rockeverm3m/airgent-play2

Airgent Play 2 gives your AI agents a voice. Instead of only sending you a text message when a background task finishes, an agent can speak the result out loud through any AirPlay 2 or DLNA speaker in your home, like a HomePod, a soundbar, or studio monitors. The idea is simple: a script or agent finishes a job, such as a download, a data processing run, or a model training session, and calls a command with a short message. That message is converted to speech using a local text to speech engine called CosyVoice 3.0, then streamed to a chosen speaker using the AirPlay protocol. Nothing is sent over the internet, so it works fully offline. If CosyVoice is not installed, the tool falls back to the built in macOS speech command. You can use it in several ways. Any agent or script that can run shell commands can simply call the tool with a message. Python based agents or scripts can import it as a module and call functions directly. It can also read messages piped in from another program's output, so you can chain it at the end of a build or test command. A webhook option that lets other programs trigger speech over HTTP is planned but not yet built. Setup requires a Mac, ideally with Apple Silicon, Python 3.12 or newer, the ffmpeg tool, and at least one AirPlay 2 speaker on the same network. Installation is a single pip command, or you can clone the repository and install it from source. You can set a default speaker so every call goes there automatically, or target a specific speaker by name each time. The project is released under the MIT license and is still early, with only one star and a small roadmap that includes Docker support for headless servers, multi room broadcasting, and Windows support through DLNA.