AI Briefingauto-ontology: A data semantics layer that reduces agent errors
NVIDIA/auto-ontology
About the project
A Python framework that addresses the reliability issues that arise when LLM agents query enterprise data. It collects database metadata to automatically generate an ontology consisting of business terms, relationships between columns and tables, and validated SQL expressions. Unlike the conventional approach of inserting schemas directly into prompts, it is designed so that agents reference only meaningful context through the ontology.
A key feature is the ability to pre-validate whether a question can be answered by the data. If an answer is not possible, it does not blindly execute SQL but instead makes a clear request to the user. When an answer is generated, it returns the corresponding SQL and reasoning steps together, enabling the source of the result to be traced. It also provides a workflow for reviewing and certifying ontology definitions through a web UI.
It exposes context to various agent harnesses via MCP, agent skills, and REST APIs. It also provides a companion repository that measures the accuracy difference compared to schema-based reasoning on benchmarks such as BIRD and BEAVER. It is released under the Apache-2.0 license and can be easily deployed with Postgres using Docker Compose.
NVIDIA/auto-ontology
Auto Ontology is a Python framework for building semantic, graph-based reasoning layers over data. It enables deterministic query generation, better context retrieval, and more explainable AI for LLM and agent workflows.
Python
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.