aipopdb

9 Nov 2026  ·  6,605 works · 976 news indexed

News / Agentic AI Foundation adds four projects from Anthropic, OpenAI, Hugging Face and Alibaba

Agentic AI Foundation adds four projects from Anthropic, OpenAI, Hugging Face and Alibaba

AIPopDB Newsroom · via linuxfoundation.org · · 3 min read

Read the original ↗
Image for Agentic AI Foundation adds four projects from Anthropic, OpenAI, Hugging Face and Alibaba

Image: Linux Foundation

The Agentic AI Foundation, a project of the Linux Foundation, announced on 22 October 2026 that it is taking in four more open-source projects: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba. The news came alongside AGNTCon and MCPCon North America, held on 22 and 23 October at the San Jose McEnery Convention Center. With the additions the foundation hosts ten projects, up from three when it started in December 2025.

That original trio was the Model Context Protocol, the goose agent and the AGENTS.md convention. Anthropic donated MCP to the foundation less than a year ago. The other projects already under its roof are A2A, Agent Router and agentgateway. The foundation presents itself as a single neutral home for shared agent standards, with no one company in control of where they go. Its CTO and co-founder, Manik Surtani, framed the growth as going from three projects to ten in a year. The Linux Foundation's own text is a press release, so the neutrality and no-single-owner framing is the foundation's description of itself.

Agent Skills is an open format for packaging instructions, scripts and other resources that an agent loads only when it needs them. Anthropic released it about a year ago. It is licensed under Apache 2.0 and CC BY 4.0. According to a post from the foundation cited by Unite.AI, more than 70 agent platforms support it, there are over a million skills in circulation and installs are close to 280 million. A technical steering committee with maintainers from several organisations is to guide it from here.

Open Responses is a shared schema with streaming events and tooling, modelled on OpenAI's Responses API. The idea is that code written against it can move between model providers without being rewritten. OpenEnv is a containerised framework and hub for reinforcement-learning environments. It exposes both MCP and the Gym protocol, and it was built by Hugging Face together with Microsoft, Meta-PyTorch, Modal, Prime Intellect and others, so that training environments can be shared rather than rebuilt by each lab. The announcement describes it as a hub as well as a framework, meaning environments can be published and reused across projects.

Generic AI illustration SiliconANGLE used as its article image

Generic AI illustration SiliconANGLE used as its article image

Image: SiliconANGLE

OpenSandbox, which came out of Alibaba, is a platform for running agent code in isolation. It offers SDKs in several languages, unified APIs and runtimes for Docker and Kubernetes, and is released under Apache 2.0. Xiaobin Xu, who heads development infrastructure at Alibaba, said the company did not want the project's future defined by a single firm, which is the case the foundation makes for all of its members.

About the Agentic AI Foundation

The Agentic AI Foundation (AAIF) is a Linux Foundation body that gives open-source agentic AI projects a neutral home, so that agents develop transparently and collaboratively.

The Linux Foundation announced it on 9 December 2025, in San Francisco. It started with three projects, each given by a different company: the Model Context Protocol from Anthropic, the goose agent framework from Block, and the AGENTS.md convention from OpenAI. The launch announcement named eight platinum members: Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI.

Anthropic's position, as it gave it, is that donating MCP was not the end of its commitment, and that it intends to keep investing in open tooling for agents. Its lead MCP maintainer, David Soria Parra, credited the foundation with giving the protocol a neutral base. For people building agents, the practical effect is that a skills format, a model-call schema, a training-environment standard and a sandbox runtime now sit under one governance body. The foundation did not name a date for any further additions.

Unite.AI's illustration for its article on AAIF reaching 10 projects

Unite.AI's illustration for its article on AAIF reaching 10 projects

Image: Unite.AI