Why I built an AI that stays on my computer

There is a strange moment before typing something personal into an AI chat.

You write the first sentence. Then you remember that the conversation is going somewhere else: through an account, across the internet, onto infrastructure you do not control. You soften a name. You remove the sensitive detail. Sometimes you close the tab.

The assistant may still be useful, but the honest version of the question never gets asked.

DT Life began with that contradiction. The subjects for which an assistant could be most helpful—private work, a difficult relationship, a medical letter, an unfinished idea, a career decision—are often the subjects people are least comfortable sending to a remote service.

I did not want to solve that with another privacy promise. I wanted the architecture to make the promise testable.

“Private” should describe a design, not a policy

Most software privacy is a relationship of trust. A company receives your information and promises to handle it carefully. The policy may be sincere and the security may be excellent, but your words still left your computer. You still need the company, its account system and its servers to keep their side of the arrangement.

DT Life takes a different route. Its language model runs on your own Windows PC. Conversations, memories, diary entries and other personal material live in an encrypted local database. The encryption key is protected by your Windows account. There is no DT Life account to create, no conversation server and no telemetry quietly describing how you use the app.

That changes the most important question from “Do I trust this company?” to “Can I verify what this app does?”

After the model has been downloaded, block DT Life in your firewall or disconnect the network cable. Ordinary conversation, memory, document reading, translation, dictation and speech continue on the machine. The proof is not a paragraph on this website. It is what does—or does not—break when the internet disappears.

A companion needs continuity

Running locally was only half of the idea. A private chatbot that forgets everything when the window closes is still just a chatbot.

DT Life is meant to become more useful through continuity. It can remember facts you approve, the people who matter, the way you prefer to write and commitments you actually make. It can keep a diary from real material in your day. It can ask how a planned event went, find an answer in documents you gave it and help draft a reply with the relevant context already present.

That memory is not a hidden profile. The app shows what it has stored. You can review suggested memories, correct them, delete individual items, export the data or remove it through scoped controls. A digital twin should not know something about you that you are forbidden to see.

Ownership also has an inconvenient consequence: there is no company copy waiting to rescue you if your disk dies. DT Life therefore includes a passphrase-encrypted recovery bundle that can carry the database and its key to another Windows installation. You have to make and protect that bundle. Local ownership means the safety copy is your responsibility too.

What local AI costs

This design is not magic, and it does not win every comparison.

A small model running on a personal computer will not match the raw ability of the largest cloud models. Its speed depends on your hardware; a capable GPU feels very different from an older CPU. Some optional features genuinely need a network. Web search has to fetch public pages. Calendar feeds have to be retrieved. The model and optional voice or document models must be downloaded before they can work offline.

The important boundary is that those connections are explicit. Web search is off by default. The reasoning still happens locally. If a feature needs the outside world, the app should say so instead of stretching the word “offline” until it becomes meaningless.

Local models can also be wrong. They can misunderstand a screen, write a weak reply or remember a suggestion that needs correction. DT Life keeps the user at the final decision: review the draft, approve the memory, send the message—or do none of those things.

Who this is for

DT Life is for people who want continuity from AI but cannot accept the usual price of continuity.

That might be a lawyer working with confidential documents, a job seeker who does not want their applications uploaded, a freelancer whose work and personal life share one machine, or simply someone who wants a place to think without creating another permanent online account.

It is also for the skeptical person. You should not have to believe the maker of a privacy product. Download it, use the full trial on the computer where it will actually run, inspect what it stores and pull the cable out.

If the app only makes sense while you trust every sentence on this page, then I have built the wrong thing.

Try DT Life on your own computer →