to inspect recorded SQL and results.
DatusNorthstar LivingNorthstar Premier demo
11 concepts · 11 relationships
Click a concept to trace its relationships and source fields
Prepared demo: ontology and mappings require team review.
Datus Agent uses the ontology to navigate concepts. DOSI resolves supported paths through a mapped semantic model and plans the query. Teams can inspect the generated SQL and its underlying definitions.
One connected context. From business meaning to execution.
Ontology names the business concepts and relationships. The semantic model defines metrics, dimensions, and source mappings in compatible Apache Ossie definitions. DOSI connects supported ontology paths to those definitions and plans SQL; Datus Agent uses that context in its work.
Existing knowledge
Documents
SQL & ETL
Warehouses
Datus
Datus Studio
A shared workspace for Agent-driven data engineering.
Datus Agent
A complete data Agent for engineering and analysis.
Ontology + semantic model
DOSI Engine
Execute compatible Ossie definitions as native SQL.
Apache Ossie Spec
An open, vendor-neutral semantic modeling specification.
Put context to work
Teams
AI agents
Applications & BI
Apache Ossie (incubating) is an ASF project. Datus and DOSI are independently developed products; the diagram shows how they use compatible definitions, not ASF ownership or endorsement.
Apache Ossie (incubating) provides open definitions. DOSI is Datus’s compatible execution runtime, licensed under Elastic License 2.0; it is not an ASF product.
Evaluate against your model and installed version. Supported constructs and interfaces are evolving.
Build a useful scope first
Start with one decision. Keep people in the loop.
Choose a business question whose answer your team can check. Start with existing analytics definitions, build the concepts and relationships needed to answer it, and expand as your team validates the model.
01
Agree on meaning
Build a shared business vocabulary with domain owners. Define terms, distinguish roles, and identify exceptions before mapping them to data.
02
Map and review
Connect business concepts to existing semantic definitions and source metadata. Treat Agent-assisted work as a draft, with mappings reviewed by the people who own the data.
03
Test and evolve
Maintain shared definitions that multiple agents can reuse. Test reference questions when definitions change, and approve corrections before applying them to later work.
Put context to work
Use Datus. Or bring execution into your own agent.
For teams using Datus
Use Studio as the workspace for data tasks and Datus Agent as the data engineering agent. Configure OntologyAnalyst with a compatible model and mapping to explore business objects and retrieve data through supported paths.
Use DOSI’s documented interfaces to discover semantic definitions and request supported queries from your workflow. Check the ontology reference for the constructs and operations exposed by your installed version.
These approaches can work together. The right starting point depends on the question you need to answer.
Approach
What it contributes
When to use it
Semantic layer
Metrics, dimensions, grain, and calculation rules.
Your question is primarily about a defined KPI.
Ontology
Business concepts, roles, and relationships.
The question depends on which entity or relationship you mean.
Knowledge graph
Connected facts about entities, often organized using an ontology.
You need to navigate a network of facts and their connections.
RAG
Retrieves supporting material for a model to use.
Your answer needs evidence from documents or other retrieved content.
A graph-shaped illustration does not imply a graph database. The Datus approach shown here maps business concepts to semantic models and queries existing data.
Start with the question your agent needs to understand.
Bring a business domain, its definitions, and the data behind it into Datus.
No. A narrow task with an unambiguous metric may need only a semantic model. Ontology becomes useful when the question depends on business entities, relationship roles, or domain vocabulary. Start with an actual ambiguity you need to resolve, and compare the added modeling effort with the value of making that meaning reusable.
Can an LLM build the ontology for us?
An LLM can help draft concepts and relationships from supplied material, but your team still needs to decide what they mean. Assign owners, check mappings, and test representative questions. In Datus, distinguish Agent-assisted semantic modeling from the ontology and mapping artifacts required by OntologyAnalyst; do not assume an unreviewed draft is an executable business contract.
Do we need to move our warehouse into a graph database?
Not for the Datus approach shown here. Ontology mappings connect business concepts to semantic models over your data. DOSI uses supported mappings to plan SQL. A knowledge graph can serve other needs, but adopting business vocabulary and relationship paths does not by itself require moving all warehouse records into a graph store.
Does an ontology eliminate hallucinations?
No. Explicit concepts and supported checks can make some mistakes detectable, but correctness still depends on the model, mappings, data, and the agent’s interpretation. Evaluate with reference questions and inspect the execution evidence. No accuracy percentage or guarantee follows simply from adding an ontology.
Who should maintain the business definitions?
Business owners should approve meanings and exceptions; data teams should own source mappings and executable calculations. Keep changes reviewable and test their effects on existing questions. This is a recommended operating model, not a promise that an agent automatically learns and approves every change.
What should we verify before using DOSI ontology support?
The ontology reference currently labels the feature experimental. Check the supported constructs and installed versions, prepare a compatible semantic model and mapping, and test the queries you need. Do not infer general OWL reasoning, arbitrary business-rule enforcement, or cross-warehouse federation from the ontology capability.
