1.0.0, and ask the release a question the way an agent would. You don’t need any SQL or warehouse access.
Time: about 15 minutes.
What you’ll build
Aretail ontology with four object types and two links:
Before you begin
- An administrator has installed the Ontology app. See Install and manage apps.
- You can create in at least one workspace.
- Open Ontology Studio: in Brain, select Apps, then select Ontology. See Find your installed apps.
1
Create the ontology
- On All ontologies, choose New ontology.
- Select the Workspace and enter
Retailas the Name. Studio fills in the API nameretail. - Optionally open More options and set an IRI namespace such as
https://example.com/retail#. Studio then suggests an IRI for each definition. Leave the Versioning policy on Backward compatible. - Choose Create ontology.
2
Add the four object types
- Under Model, open Definitions and choose New definition.
- Select the Object type kind, enter
Customeras the Name, and keep Object kind on Entity · something that exists over time. - Choose Create draft.
- Repeat for
ProductandStoreas entities, and forOrderwith Event · something that happened.
3
Add properties
- Open Customer and select the Properties tab.
- Choose Add property. Enter the API name
customer_id, choose the Data type Text, and turn on Required and Identifier. Choose Add property. - Add
statusas Text. Optionally add Allowed values such asactiveandchurned. - Choose Save draft, or press ⌘S or Ctrl+S.
- Repeat for the other types, using the properties in What you’ll build. Use Date and time for
ordered_at.
4
Connect the types with links
- In Definitions, choose New definition and select the Link kind.
-
Enter
placesas the Name, choose Customer under From and Order under To, and set the Reverse name toplaced_by. Choose Create draft. A new link starts as One to many, which is whatplacesneeds. The link page reads “A Customer has many Orders · an Order has exactly one Customer”. -
Create
containsfrom Order to Product with the reverse namecontained_in. -
On the
containspage, set Relationship to Many to many and choose Save draft.
5
Validate and review
Under Release, open Publish. Studio checks every draft together as soon as the page opens.
- Issues shows errors and warnings. A link or type without an IRI gets a warning, because RDF, OWL and SPARQL tools can’t address it. Warnings don’t block publishing.
- Fix any errors. Each issue links to the definition it names.
6
Publish release 1.0.0
- Under Version, keep the suggested Version number
1.0.0. - Choose Publish 1.0.0.
rel_. Choose Copy release ID and keep it for the last step. The release can’t change: to fix something later, publish 1.0.1 or 1.1.0.7
See the context an agent gets
Under What’s next, choose Ask this release. Studio opens Agent context on the Ask tab with release
1.0.0 selected.- Keep the scope on Logical model.
- Enter the Question
customers and their ordersand choose Ask.
places link can appear as linked to a match. The header shows a token estimate such as ≈ 900 of 12,000 tokens and fits, which means nothing was cut.An agent gets the same result by calling the extension__ontology__get_context MCP tool. The ontology ID is the part of the Studio address after #/o/. Open a session with mcp__initialise first, then replace the example IDs and session_id with your values:Verify
- Releases lists
1.0.0with its date, author, definition count and checksum. - The toolbar status shows Live 1.0.0.
- The Data bindings, Knowledge records and Agent context pages now appear in the navigation. They only appear once a release exists.
What you accomplished
You built a six-definition model, validated it as one graph, and published an unchangeable release. You then saw the bounded slice an agent receives for a question, with the release ID and token estimate that prove which model it read.Next steps
- Bindings: map Customer, Order and Product to your warehouse tables.
- Use Ontology with agents: connect an MCP client and pick the right tool.
- Ontology concepts: interfaces, shared properties, actions and terminology.
- Publish an ontology through the API: automate the same steps.