NVIDIA open-sources Auto Ontology to fix text-to-SQL's biggest problem
Roger_M_Taylor · x · 2026-11-09
NVIDIA released Auto Ontology (10.7k stars on GitHub), attacking the core failure mode of text-to-SQL: agents write valid SQL but return wrong numbers because they guess at what columns mean — and different systems in the same company may define 'revenue' differently.
The framework auto-drafts a semantic ontology layer from database metadata, certified by humans, containing:
- terms: business concepts (customer, order, revenue) with descriptions and synonyms mapped to exact columns
- SQL attributes: reviewed expressions such as margin calculations
Exposed via MCP, CLI, and API to any agent harness, it enables deterministic query generation, better context retrieval, and explainability. Includes natural-language querying powered by NVIDIA NIM, a Postgres-backed semantic catalog, and a web UI.
More from coding & agent
- Agent researcher asks: how valuable is training beyond 256k context on Tinker? — cHHillee · 2026-11-09
- Capwise adds CLI + Doubao voice for auto screen-recording demos, open-sourcing soon — vista8 · 2026-11-09
- Multi-agent AI sales pipeline: 6 agents from lead discovery to supervisor review — atullchaurasia · 2026-11-09
- antirez: devs refusing AI coding help are visibly falling behind — antirez · 2026-11-09
- Coding agent pauses to ask its human: 'I need you to create one Hugging Face key' — Sauers_ · 2026-11-09
- Real-world coding agent benchmark: best model fully solves only 26.4% of tasks — emmanuelvivier · 2026-11-09