I think the the key is to have multiple sensing and acting devices working together to create and act on a high-quality model of the interior environment.
My preference would be to self-host that model and the associated controls. I suspect this would require some meaningful compute capability in-house.
I would hope. There's some usefulness for alarm-level conditions to communicate with the outside world...but it is really disappointing how many devices feel the need to phone home about nothing in particular.
True, I'd be happy to self-host the meat of a system like this.
However, I suspect that the problem would benefit from cloud-scale compute. It's a fluid dynamics problem, which is just a lot of numbers to crunch. This might require more horsepower than a little ARM device can reasonably push out.