First thing is be aware that any system where you have local access only will not perform as well as Google. In the real world you will only get Goggle performance from google or alexa.
That being said HA Vpe is good and combined with AI is very good, there are still limitations but it works and is mostly good enough or better.
There are so many options from DIY, through various chinese boxes to HA VPE, best bet is to spend time searching this forum.
At the end of the day if you want 100% reliable and something that is hassle free and responds to everything then alexa is the only option to replace Google right now.
For me it’s DIY all the way and I have replaced Google with DIY so it can be done but be realistic with your expectations and the amount of setup needed.
I’m aware of the required work necessary. I’m currently running three proxmox hosts with several lxc, vm and Docker containers (too give just 1 example of my home lab), so I’m up to the task. I just want to know what the general direction is.
So far, I haven’t found any proper solution despite this tiny Atom dev device. I bought that, but found it extremely limited. So if you can point me in some direction, I’ll be quite happy!
It will come down to adding LLM’s the hardware can use to understand your requests. Check out the Cookbook under number 7 and read through those guides to get an idea of what is involved.
Then regarding hardware what Arh is mentioning is the ‘Voice PE’ hardware, there are multiple other options around, namely Seeedstudios has their own hardware. There will be other options if you check the forum.