Skip to main content
Preview Ontology is one shared model of your business that agents read before they query your data. Data and platform teams define customers, orders, and their relationships so agents share the same meaning. You describe the model once, publish it as a fixed release, and map it to your warehouse tables. Agents then ask for just the part of the model a task needs. The running example on these pages is a retail company. Its model has four object types (Customer, Order, Product, Store) and two links: a Customer places an Order, and an Order contains Products.

​
How it works

Logical definitions follow five steps. The context tools read published releases.

​
What you can do

Model your business

Add types and links, then publish your first release.

Map the model to tables

Tell agents which tables, columns and joins hold each type.

Give agents bounded context

Agents fetch only the definitions a question needs.

Govern policy records

Keep policies and controls with effective dates.

Review drafts from agents

Agents propose definitions; a person approves the save.

Automate with the API

Import, validate and publish releases from scripts.

​
Get started

Build your first ontology

Publish a retail model and inspect its agent context in 15 minutes.

​
Key concepts

  • Object type: a kind of business thing, such as Customer, with its properties.
  • Link type: a named relationship between two types, such as Customer places Order.
  • Release: a published, unchangeable snapshot of the model that agents read.
  • Bindings: a published mapping from one release to tables, columns and joins.
  • Knowledge record: a policy or control stored with the dates it is in force.

​
Studio or API

Ontology Studio and the API share the same release rules, so you can start in one and continue in the other.

​
Common tasks

​
Availability and limits

  • Preview, opt-in app. An administrator installs the Ontology app; it is not part of default onboarding. Bindings need the Ontology Binding app too. See Install and manage apps.
  • No data copy. Ontology stores definitions and a small set of governed policy records. Customer, order and other warehouse rows stay in your warehouse.
  • No SQL runs. Neither Ontology nor bindings connect to a source, generate SQL or run a query. Agents query with their own connection.
  • Actions are described, not run. An action type records what an operation needs and changes. Nothing executes it.
  • Release size. One release holds up to 256 definitions.

​
Reference

​
Troubleshooting

See Ontology errors for import, validation, publication and context errors.