I've been experimenting with pushing local AI fully into the browser via Web Assembly and WebGPU, and finally have a semblance of a working platform here! It's still a bit of a PoC but hell, it works.
You can create assistants and specify:
- Wake word
- Language model
- Voice
I published a little post on my site with some technical details as well if anyone is interested: https://shaneduffy.io/blog/i-built-a-voice-assistant-that-ru...