The idea is to use a voice interface to actually perform "skills", like Alexa. Check out the prompt and you can get an idea of how they've set it up. So the LLM generates code that can then be executed to perform actions.

https://github.com/OpenInterpreter/01/blob/main/software/sou...

There don't seem to be many skills defined in this repo though, but I didn't look too hard. This software is the server, and then they have a small remote device that can talk to the server and it appears to use ngrok for that.

[replying to comment here]

I totallly agree that the term "OS" is used for a lot of projects that are not actually what I would call an OS. My own employer is guilty of this. :) Yes, this seems to be just like UFO, and Rabbit, and other similar projects.