> When you enter home - based on presence, time of day, ambient illuminance, barometric pressure, temperature what do you want to happen? Usually this involves a chain of devices, working together inside a blackbox running on an OS that no one single person can understand.

A fair bit of this can already be implemented with a workflow that looks like:

Sensors -> zigbee or zwave -> home assistant

Or Internet based data -> home assistant (weather, external data, the humidity sensor at your local airport might be more accurate than the one on your deck, etc).

And then open source extensions/tools built into home assistant to cause things to happen, send external signals, run scripts, call APIs, talk to a local LLM, etc.

It's very much not a black box of opaque closed source software, rather the opposite, and it has a very active and technically knowledgeable enthusiast development community behind it.

If we posit for a minute that the data acquisition into locally-run system problem is not impossible to solve, then all kinds of things can be built atop that to serve your own unique home preferences.

> Having the logic run local

At the smallest scale people are doing this on raspberry pi, though I have seen people have much more satisfactory results on something with somewhat more system resources like an Intel N150 CPU based mini PC with 12GB RAM.

Going a bit up a bit in budget from there, people are running home assistant and frigate together on the same tiny system that is a KVM hypervisor (again all open source, no black boxes involved), so that Frigate can have a GPU passed through to it for image object detection/video analysis pipelines.

"Sensors -> zigbee or zwave -> home assistant"

Home assistant gets updated quite frequently and so does the OS that runs beneath it. My point is that the chain you are describing here can be (is) one single device, with no outside dependencies that runs the logic on a room level. Quieter computing is if it works without the constant upkeep.