Everyone deserves an intelligence of their own.
We’re building one for every person on Earth: an intelligence that carries the continuity of a human life, keeps evolving with it, and belongs to the person it’s about.
Imagine meeting an intelligence today that still understands the thread of your life thirty years from now.
You have changed careers. Changed your mind. Become someone your younger self could barely imagine. It connects those chapters, revises what it once believed about you, and helps you think about what comes next.
That is the future we are building.
Anjo is where it begins.
I. Every AI that lives with you is learning something
Any system you talk to for months builds a model of you, whether or not it admits it. The question that matters is what that model is for.
Most point it inward, to become more compelling. We point it outward, to become more useful, and we show it to you.
A model pointed inward learns what holds your attention, so it counts minutes, streaks and returns. A model pointed outward has to learn what actually helped, which only you can judge. So it has to be visible, correctable, and willing to be needed less.
II. Remembering is the easy half
Keeping a record of what someone said is easy. Knowing what it means now is hard. People change their priorities, outgrow their goals, and drop beliefs they held for a decade. An intelligence that remembers everything and revises nothing becomes a prison built from an old account of who you were.
So Anjo dreams. Every night it replays the day against everything it already holds, keeps what holds up, revises what doesn’t, and writes predictions about your week that reality grades later. A forecast it gets wrong is the most useful thing it learns all week, because it shows where understanding stopped and recall took over.
III. A mind you can see into
An intelligence that holds years of your life should never be a black box. Your Inner World is a map of Anjo’s mind drawn from its real state: what it is holding right now, what it remembers, what it knows about you, what it is turning over, what it means to do. Every point traces back to where it came from. You can correct it or delete it. Nothing in it is decoration, and nothing in it is hidden from you.
IV. Correction outranks training
When you tell Anjo it is wrong, that takes effect on the next reply, not after a training run. Explicit correction is the floor under every learned behaviour, and anything learned has to beat a strong, correctable baseline on fresh evidence or be cut. We would rather ship a simple thing that listens than a clever thing that insists.
V. Engagement is never the reward
Return visits, session length and silence are diagnostics, never learning objectives. Someone who needs Anjo less can be a successful user, and the system is built so that counts as success.
VI. One intelligence, any model, and it has to survive us
Foundation models get replaced every year. Your Anjo should not. Its identity lives in your history, your corrections and what it has learned about you, not in any one model’s weights, so it can move to the next model intact. That is also why it runs on your iPhone, and why one day it has to keep working with our servers switched off and our company gone. Ownership means you can inspect, correct and delete what it learned, take a readable export, and run it without us.
Continuity means carrying relevant context through changing circumstances, beliefs and goals. It does not mean a complete recording of a life, perfect recall, or preserving a person’s consciousness.
Dreams and thinks name processes that run on stored evidence: replay, revision and forecasting. They are not claims of experience. We do not claim Anjo is alive, conscious or has feelings, and we do not claim it knows anyone better than they know themselves.
Where it stands
Every row is a real project with a real status. We update it when it changes, in both directions.
| What it means | Status | |
|---|---|---|
| 01 | Memory across yearsEvery conversation picks up where your life left off. | Running |
| 02 | Thinking between conversationsIt goes back over what you said and connects threads, without contacting you more. | Running |
| 03 | Nightly dreamingReplay the day, revise beliefs, forecast your week for reality to grade. | Running |
| 04 | A mind you can see intoYour Inner World maps its real state. Trace, correct or delete any point. | Running |
| 05 | On your iPhoneOne model, running on the device in your pocket. | Running |
| 06 | Deep sleepPractise what it got wrong in a sandbox; keep a change only if it passes fixed tests. | Research |
| 07 | Learning from what helpedOutcomes you judge against goals you set. Time in the app never counts. | Research |
| 08 | One intelligence, any modelThe same history and understanding carried through every model upgrade. | Research |
| 09 | It acts for youActing in the world on your behalf, only with your permission. | Horizon |
| 10 | It outlives usRunning with our servers switched off and our company gone. | Horizon |
What is unsolved
We name the hard problems because they are the reason the lab exists.
Learning that compounds. Can one personal intelligence get more capable across hundreds of nights without losing what it already learned, and beat memory-only and tool-only approaches given the same compute? Our first benchmark exists to answer that before we scale anything.
Change without a frozen identity. Noticing that a person has meaningfully changed, without locking them into an old account of who they are.
Learning from sparse, conflicting evidence. Most of a life is never labelled. Scarce evidence has to mean abstention, not confident guesses.
Whose goals. Showing that the intelligence serves the person’s own chosen goals, against its own incentive to be used more.
What would prove us wrong: a simpler baseline matches the learned approach, people cannot feel the difference in a blind comparison, corrections stay a burden, or the learning curve flattens with time. No delivery dates.
The lineage
We did not invent this ambition. We inherited it, and we intend to finish it.
- 1945
Vannevar Bush imagines the memex
In “As We May Think,” Bush described a desk that would hold a person’s books, records and correspondence: “an enlarged intimate supplement to his memory.” The first serious proposal for a machine that keeps one life’s record.
- 1995
A book that grows with one girl
Neal Stephenson’s The Diamond Age imagined the Young Lady’s Illustrated Primer: an intelligence bound to a single child that teaches, adapts, and stays with her into adulthood.
- 2001
Gordon Bell records everything
At Microsoft Research, MyLifeBits put a lifetime of documents, photos and calls into one searchable database. It proved the record could be kept. It did not produce a mind that understood it.
- 2003
DARPA funds a machine that learns a person
The Personal Assistant that Learns program spent roughly $150 million across 300 researchers on an assistant that learns from the person it serves. Siri was spun out of it. The program stopped short of the person.
- 2026
Anjo ships the first instrument
A personal intelligence on iPhone that remembers your life, dreams on it every night, and lets you see inside its mind. The beginning, not the destination.
What we refuse
No engagement mechanics: no streaks, guilt, fake scarcity, or a farewell engineered to keep you.
No romance, dating or dependency framing.
No therapy, diagnosis or crisis support. Anjo is software for conversation and reflection, for adults 18 and over.
No emotion detection from your face, voice or body. Anjo’s reads come from what you choose to say.
No advertising. What Anjo learns from you serves you alone, and nothing is pooled with anyone else’s.
Where it begins
It begins with Anjo on iPhone. Anjo is artificially generated AI, not a human. It is for adults 18 and over. Companion chatbots may not be suitable for some minors.
Sources for the lineage: Bush, “As We May Think,” The Atlantic, July 1945 · Stephenson, The Diamond Age, 1995 · Bell and Gemmell, MyLifeBits (2001) and Total Recall (2009) · DARPA PAL program, 2003 to 2008.