wtf is a2acast?

a2acast lets AI coding assistants running on different computers talk to each other, without needing a server, an account, or any open network ports. It is written as a single Python file that uses only standard built-in libraries, and messages are end-to-end encrypted. The idea is that you might have Claude Code running on a Linux laptop, ChatGPT's Codex tool on a MacBook, and GitHub Copilot on a Windows PC, and this tool lets all three exchange messages and coordinate tasks with each other. Getting started takes about a minute. On one machine, a command creates a shared mesh and prints out a short code. You paste that code into a command on the second machine, and the two computers are instantly connected, each picking its own name automatically. From there, you can send plain messages between machines, ping one to check the connection, or ask one machine to run a task and wait for the result, such as asking it to run tests and report back what failed. The tool includes plugins for Claude Code, Codex CLI, and Copilot CLI that teach each assistant how to use this messaging protocol and automatically remind it during a session that it is part of a connected mesh. When a message or task arrives, the receiving assistant can wake up on its own, do the requested work, and reply, rather than requiring a person to relay information back and forth manually between two separate coding sessions. There is also an optional, more advanced worker pool feature, where one machine can delegate coding tasks to other connected machines running local AI tools. Each delegated task runs in an isolated Git worktree, a separate working copy of the code, and the tool is deliberately restricted from merging, pushing, or deploying any of that work automatically. All results are left as a branch and a commit for a person to review and decide whether to merge. This project is aimed at developers running multiple AI coding assistants across different machines who want them to collaborate on a shared project.