Agentic AI Foundation adds four more open-source AI tools
The Agentic AI Foundation, a group that looks after open-source artificial intelligence tools, announced today that it has taken on four new projects.
OpenAI Group PBC, Anthropic PBC, Hugging Face Inc. and Alibaba Group Holding Ltd. donated the tools. The additions bring the total number of projects managed by the AAIF to 10. When the group was launched by the Linux Foundation last December, it managed just three projects.
Developers use application programming interfaces (APIs) to connect their software to large language models (LLMs) that run in the cloud. Each LLM provider has its own API. This means software built for one provider may not work easily with another provider's API. Switching providers can require developers to make many changes to their code.
Open Responses, the first of the four projects donated to the AAIF, aims to solve this problem. It provides a common format for building APIs that connect to AI models. Developers can move code between APIs that follow the Open Responses format without making major changes.
OpenAI donated the technology and based it on the API it uses to give people access to its LLMs. As a result, software that uses OpenAI can be moved more easily to APIs that follow the Open Responses format, and software using those APIs can be moved back to OpenAI. Many developers already know how to use OpenAI's APIs, which could also make the new format easier to adopt.
The second project announced by the AAIF today came from Anthropic. The company donated Agent Skills, an open-source project it released last December. Agent Skills provides a common format for creating skills, which are groups of resources that help AI agents produce better results. These resources can include instructions for completing a task, examples that show how to do it and scripts that automate steps.
Using a common format for skills makes it easier for development teams to share them. Teams do not need to change the way files are organized before connecting them to a software system. Keeping information organized in a consistent way also makes it easier for AI agents to process and reduces the risk of hallucinations, when AI produces incorrect or made-up information.
“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 research teams train their models in sandboxes. These are test environments that imitate the computer systems where an LLM is expected to run when it is put into real use. For example, a company building an AI model that writes code from natural-language instructions might create sandboxes that imitate developers' computers. The two other projects announced by the AAIF today, OpenEnv and OpenSandbox, focus on these test environments. Hugging Face donated OpenEnv, while Alibaba donated OpenSandbox.
A training system coordinates the sandboxes used in an AI project and automates repeated management tasks. OpenEnv reduces the work needed to switch from one training system to another. OpenSandbox provides a set of APIs for managing sandboxes. It also makes tasks easier, such as connecting AI agents running inside sandboxes to outside applications.