It can get data from the internet, I think the point is that it "lives" on-device and hence can be interacted with offline, unlike Siri/Alexa/Cortana which live entirely on the company's servers.
Siri/Alexa/Cortana are assistants, while the part you are talking about is speech recognition. Sure, it’s great that this is done offline, but I don’t think it’s fair to compare it to more services because at that point you do need to get data from the internet just like any other service.
I would say your comment was the unfair one. Sure, it cannot tell the weather without internet. But it can still do plenty of more important home automation functions, namely automation. No one owns one of these home assistants for it to tell the weather. You get it for controlling other devices with your voice and this can do that without access to thr world wide web.
Voice Control in iOS can do some of this. It’s older than HomeKit, though, so it can’t do home automation-it’s limited to running commands on-device.
I believe the intention is providing a voice recognition system you can use to interface with other things, like APIs. Products from Big Data (in my usage like Big Oil, Big Tobacco, etc.) offer convenience at the cost of privacy. For many people, this is a minor trade-off. For others, alternatives with perhaps less convenience.
I think it’s misleading to sell this as a “privacy-focused AI” (specifically, similar to Siri or Google Assistant) when it’s really an offline speech recognition tool. This keeps the “privacy” part accurate, and makes it clear that intents are still being sent out from the device (as opposed to audio recordings).