TL;DR: We're donating the FastA2A repository to Datalayer, effective today.
Context
FastA2A started as a lightweight implementation of the A2A protocol: a way for AI agents, tools, and runtimes to communicate through structured, streaming interactions. Marcelo Trylesinski built and maintained it, and the project attracted real community interest. It works. People are using it.
But we've been honest with ourselves about where our focus is now. Pydantic AI, Pydantic Logfire, and the broader validation and AI engineering stack are where we're investing. FastA2A is a good project that deserves more attention than we can give it right now.
Datalayer is the new home for it. They've already been contributing, including recent work on streaming support, and agent-to-agent infrastructure is genuinely central to what they're building. Their roadmap around generative UIs, Jupyter-native AI workflows, and collaborative agent systems maps directly to where FastA2A needs to go.
What changes and what stays the same
Everything stays the same for existing users: MIT license, open source, community contributions welcome. Datalayer will take over maintenance and drive new development, with our support during the transition.
A big thank you to Marcelo for creating the library and seeing it through to this point. And to the Datalayer team, who we're glad to hand this to.
The repo now has a new home. And we're glad it's in good hands.
More details on Datalayer's blogpost.
