@jonathan859 Nope, that is basically what happens in the end. You need a part that processes speech to text, interprets it, acts on it and generates either text output that then again is output via TTS, or voice directly. Technically all of that can happen locally, but you can off-load it into cloud services like OpenAI too. I'd like to run as much as possible locally, or on hardware controlled by me. If push comes to shove, i'll be getting a VGPU server I guess.

@jonathan859 You can integrate a model with toolcall, but I think the risk of hallucination can be pretty costly. For example, if you asks to turn off all the lights, it might shut off all the smart plugs including router. If you ask who's at the door, it might open the door, etc. Also you can only run tiny tiny llms on Pi, and it will be pretty useless at toolcalls.

@Jage Thinking about the same, but renting a VGPU server would be way more expensive than buying even 1 graphics card now, at least in the long run. That is why i'm asking if people got experiences already. I don't need to interpret live video, I don't think images either. Mostly just text to control entities and such, at least for now. But having a bit of GPU power in reserve won't be bad either.

Ya good point I could just build a desktop, and that would give me another excuse to do it. and for you with text that's way easier. Sorry I haven't gone further down the road yet.