Mattj: I've already thought about most your points.
* Which devices should run a full XMPP stack?
For this I more or less agree with you it does not make sense for every light bulb to have it's own XMPP stack (to may individual devices) so they should (like they are at the moment) connect to some kind of Hub via LoRa/ZigBe/... and than this Hub acts as a proxy to XMPP and or other services with it's own JID. Larger devices like for example a fridge/heater/CCTV/... should have their own XMPP stack and JID since they usually provide a group of options and information to the end user.
* How powerfull has a microcontroller to be to be able to run XMPP?
That's something I'm not really sure about atm. I'm working atm with a ESP32 for my thesis as a prototype. Lets see if I actually need both cores 😉
* Running your own server.
Well. Yes. At the moment you can not hope that everybody has it's own XMPP server so people have to rely on external servers. Those servers could be provided instead by the provider? Or an other approach I thought about for an indealized world is if your router runs one central XMPP server. Routers already run all kind of stuff. Problem here server licence cost/maintenance!
* Should everything run via the Server.
Yes to be able to provide as may features without reinventing stuff using existing XMPP stuff you would have to run everything via the server (no direct connection). Problem you rely on the server (latency!?)
That's something I actually haven't thought about. It's really annoying if you tap a button and then have to wait ~1sec until the state/light changes.
I have to do some testing here and see how bad it actually is!
Wow it got a really big wall of text. Zorry 😉
Hope I didn't forgot any of your points
I'd say that I really was impressed how fast my google home mini activated via voice command a wifi power outlet
(i sold the power outlet later, but for differnt reasons)
anyway, I am curious if the google home mini communicated directly with the outlet, without a message leaving my home network, or not
flow: Yes, this would be interresting to know. It allways feel to me different if I activate something with speach instead of pushing a button. When you tell somebody something you usually are aware that it takes them an instance to do so. On the other hand if you switch a light switch it makes ding and on/off is your light.
COM8, I suspect that the google homes are at least able to perform some speech recognizion locally and then directly talk to the required device/thing in your household and tell it what to do.
Yes, this sound plausible
COM8: sensible answers. I still have my doubts about some parts (where the server will be hosted is a key part), and... I have my doubts about ESP32 being able to manage XMPP, but I'd be very keen to be proven wrong :)