- Published on
Nvidia – open-source auto-ontology
- Authors
- Name
- DeepHorus
NVIDIA has released Auto Ontology, an open-source library designed to give AI agents a reliable context layer over enterprise data. The Apache-2.0 project builds ontologies from existing metadata and reconciles definitions across systems, aiming to prevent fast but incorrect agentic analytics. It complements semantic models customers already own rather than replacing them.
The library addresses a central reliability problem: agents can retrieve documents and query databases, but a revenue metric is useless if the number is wrong. Auto Ontology tells agents where data lives, what it means, whether to use it, how to use it, where it came from, and what is similar. It exposes that context to any agent harness through MCP, a CLI and an API.
Developers can use it to discover enterprise ontologies, connect existing semantic models and build AI-ready knowledge graphs. It grounds agents in trusted business meaning for metrics, entities and relationships, and reconciles inconsistent definitions. It also validates generated queries against certified definitions and lineage, returning provenance with every answer. A learning loop folds corrections and usage back into the ontology.
The project supports natural-language querying of structured data by translating questions into SQL and executing them against connected databases, powered by NVIDIA NIM. Its semantic catalog sits over relational sources in Postgres. Authentication and role-based access control use Better Auth, and a web UI covers chat, analysis, catalog, analytics and settings.
Deployment options include a Helm chart for Kubernetes and Docker Compose for local stacks. Defaults include the nvidia/nemotron-3-nano-30b-a3b language model, nvidia/nemotron-3-embed-1b embeddings and nvidia/llama-nemotron-rerank-vl-1b-v2 reranking. Users need an NVIDIA API key and connection details for Databricks, Postgres, Snowflake or DuckDB.
Auto Ontology is not accepting external contributions. It is distributed under Apache License 2.0. Its release signals NVIDIA’s push to standardize a trusted agent context layer.
