The voice was never the hard part. What made every talking assistant feel small is that the conversation ended when you stopped speaking, so the only jobs it could take were the ones that fit in one sentence.
Timers, weather, the next song. Each request lands on an assistant that has never met you, which is why after five years of owning one you still only ask it three things.
You can say “chase the thing we talked about on Tuesday” and be understood, because Tuesday is in there, along with who was on it and what was decided. The voice is just the fastest door into a mind that already knows the situation.
There is no public address and nothing is exposed to the internet. Your phone reaches it over your own private network, which is why it can be this open without being open to anybody else.
The half hour that was dead is the half hour where you hand over the day. You describe what needs doing, and it is doing it before you have parked.
Say the thing rather than type the thing. Talking is three times faster than typing and you were going to have to explain it anyway.
Hold to talk, or just start talking. The machine doing the work is at home; what is in your hand is a way to reach it.
Start a thing by voice in the kitchen and pick it up in a browser an hour later. Same conversation, same memory, no repeating yourself.
There is no prompt engineering and there is no course to take. If you have to learn a way of phrasing things, the tool has handed you its problem.
You are allowed to change your mind halfway through the sentence, to say it out of order, to forget the name of the thing and describe it instead. It has the context to work out what you meant, and if it genuinely cannot, it asks rather than guessing. What Maison is.
Yes, from your phone, your desk or the car. Hold to talk or just start talking. The speech goes to your own machine and the voice that answers is generated there too, so the whole exchange happens on hardware you own.
No. Turning your speech into words, searching your memory, and generating the voice that answers all run on your own chip, and all three keep working with the network off. Only the hard reasoning leaves, through your own Claude account.
Those hold one request at a time, which is why after years of owning one you still ask it three things. Maison is a conversation that has been running for months, so you can say chase the thing we talked about on Tuesday and be understood.
That is one of the places it earns its keep. The half hour that was dead becomes the half hour where you hand over the day, and the work is underway before you have parked.
No. There is no prompt engineering and no course to take. You can change your mind halfway through the sentence, say it out of order, or forget the name of the thing and describe it, and if it genuinely cannot work out what you meant it asks rather than guessing.
Yes. It is the same Maison and the same memory whether you use your voice, your phone, a browser or your desk. Start it in the kitchen and pick it up an hour later without repeating yourself.