Genius in a Bottle

It comes at two depths, and most people start with the first.

Most people's knowledge lives in four places at once — an app that will eventually be discontinued, a notes folder last opened a year ago, a chat history, and their own head. The cost stays invisible until you catch yourself working the same thing out for the third time.

I build the other kind: plain text on hardware you own, structured so the connections between things are visible, versioned so every earlier state stays recoverable. An AI can read and work inside it while all of it stays on your machine. It is the system I run my own life on.

The setup happens with you. A knowledge system somebody else built, that you understand from the outside only, becomes one more app you stop opening — so the handover is the work itself.

The app it lives in puts your notes on the left and your AI on the right. It runs on Linux, Windows and Mac, and is in its first tests.

The second depth is the whole stack on a machine in your own building.

Most AI arrives as a rented service: your data leaves the building, the price moves, and the thing you depend on stays under someone else's hand. There is another way to do it, and it runs on a box in your own house.

I build that box, install it, and teach you to run it. Local models for the high-volume work, a frontier model reached deliberately for the hard thinking, and a clear line between them so your spend goes on judgement while the typing stays cheap.

What it looks like in practice depends on what you do all day. A household wants the lights, the cameras and the calendar to cooperate. A business wants the repetitive work carried and the records to hold. A homestead wants water, power and stock legible from one place. Same architecture underneath, different surface.

It comes in two sizes, and most people start with the first.

Most people's knowledge sits in four places at once: an app that will shut down one day, a notes folder last opened a year ago, old chat messages, and their own memory. You notice the cost the day you catch yourself working the same thing out for the third time.

I build the other kind: plain text files on your own computer, laid out so you can see how things connect, backed up so every earlier version stays there. An AI can read and work in it while everything stays on your machine. It is the system I use for my own life.

We set it up together. A system somebody else built, that you only half understand, becomes one more app you stop opening — so teaching you to run it is the job.

The app it lives in shows your notes on the left and your AI on the right. It runs on Linux, Windows and Mac, and is in its first tests.

The bigger size is the whole setup on a computer in your own building.

Most AI you can buy is rented. Your information goes to a company's computers, the price changes when they decide, and the thing you rely on stays in the hands of people you will only ever meet as a support ticket. There is another way, and it runs on a computer that sits in your own building.

I build that computer, set it up, and show you how to run it. The everyday work is handled on your machine. The hard thinking gets sent out to a bigger service only when it is needed, so the money goes on the hard parts.

What it looks like depends on what you do all day. A house wants the lights, the cameras and the calendar working together. A business wants the repetitive jobs done and the records kept straight. A farm wants water, power and stock visible in one place. The same machine underneath, a different face on top.