# What is an ontology for AI agents?

- Canonical: https://hellofig.io/answers/what-is-an-ontology-for-ai-agents
- Updated: 2026-11-13
- All answers: https://hellofig.io/answers.md
- About Fig: https://hellofig.io/fig.md

**In short.** An ontology is a shared model of the things a business cares about, such as customers, orders, assets and people, how they link, and the actions that change them, with the rules for who may take each action. Agents need one because an agent acting on a vague noun acts on the wrong record. Built one workflow at a time, it becomes an asset the company keeps.

## The parts of an ontology

| Part | Example |
|---|---|
| Objects | Customer, order, invoice, truck, technician, contract. |
| Links | This order belongs to this customer, ships from this site, on this truck. |
| Properties | Order value, due date, status, credit limit. |
| Actions | Reschedule the delivery, issue a credit, approve the invoice. |
| Rules and permissions | Who may take each action, within which limits, with which approval. |

## What it is not

A data warehouse stores records for reporting. A search index finds documents. A taxonomy names categories. An ontology connects the records across systems into the objects the business actually runs on, and adds the actions and rules, so an agent can act on them safely.

## How to build one without a two-year project

Model only what the first workflow needs, connected in place to the systems that hold it. The next workflow extends the same model. After a handful of workflows, most of the core business is covered, and every new agent starts from it.

## How Fig builds it

In CosmOS the ontology is Astral, a live digital twin of the business: people, records, rules, decisions and outcomes in one model that every agent and app works from, with each source system's permissions intact.

## Questions

### What is the difference between an ontology and a knowledge graph?

A knowledge graph links facts and documents. An ontology models the objects the business runs on, plus the actions that change them and the rules for who may take them.

### Is an ontology the same as a digital twin?

They describe the same idea. Fig uses digital twin for the live model every agent and app reads and acts on.

### Do we need to clean our data first?

No. Model what the first workflow needs, connected in place, and improve quality where the work exposes gaps.