Auto Ontology: NVIDIA’s Glossary of Business Terms That Checks AI Agents’ Database Queries

Auto Ontology is an NVIDIA library that drafts a glossary of a company’s business terms from its database, lets people approve each entry, and checks AI agents’ database queries against it. NVIDIA reports it cut the time to create such a glossary by 90% in its own testing.
artificial-intelligence
data-science
Author

Kabui, Charles

Published

2026-11-11

Keywords

auto-ontology, semantic-layer, text-to-sql, ai-agents, nvidia

Download as PDF

Auto Ontology is an NVIDIA library for data teams that let AI agents query company databases. An ontology works like a glossary: it defines business terms such as “revenue”, names the database columns that hold each one, and shows how tables connect. The library drafts it from the database’s tables and columns. People test the draft with real questions and approve each entry. An agent then writes SQL (the language of databases) from the matching terms, checks it against the database and repairs it when a check fails. On a Microsoft sample sales database, one run produced 31 terms. Asked for one product’s June 2015 revenue in Pennsylvania, the agent joined six tables and returned $588.80, with its SQL.

NVIDIA’s Adel El Hallak told The New Stack that people know a “tier” and a “plan” mean the same thing at a mobile carrier, but language models “mess those up.” Without a glossary, an agent guesses which columns count, or a data team writes every definition by hand, for example in dbt’s Semantic Layer. Auto Ontology drafts the definitions for people to approve, which NVIDIA says cut creation time by 90% in its own testing. Asked for return on ad spend in a database with no advertising data, the agent asks what to do instead of inventing a number. Agents reach it through MCP, a standard plug-in format for AI agents, a command line or a web API. It reads 10 database types, and each answer typically takes tens of seconds.

Read More: Apache Ossie: One Semantic Model for BI Tools and AI Agents, a shared file format for business definitions that Auto Ontology’s README lists for import and export

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Auto {Ontology:} {NVIDIA’s} {Glossary} of {Business} {Terms}
    {That} {Checks} {AI} {Agents’} {Database} {Queries}},
  date = {2026-11-11},
  url = {https://toknow.ai/posts/nvidia-auto-ontology-business-terms-ai-agents-sql/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Auto Ontology: NVIDIA’s Glossary of Business Terms That Checks AI Agents’ Database Queries.” https://toknow.ai/posts/nvidia-auto-ontology-business-terms-ai-agents-sql/.