NVIDIA Auto Ontology is an open Python framework that builds semantic, graph-based reasoning layers over enterprise data. It addresses a critical gap in agentic workflows: while agents can retrieve documents and query databases, they often produce incorrect results due to inconsistent data definitions and lack of context. The library reconciles metadata across systems, provides trusted business meanings for metrics and entities, and validates generated queries against certified definitions. It exposes this context layer to any agent harness via MCP, CLI, and API, enabling deterministic query generation and explainable AI. Key capabilities include discovering enterprise ontologies from metadata, grounding agents in accurate business semantics, validating SQL queries with provenance, and serving context flexibly. The system learns from corrections, improving shared context over time. It complements existing semantic models and supports natural-language querying powered by NVIDIA NIM, with deployment options including Docker Compose and Kubernetes.
Key Features include natural-language to SQL translation, a semantic catalog stored in Postgres, authentication via Better Auth, and a web UI for chat and analytics. Components span Next.js frontend, FastAPI backend, ingestion workers, and optional HashiCorp Vault. It uses NVIDIA NIM endpoints for inference, with configurable model triplets for reasoning, embeddings, and reranking. Licensed under Apache 2.0, it is currently not accepting external contributions.
Source: Hacker News · Summarized by HeadlinesBriefing