Skip to main content
Version: v5.3

AI assistance and reasoning

AI assistance​

The Application Ontology Service can use the AI Service to help you build an ontology and prepare data for it. You don't need these features to build or use an ontology.

The AI assistance calls can:

  • Generate a model: Propose classes, relationships, and rules from a natural-language description.
  • Infer an item type schema: Propose a schema from sample JSON data.
  • Review a dataset: Review an import dataset against a model.
  • Prepare a source dataset: Propose an import dataset from files in the File Service.

The calls return proposals. They don't change your ontology. Validate a proposal, then apply it with the classes, relationships, and rules calls. The AI Service output is returned as is.

Each call runs on an AI Service team. If you don't send a team ID, the platform sets up a team for the call and returns its ID so you can reuse it.

The platform checks that the ontology exists before it calls the AI Service. An unknown ontology returns 404, and no AI call is made.

Note: For more information, refer to the Generate a model, Infer an item type schema, Review a dataset, and Prepare a source dataset Rest API sections.

Class MCP tools​

You can expose a class as an MCP tool, so that AI agents can query the class's individuals. For more information on MCP tools, see MCP overview.

A class MCP configuration declares the tool's name and optional query criteria. You declare it in a draft, and the platform creates the tool in the AI Service when you promote the draft. Until then, its status is PENDING. After promotion, it's PROVISIONED, or FAILED if the tool couldn't be created. Any change to the configuration sets it back to PENDING.

The tool lists the individuals in the class's collection, so map the class to a collection before you promote. If a promoted version drops or renames a tool that the live version had, the platform deletes the old tool. Provisioning also fails if a tool with the same name already exists in the AI Service and the platform didn't create it.

To retry a FAILED tool, provision the live declaration again.

Note: For more information, refer to the Declare a class MCP config and Provision a class MCP config Rest API sections.

Reasoner​

The reasoner applies an ontology's classes, rules, and constraints to a set of instances. It infers which classes each instance belongs to, derives relationships from rules such as inverse and transitive properties, and reports rule and constraint violations. For more information, see Rule enforcement.

You send the ontology, rules, constraints, and instances in the request. The reasoner doesn't read or change stored ontologies or individuals, and doesn't check ontology permissions. A result with violations still returns 200.

Note: For more information, refer to the Run the reasoner Rest API section.

OWL preview​

The OWL preview parses and classifies OWL documents that you send in the request, and returns their axioms. Use it to review an existing OWL ontology before you turn its axioms into rules. Nothing is saved. To keep an axiom, create a rule from it. For more information, see Rules.

To see why the preview inferred an axiom, ask for an explanation. It returns the asserted axioms that support the inferred one.

Note: For more information, refer to the Preview OWL documents and Explain a previewed axiom Rest API sections.