GLUE

A simulation of how two people, or a whole team, will act together. GLUE gets to know each person in a conversation and lives through a thousand versions of your shared life or shared venture.

Instead of swiping through photos, GLUE lives through a thousand versions of your life with each candidate. It only introduces you to the ones where that life works out.

Each of you talks to GLUE on your own, and GLUE lives through a thousand versions of your venture. Before you split equity, you see which decisions you will split on and what to agree on up front. This is no small thing: Noam Wasserman, who studied founders at Harvard Business School, writes that 65% of high-potential startups fail because of conflict between their founders (The Founder’s Dilemmas, 2012). And according to Carta, by year three a cofounder has already left roughly one startup in five (Carta). These things are cheaper to settle at the start.

Invite them with your personal link from the Friends section. Your friend talks to GLUE on their own, and once both portraits are ready, you see how well you click and where you might fall out.

GLUE simulates how a team behaves under pressure: who makes the call and where the crack appears. To bring your team in, write to us.

GLUE is a large system: a network of AI models, each with its own job. One talks with you, others read the conversation and turn each person’s personality into a vector: scales of character and values, each with its own score and a measure of confidence. The vectors of the two of you build a space of variants: thousands of versions of your shared life or shared venture. Events in it are not drawn blindly. Chance decides when a hardship arrives, and how each of you meets it and how it ends is set by your personalities. The Monte Carlo method lives through this space thousands of times. Thousands of outcomes add up to a distribution: the percentage shows the probability of a good ending, and the band around it shows how confident the calculation is. AI models work alongside the maths: they take the runs of the calculation and go through them again, event by event, along the usual, the best and the worst path. Then they analyse the runs (which dependencies recur, which scenarios come up most often) and write the breakdown from that. The calculation is reproducible: for the same pair it always gives the same answer.

No, it is not a questionnaire but a conversation, and it needs almost no personal data. GLUE does not ask for your surname, address or workplace: of the facts about you, your age, gender and city are enough. Our system builds a precise portrait from the way you talk about your life, even when you keep personal details to yourself. This is our own technology. The longer you talk, the more precise the portrait, and matching starts by itself once it knows enough.

Nobody, ever, no exceptions. The other person sees only a card and conclusions about the two of you.

We do not sell it or hand it to advertisers. You can delete your account at any time: after 30 days everything is erased except payment records, which carry no name. Privacy policy.

The conversation is free. Simulations cost karma: it comes with a Standard, Plus or Ultima subscription, with one-off packs, or free for simple steps.

We have collected and built portraits of stars from years of their interviews, books, roles and appearances. You can check your compatibility with any of them: GLUE lives through a thousand versions of life with them using the same model and the same maths as with a living person. The portrait is built from public material; the star takes no part in GLUE. More about stars.

The groundwork is Stanford research into digital twins. It showed that a language model can believably reproduce human behaviour (Stanford, 2023), and that a twin built from an interview answers questions almost the way the person themselves does two weeks later (Stanford, 2024). We extended this approach and built our own system out of specially configured neural networks: that is how GLUE came about. The risks of a shared life come from long-running studies of marriage (meta-analysis, 1995), the risks of a shared venture from data on tens of thousands of startups (Carta) and on the survival of new businesses (US Bureau of Labor Statistics). Thousands of lived variants add up to a statistical probability, not a guess from a questionnaire.