Skip to content

AI in the house that stays in the house.

A voice assistant that ships every request to a company's servers is a reasonable trade for a kitchen timer and a poor one for a system that knows when your house is empty, which doors are unlocked, and who is home. The value of AI in a home comes from context — and context about a household is exactly what should not leave it.

So we run the model locally. A language layer on hardware in your house lets you ask for things in plain words, compose routines by describing them, and get useful summaries — without a record of your family accumulating somewhere you cannot see.

What the work covers

A model running on your hardware
Specified for the house, sitting alongside the automation server rather than in someone else's data centre.
Natural-language control
Ask for what you want in ordinary words instead of remembering the exact name of a device.
Household memory, held locally
Preferences and routines the system remembers, stored on your equipment.
Routines you describe rather than build
"Get the house ready for guests on Friday" composed into a real automation you can inspect and edit.
Summaries instead of notification spam
What happened while you were out, in a sentence, rather than forty alerts.
Optional cloud, always your call
A larger cloud model can be added for tasks that benefit — deliberately, and never as the default.

Questions people ask

Does the AI send my home data to the cloud?

Not by default, and that is the entire point of the design. The language model runs on hardware in your house, so requests, device state and household context stay on your equipment. A cloud model can be added for specific tasks that genuinely benefit from one, but that is a deliberate choice you make, not the default behaviour.

What hardware does a local AI model need?

Less than most people expect. A small dedicated machine with a capable GPU handles natural-language control and summarisation for a household comfortably. We size it against what you actually want it to do, and it sits alongside the automation server rather than replacing it.

How is this different from Alexa or Google Assistant?

Those assistants process requests on the manufacturer's servers and are limited to what that manufacturer supports. A local model runs in your home, works with the full state of your own system, and does not build a record of your household somewhere else. It can also act on context — who is home, what is open, what usually happens at this hour.

Tell us what is already in your house.

Bring us the apps that will not talk to each other, the cameras you do not trust, and the routines that almost work. We will design the rest around how you actually live.