NVIDIA Releases Auto Ontology Open-Source Semantic Layer Providing SQL Query Foundation for AI Agents

Techub News reports that NVIDIA has released Auto Ontology, an open-source semantic layer framework under the Apache 2.0 license. The framework builds a context layer from enterprise metadata, enabling AI agents to write SQL queries based on certified business definitions. Its codebase explicitly points out the core issue: if the numbers are wrong, quickly obtaining revenue answers is worthless. Auto Ontology includes five core features: generating ontologies from metadata, providing agents with trusted metric and relationship definitions, validating generated queries based on certified definitions and lineage, offering services via MCP, CLI, and API, and feeding certified corrections back to the ontology. Its tech stack includes a Next.js frontend, FastAPI backend, workers that handle embeddings, and a Postgres database using pgvector. NVIDIA stated in tests that using this framework can reduce ontology creation time by 90%. The framework is compatible with data sources like Databricks, Postgres, Snowflake, or DuckDB and can configure model endpoints via environment variables.