Rank #02 · Weekly · AI
auto-ontology: NVIDIA's Graph Ontology Layer for Deterministic Agent Analytics
Give agents a certified business meaning layer — terms, joins, and SQL attributes — so analytic answers come with SQL, steps, and a clear "unanswerable" path when coverage is missing.
Canonical digest: /daily/auto-ontology (also on Weekly board).
What it does
Auto Ontology is NVIDIA's open library for a semantic, graph-based reasoning layer over structured data. The README frames the reliability problem directly: retrieving documents or querying a database faster is useless if the metric is wrong.
Core workflow from the docs:
- Generate. Ingestion profiles schemas, samples values, extracts terms, infers relationships, and suggests SQL attributes. Ontology and embeddings land in Postgres + pgvector.
- Certify. People refine terms and attributes, then certify names, descriptions, and SQL.
- Answer. The agent retrieves candidates, writes SQL, validates/repairs, and returns the answer with SQL and steps.
check_answerablecan refuse when coverage is missing.
Surfaces: MCP server (FastMCP), agent skills (AGENTS.md), REST chat completions API. Stack: Next.js + FastAPI + Postgres/pgvector; inference defaults to NVIDIA NIM; connectors include Postgres, MySQL, Databricks, Snowflake, ClickHouse, DuckDB, and more. Docker Compose and Helm documented. Project notes it is not currently accepting external contributions.
Who it's for
Teams building agentic analytics / text-to-SQL who need certified measures and provenance; platform engineers wiring ontology tools into MCP clients. Not a zero-config chat-over-CSV toy.
Stars / language / license
10,407 stars · Python · Apache-2.0
Data verified: 2026-11-09 09:19 Asia/Shanghai (GitHub API get_repository + prior README structure)
Trending note: +9,327 stars this week (weekly Trending; selection captured 09:10 Asia/Shanghai)
Category tags
AI · ontology · text-to-SQL · RAG · MCP · NVIDIA · enterprise-data · explainable-AI
How it differs from similar projects
Compared with plain RAG. Retrieves certified terms/SQL attributes, not just text chunks.
Compared with dbt / LookML / Cube alone. Those define BI metrics; Auto Ontology adds an agent harness (chat, MCP, answerability, provenance).
Compared with generic LangChain text-to-SQL demos. Here SQL is grounded in a compiled, certifiable ontology with repair loops. TidyHow has not reproduced vendor eval numbers.