Agentic AI Foundation 新增四个开源 AI 工具，项目总数达 10 个
Agentic AI Foundation adds four more open-source AI tools - SiliconANGLE
Agentic AI Foundation 新增四个开源项目，分别由 OpenAI、Anthropic、Hugging Face 和阿里巴巴捐赠，使其维护的项目总数从 3 个增至 10 个。OpenAI 捐赠的 Open Responses 提供统一的模型 API 实现模板，让代码在不同 API 之间迁移时无需大改；Anthropic 捐赠的 Agent Skills 为技能提供标准化格式，便于团队共享并降低模型幻觉风险。Hugging Face 的 OpenEnv 和阿里巴巴的 OpenSandbox 则面向模型训练所用的沙箱环境，前者减少切换训练系统的工作量，后者提供管理沙箱的 API 集合。
正文
The Agentic AI Foundation, a consortium that maintains open-source artificial intelligence tools, today took four new projects under its wing.
The technologies were donated by OpenAI Group PBC, Anthropic PBC, Hugging Face Inc. and Alibaba Group Holding Ltd. They bring the total number of projects maintained by the AAIF to 10. That’s up from the three the group managed when it was launched by the Linux Foundation last December.
Developers access cloud-based large language models through application programming interfaces. Each LLM provider uses a different API, which means that software written for one provider’s API can’t be easily ported to another. Switching providers often requires developers to make extensive code changes.
Open Responses, the first of the four projects donated to the AAIF, addresses the challenge. It provides a standardized template for implementing AI model APIs. Developers can move code between APIs that use Open Responses without making major changes.
The technology was donated by OpenAI, which modeled it after the API through which it provides access to its LLMs. That means OpenAI-powered workloads can be easily moved to APIs based on Open Responses and vice versa. Plus, the fact that many developers are already familiar with the company’s APIs should ease adoption.
The second contribution that AAIF announced today came from Anthropic. The company donated Agent Skills, an open-source project it released last December. It provides a standardized format for implementing skills, which are collections of resources that improve an AI agent’s output quality. Those resources can include instructions that explain how to carry out a task, illustrative examples and automation scripts.
Packaging skills in a standardized format makes it easier to share them between development teams. There’s no need to change how the files are structured before connecting them to a workload. Organizing data consistently also makes it easier to process for AI agents, which lowers the risk of hallucinations.
“The AAIF has done important work in giving the Model Context Protocol (MCP) a neutral home since we donated it less than a year ago, and alongside the AAIF, we’re committed to continuing to invest in open source agentic tooling for builders everywhere,” said Anthropic engineer David Soria Parra.
AI labs train their models in sandboxes, technology environments that simulate the infrastructure on which an LLM will be expected to run in production. For example, a company building a vibe coding model might create sandboxes that simulate developer workstations. Sandboxes are the focus of OpenEnv and OpenSandbox, the two other projects that the AAIF announced today. They were donated by Hugging Face and Alibaba, respectively.
The sandboxes used in an AI project are coordinated by a training system that automates repetitive management tasks. OpenEnv reduces the amount of work involved in switching training systems. OpenSandbox, in turn, provides a set of APIs for managing sandboxes. It eases tasks such as connecting agents that run inside sandboxes to external applications.
Image: Unsplash