NVIDIA Auto Ontology: How It Works

NVIDIA Auto Ontology: Help AI Agents Understand Your Data Before Writing SQL

What Is NVIDIA Auto Ontology?​

AI agents can generate SQL queries, but understanding a database is more than knowing table and column names. They also need to understand what business terms mean, how tables are connected, and which definitions should be trusted.
That's the problem NVIDIA Auto Ontology is designed to solve.
Auto Ontology is an open-source Python framework that builds a semantic layer and knowledge graph from enterprise data metadata. It helps AI agents understand business concepts, discover relationships between tables, generate SQL queries, and validate their answers before returning results.

---

🧠 How Does Auto Ontology Work?​

Instead of giving an AI agent a database schema and expecting it to figure everything out, Auto Ontology creates an organized knowledge layer containing three main elements:

• 📊 Business Terms: Definitions of concepts such as customers, orders, revenue, and products, mapped to their actual database columns.
• 🔗 Relationships: Connections between tables and entities, including the joins needed to retrieve related data.
• 🧮 SQL Attributes: Reviewed SQL expressions that help agents calculate metrics consistently.

This information is organized into a semantic graph that AI agents can use when answering questions.

---

⚙️ How Does It Work in Practice?​

Imagine asking an AI agent:
How much revenue did a specific product generate in California during June 2025?
Auto Ontology helps the agent follow these steps:

1. 🔍 Identify the business terms in the question.
2. 🔗 Find the relevant tables, columns, and relationships.
3. 🧮 Generate SQL using the available definitions.
4. ✅ Validate the query and check its filters.
5. 📊 Execute the query against the connected database.
6. 📋 Return the result along with the SQL and execution steps.

If the available data cannot answer the question, the system can report that instead of making up a result.

---

🔌 How Can You Connect AI Agents to It?​

Auto Ontology supports several integration options:

• MCP (Model Context Protocol).
• REST API.
• Command-Line Interface (CLI).
• A Web UI for exploring data, managing definitions, and asking questions.

It can also work with MCP-compatible clients such as Cursor and Claude Desktop.

---

🗄️ Supported Databases​

The project supports connections to several popular database platforms:

• PostgreSQL
• MySQL
• Snowflake
• Databricks
• ClickHouse
• Trino
• Apache Kyuubi
• HeavyDB
• DuckDB
• SQLite

---

🖥️ Web Interface and Human Review​

Auto Ontology includes a Web UI where teams can:

• Explore the data catalog and semantic graph.
• Review and edit business definitions.
• Manage SQL attributes and relationships.
• Test questions before relying on the results.
• Certify definitions so agents can use trusted business logic.

This human review step is important because automatically generated definitions may need corrections before they can be trusted.

---

⚠️ Is Auto Ontology a Replacement for RAG?​

Not exactly. Auto Ontology focuses on structured data, database relationships, and business definitions. Traditional RAG is commonly used to retrieve relevant information from documents and other unstructured content.
They solve different problems and can complement each other in the same AI system.

---

🚀 Why Is This Project Interesting?​

The important idea is that connecting an LLM to a database is not enough to guarantee reliable answers.

AI agents also need:

• Clear business definitions.
• Accurate relationships between data sources.
• Reusable and reviewed SQL logic.
• Query validation.
• Data provenance so results can be traced back to their sources.

Auto Ontology brings these elements together, making AI-powered analytics more explainable and easier to verify.

---

🔓 Open Source and Getting Started​

NVIDIA Auto Ontology is available under the Apache 2.0 license.
You can explore the source code, installation instructions, documentation, and integration examples on GitHub.
NVIDIA Auto Ontology - GitHub Repository
Official Documentation and Quickstart

---

💡 Final Thought​

The future of AI-powered databases isn't just about making LLMs better at writing SQL. It's about giving AI agents the context they need to understand what the data actually means before they query it.
That's exactly the problem NVIDIA Auto Ontology is trying to solve.