The Memory Worth Building Is Yours

Knowledgebooks exist to make one thing true: what an AI knows about you belongs to you. Not to the model. Not to the company behind it. To you, on a page you can read, in words you can edit, in a document you can take anywhere. This is the case for why that matters, and what we're building to make it real.

The only question left

Start with what's coming, because it's coming either way. Personal AI is going to know everything about your family. It'll know the kids' ages and the allergy and the mortgage math. It'll know what you asked at midnight and what you were too embarrassed to ask a person. The technology is too useful for this not to happen; being known is the whole point.

The only question left is who holds the knowing.

Today there are two answers, and both are wrong. The first: nobody holds it. The AI forgets you, and you re-explain your life in every chat. The second: a company holds it, in a memory you can't read, can't correct, and can't take with you. A file about your family on someone else's servers, useful to their models before it's useful to you. One road makes AI weak. The other makes you the product.

We're building the third road: your AI knows you, and you hold the knowing.

Knowledge lives at home

There's an old insight from economics that the AI industry keeps relearning. Knowledge is local. It lives with the people who use it, and it dies when you try to centralize it. It's why planned economies lose to markets, and why a chef's recipe can't be written into a database by someone who's never run her kitchen. Some of the smartest people in AI are now making this argument for organizations: a company's knowledge should be cultivated where it lives, not extracted into a standard offering.

We think the argument lands hardest at a smaller address. The most local knowledge in the world isn't inside a company. It's inside a household. Which kid is 9 and which is 12. The vegetarian rule at dinner. What Saturdays need to feel like so nobody melts down by mid-afternoon. No focus group holds it, no scraper can find it, and it changes weekly. The person who runs the household carries all of it, mostly in her head, and re-types it into a chat window more often than anyone should have to.

People already know this. The proof is that they built the fix by hand. The most resourceful AI users we've met keep a note called "About Me" and paste it into every serious chat. Nobody told them to. The need was real and no one had built the product. A Knowledgebook is that note, made real: a living document of what you know, written in plain words, that fills as you chat and only changes when you say so.

Nothing lands without your yes

Whether personal AI respects you comes down to design decisions, made one at a time. Here are ours.

Chats delete themselves. Every conversation with Ask Safely clears on a timer you choose, 8 hours by default, because most of what you say to an AI is scaffolding: the question about the rash, the vent, the budget math at midnight. It helped in the moment and doesn't need to follow you around. When every word is kept forever, you start editing yourself, and an assistant you can't be honest with isn't much of an assistant. When a chat clears, it's gone from your history. Encrypted in transit and at rest, and never used for training.

Keeping is consent. When something in a chat seems worth keeping, Ask Safely proposes it. Nothing lands without your yes. The machine drafts; you decide. That's not a settings toggle. That's the relationship.

Answers show their work. Under every reply sits a receipt showing which Knowledgebook each fact came from. Tap it and you're reading the source. Edit the line and the memory updates. Delete it and it's gone. It's your handwriting, in effect, even when we hold the pen.

None of this is how you'd build AI memory if your goal were to accumulate data. All of it is how you'd build it if your goal were to be trusted with a family.

Moral and practical, and honest about both

We should say plainly why we're doing this, because the answer is both moral and practical, and pretending otherwise would be its own kind of dishonesty.

The moral part is old.

In 1890, Louis Brandeis called privacy "the right to be let alone."

And people have understood ever since that some rooms need a door. We'd put it differently for the AI age: privacy is the permission to be honest. Doctors, lawyers, and clergy keep confidence not to help people hide, but because honesty needs protection to exist. People now bring those same conversations to AI. They deserve the same protection, built in.

The practical part is incentives. You pay us with money, not data. That single decision does more than any privacy policy: we have no reason to keep more of you than you chose, nothing to gain from a chat you forgot about, no advertiser waiting for the profile. And we bound ourselves to it publicly. Exporting your Knowledgebooks is free, forever. The moment owning your knowledge costs money, the trust story collapses, so ownership is never the paywall.

That commitment weakens our own lock-in, on purpose. Your Knowledgebooks can leave with you any time, which means the only way we keep you is by being the best home your knowledge ever had. We think that's a harder business to build than a cage. We also think it's the only one worth building.

Where this goes

Later this year, Knowledgebooks travel. Choose what another assistant can read: your People and Lifestyle, say. It answers like it knows you too. You'll see every recall and its source, and you can stop sharing any time. Your other Knowledgebooks stay home. The chat is the front door. The document underneath is the point: knowledge that works everywhere and belongs to one person.

The road we're on

The future of personal AI hasn't been decided. Being known by a machine doesn't have to cost you anything. It just has to be built that way, one deliberate decision at a time, by people who'd rather be trusted than fed. A book is the oldest technology we have for keeping what you know yours. We built the new one on top of it.