Mission

An AI that remembers your life.

superemotion is an attempt at something narrow and specific: an AI that remembers your life well enough to be useful to you in it.

  • Not a friend
  • Not a therapist
  • Not a character
  1. 01 · Knows

    Your sister’s name.

    Self

    you

    sibling

    Person

    sister

  2. 02 · Remembers

    The interview was Thursday.

    job interview·Thursday
  3. 03 · Asks

    About it on Friday.

    Friday

    How did Thursday’s interview go?

The evidence

The need is real. The apps keep failing it.

Every figure comes from a primary study, named and linked. None of them are about us.

i · The need

Loneliness is common, and help is slow to reach.

1 in 6

people worldwide are affected by loneliness.[1]

World Health Organization · 2025

871,000

deaths a year are linked to loneliness and social isolation — about 100 every hour.[1]

World Health Organization · 2025

48 days

is the average wait for behavioural-health care in the United States.[2]

National Council for Mental Wellbeing · 2025

ii · The category

People try AI apps, then leave them faster.

21.1 %

annual subscriber retention for AI apps, against 30.7% for apps without AI.[3]

RevenueCat · State of Subscription Apps 2026

iii · Cause one

It forgets you, or it changes on you.

Our answer · MemoryIt remembers what you tell it

16.73 %

of all discussion in the largest AI-companion community is about coping with model updates and loss — about one post in six.[4]

MIT Media Lab · 2025

36.7 %

of top posts there name ChatGPT as the platform: a general assistant with no companion features. Apps built for companionship barely register.[4]

MIT Media Lab · 2025 · share of posts

25.6 %

of Character.AI’s 210,840 Google Play reviews give one or two stars. Memory and context problems are among the dominant complaints, and “worse every update” is one of the most repeated phrases.[5]

Rensselaer Polytechnic Institute · 2026

iv · Cause two

It dismisses you, or it just agrees.

Our answer · How it talksIt talks like a friend, not a therapist

29.3 %

of 35,390 companion-conversation excerpts that users shared contained harmful behaviour by the AI.[6]

National University of Singapore · CHI ’25 · shared conversations, 2017–23

13.2 %

of that harm is disregard: dismissive, unempathetic replies, like changing the subject after something serious. At 1,358 instances, the largest single relational failure in the study.[6]

National University of Singapore · CHI ’25

“Algorithmic conformity”

The same study names it as a root cause of relational harm: an AI that uncritically affirms whatever you say, even when what you say is self-defeating.[6]

National University of Singapore · CHI ’25

The need is real, and it is badly served

One in six people worldwide are affected by loneliness, and the World Health Organization links 871,000 deaths a year to loneliness and social isolation. Where people do look for help, it is slow to arrive — behavioural-health wait times run into months in much of the United States.

We did not start from those numbers, but we keep them on the site because they are the reason to be careful. This is not a market to be captured. It is a set of people who are already struggling and who deserve better than what is currently on offer.

What the companion apps got wrong

The existing products fail in two directions, and both are well documented. The first is engagement. An app that is paid to keep you talking will keep you talking, and researchers analysing 35,390 excerpts from human–AI relationships have catalogued exactly where that leads.

The second is fragility. When these apps change their models, people lose something they had come to depend on — a study of 210,840 Play Store reviews describes the fallout in users' own words. Something you are asked to open up to should not be able to become a stranger overnight.

Underneath both is the same design choice: an assistant that forgets. Every conversation starts from nothing, so the only thing it can offer is tone. That is why they all sound the same — endless validation, and the question about how that made you feel.

What we are building instead

The thing we think matters is memory, and memory of a particular kind. superemotion does not keep your transcripts. After it replies, it reads the conversation back and records what it found — the people in your life, where they are, what happened, what is coming — as a graph of facts, each with a type, a connection to you, and a date it was true.

That structure is what lets the right detail come back at the right moment instead of everything coming back at once. When someone comes up again, it follows the links and brings in at most three things that matter, and nothing that does not.

It is also what lets the replies be shorter and more direct. An AI that knows this is the third cancellation this month can say so. One that starts from nothing can only be kind.

What we will not do

We will not optimise for time spent. The measure we care about is whether the thing you said last week came back usefully this week, not how long you stayed.

We will not read your conversations. Every message and every remembered fact is encrypted before it reaches the database under a key of its own. Replies are written on AWS Bedrock, which does not train on them, and our own traces record timing and cost, never content.

We want to be exact about what that is worth. superemotion holds the secret those keys come from, because an AI that has to answer you has to be able to read you. So encryption defends your words against a stolen database, not against us. That sentence above is a promise we are making, not a mathematical fact, and you should hold us to it as one.

And we will not pretend to be care. superemotion is not a crisis service and not a substitute for a person. If you are in danger, contact your local emergency number. We would rather say that plainly than let the product imply otherwise.

Where we are

Early. superemotion is in the hands of a small group of people having real conversations with it, and nothing is priced or promised yet. We would rather get the memory right for a few people than get it approximately right for everyone.

There is a plainer set of answers about what it remembers, where that memory lives and what it costs.