---
title: "Ontology Agent"
description: "Generate and maintain the primitives that model your domain."
icon: "wand-magic-sparkles"
---

> **For AI agents:** the complete documentation index is at [llms.txt](/llms.txt). Append `.md` to any page URL for its markdown version.

<Note>The Ontology Agent is in **beta**.</Note>

The **Ontology Agent** generates and maintains the instances of ontology primitives that model
your domain. Generation and ongoing reconciliation are handled by a single agent through a
single review surface; there is no separate tool for the initial build versus later updates.

## Inputs

The agent accepts uploaded documents and connects to external data sources such as fileshares
and databases. Together with a prompt and the surrounding context, these inputs define the
material the agent draws on when proposing primitives.

## Generation

From its inputs, the agent instantiates three kinds of primitive:

| Primitive | Models | Example (OEM hardware manufacturer) |
|-----------|--------|-------------------------------------|
| [Thing Type Definitions](/platform/data-model#thing-type-definition) | The entities in your domain | Product models, assemblies, and parts |
| [Metric Definitions](/platform/data-model#metric-definition) | Measured or derived quantities | Tolerances, mean time between failures, test thresholds |
| [Templates](/platform/templates) | Reusable structures for instantiating these consistently | A mapping from a supplier's data sheet to a part type |

Every proposed primitive is shown in a **preview** before it enters the ontology. You can
inspect what the agent generated, edit it, and publish it. Nothing is written to the ontology
without review.

## Drift detection

Once primitives are published and their sources remain connected, the agent monitors those
sources for change. When a source diverges from what the ontology currently records — a
specification is updated, a field is added, a part is superseded — the agent detects the drift
and proposes updates to the specific affected primitive instances.

These proposals appear in the **same preview, edit, and publish surface** used for initial
generation, so reviewing an update is identical to reviewing a new primitive.

Because generation and drift detection share one agent and one review surface, the ontology is
treated as a **living model** that the agent continuously proposes changes to, rather than a
one-time output that falls out of date.

## Related docs

- [Agentic overview](/agentic/overview) explains how the agent fits the build–maintain–surface lifecycle.
- [Knowledge Base](/agentic/knowledge-base) renders the published ontology as a portable wiki.
- [Data model](/platform/data-model) defines the primitives the agent generates.