NVIDIA 发布了一款名为 Auto Ontology 的开源库,旨在为 AI 代理构建语义上下文层 1。该工具通过 MCP、CLI 和 API 接口,帮助开发者从企业元数据生成知识图谱,以解决自然语言查询结构化数据时的准确性与定义冲突问题 1。目前该项目以 Apache License 2.0 分发,暂不接受外部贡献者提交代码 1。
在功能特性方面,Auto Ontology 支持从元数据生成企业本体,调和不同系统间的定义,并通过 MCP/CLI/API 暴露上下文 1。技术栈依赖上,该库使用 NVIDIA NIM 进行推理,默认模型包括 nemotron-3-nano-30b-a3b 等;数据库存储于 Postgres,认证使用 Better Auth 1。部署方式支持 Kubernetes (Helm chart) 或本地 Docker Compose 1。此外,其安全机制规定 API Token 仅存储 SHA-256 哈希值,权限继承自拥有者的角色 1。
NVIDIA has introduced Auto Ontology, an open-source library designed to construct semantic context layers for artificial intelligence agents 1. The tool enables developers to generate knowledge graphs from enterprise metadata through Model Context Protocol (MCP), command-line interface (CLI), and API endpoints 1. By facilitating the creation of enterprise ontologies, Auto Ontology aims to reconcile conflicting definitions across different systems and improve the accuracy of natural language queries against structured data 1.
The project is distributed under the Apache License 2.0 but currently does not accept external code contributions 1. Technically, the system relies on NVIDIA NIM for inference, utilizing default models such as nemotron-3-nano-30b-a3b, while employing Postgres for database storage and Better Auth for authentication mechanisms 1. For deployment, users can choose between Kubernetes via Helm charts or local Docker Compose configurations 1. Security protocols are implemented by storing API tokens solely as SHA-256 hashes, with permissions inherited directly from the owner's role 1.
评论
还没有评论,欢迎留下第一条。