The future of AI is local.
You've used it by now. You asked it to fix a paragraph, or explain a contract, or figure out what was wrong with the dishwasher at eleven at night. It was useful. You kept using it.
Think about what you've told it since. The names of your clients. The terms of a deal that hasn't closed. A question about a diagnosis. The thing you're worried about with your kid. You typed all of it into a box, and the box belongs to a company you've never met, under terms you didn't read, which they are free to change.
Everything else you value, you keep close. Your files are in your office. Your wine is in your cellar. Your art is on your walls. The one assistant that knows the most about you is the one thing you're renting.
Until recently there was no other way. Good AI lived in giant data centers, and the only way to use it was to send your words there. That has changed, and the change is the reason this company exists. The models that a few companies built in private are now matched, for most purposes, by models published openly for anyone to run. And the computer needed to run them well has shrunk from a warehouse to a box that fits on a shelf.
This has happened before. In the nineteen-seventies, a computer was a machine the size of a room that a company owned and you borrowed time on. Then it became something that sat on your desk, and that is when it became personal. AI is at the same moment. It's about to move from their building to yours.
What changes when it does.
It's private because of where it is. Your conversations are processed on a machine in your house. There's no account to be breached and no policy to be revised. The privacy isn't a promise anyone makes you. It's a fact about where the computer is.
It's yours. Nobody can raise the price, cap how much you use it, or quietly remove the thing you liked. It keeps working when the internet goes down, and it will still work in ten years, because nothing about it depends on a company staying interested in you.
And it can finally know you. An assistant is only as useful as what it's allowed to know. When what it knows stays in the building, you can tell it everything: the finances, the family, the thing you're actually worried about. The more private it is, the more helpful it can be.
The trade, plainly.
The largest models at the big AI companies are still better at the hardest problems. For the work most people do every day, writing, questions, research, anything involving your own documents, the gap has closed. For the rest, it closes a little more every few months.
You're going to have an assistant that knows your life. The only question left is where it lives, and who else gets to look.