NVIDIA/auto-ontology: A Python framework for building semantic, graph-based reasoning layers

Detailed Features

Auto Ontology acts as a bridge between your raw databases and AI agents. It automatically ingests database schemas, profiles columns, and infers table relationships to build a semantic graph of your enterprise data. Developers can use it to translate natural language questions into accurate SQL queries powered by NVIDIA NIM, ensuring that every answer comes with verifiable provenance and execution steps.

System Architecture

The project is designed with a modern microservices architecture. It includes a Next.js frontend for the Web UI, a FastAPI Python backend for chat and catalog management, an ingestion worker for vector embeddings, and a PostgreSQL database equipped with pgvector to store both application data and the semantic ontology.

Installation and Usage

Getting started is straightforward using Docker Compose. You simply need to clone the repository, configure your environment variables including your NVIDIA API key in a .env file, and run Docker Compose to spin up the local stack. Once running, you can connect your databases and start asking questions through the built-in chat interface or integrate it with other tools via the MCP server.