Key takeaways

  • Agents get trusted business meaning for metrics and entities, so a revenue question is not left to guesswork.
  • Generated queries are checked against certified definitions, and every answer returns its provenance.
  • Each answer comes back with its SQL and the steps taken, so a person can verify it against the source.
  • It works with the semantic models you already own and serves any agent through MCP, CLI and API.
  • The catches: no outside contributions, internal services need a private network, and one option uploads tables.

What it is

NVIDIA Auto Ontology is an open library that builds a context layer from enterprise metadata and serves it to agents through MCP, CLI and API. The code is on GitHub at NVIDIA/auto-ontology under the Apache licence.

The problem it solves

Analysts and engineers who let an agent query company databases know the risk: the project's own framing is that a revenue metric answered faster is worth nothing if the number is wrong.

Auto Ontology tries to close that gap by giving agents trusted business meaning for metrics, entities and relationships, and by reconciling inconsistent definitions across systems.

What you get

  • Shared meaning: agents work from defined terms, rather than guessing what a column holds.
  • Checked queries: generated SQL is validated against certified definitions and lineage, with provenance returned.
  • Answers you can audit: each one comes back with its SQL and the steps that produced it.
  • Human sign-off: people certify names, descriptions and SQL attributes before they are relied on.
  • Honest gaps: when the data cannot support a question, the agent says so instead of inventing an answer.
  • Reach: it connects to ten database engines, and an MCP server gives agents read-only tools over the ontology.

The project says it complements the semantic models customers already own, rather than replacing them.

The catches

  • Community: the project is currently not accepting external contributions, so it is a vendor-maintained tool you can use but not shape through pull requests.
  • Security: the ingestion service has no authentication of its own, so the internal services must sit on a private network.
  • Privacy: the optional KumoRFM prediction route is a hosted service, and the agent uploads the connected tables to it.
  • Dependencies: inference runs on NVIDIA NIM endpoints, hosted or self-hosted, so you need a model endpoint and key.

What we did not test

We did not run Auto Ontology, connect a database or measure how accurate its answers are. Every claim here comes from a literal quote in the project's README or the GitHub API, and the feature descriptions are the project's own.

Verdict

Editors' view: this is aimed at teams building agents that query structured business data and who need answers they can audit. If you only need ad hoc queries on one small database, it is probably more machinery than you need. Judge accuracy on your own data before trusting it.

Repository
NVIDIA/auto-ontology
Python · Apache-2.0
View on GitHub ↗

Sources

Need help adopting this, or have a software project in mind?

Tell us what you are building. The initial consultation is free. Email [email protected] or follow @zolnaltd.

Start a conversation →