NVIDIA open-sources Auto Ontology: an enterprise context layer that aligns metric definitions before an agent answers
Synopsis
NVIDIA released Auto Ontology, an open library that builds a context layer from enterprise metadata and tells AI agents where data lives, what a metric means, whether to use it, how to use it, where it came from, and what is similar. It generates enterprise ontologies from metadata, connects existing ontologies and semantic models, and reconciles conflicting definitions across systems. It then exposes that context to any agent harness through MCP, CLI, and API, returning provenance with every answer. The library is distributed under Apache License 2.0 and is not currently accepting external contributions.
Interpretation
NVIDIA released Auto Ontology, an open library that gives AI agents an enterprise context layer so they know where data lives, what a metric means, whether to use it, and how to use it before retrieving anything. Agents could already retrieve documents and query databases, but reliability gates agentic analytics: a revenue metric answered faster is worth nothing if the number is wrong. The library targets the layer built from enterprise metadata and is positioned to complement the semantic models customers already own rather than replace them.
That context is carried by enterprise ontologies generated from metadata, in which definitions from different systems are reconciled into one consistent meaning. An ontology is a knowledge structure that writes business concepts and their relationships down explicitly; the library also connects existing ontologies and semantic models and builds AI-ready knowledge graphs. Generated queries are checked against certified definitions and data lineage, the record of where data came from and how it was processed, and every answer carries provenance.
The context is exposed to any agent harness through MCP, CLI, and API, and certified corrections and usage can be folded back into the ontology. MCP is an interface convention that lets agents call external tools and data sources, CLI is a command-line entry point, and API is a programmatic entry point. The library is distributed under Apache License 2.0 with a third-party notices file, and the project is currently not accepting external contributions.
Perspective
The library targets enterprise metadata and relational data sources, stores the ontology in the same Postgres instance as the catalog it describes, uses NVIDIA NIM endpoints for inference, and can be deployed on Kubernetes or a local Docker Compose stack. The example environment given is Wide World Importers, Microsoft's sample sales database.
The material gives no quantitative accuracy or error-rate results for the context layer, and does not say how definition reconciliation behaves when conflicts are numerous. The project is not accepting external contributions, so adopters work through the existing interfaces.