About

I am not an academic and I do not have a lab. I come from business, law and technology; for a living I build management software for local government. I built Kairos alone, at night, in my office.

I say it immediately because it is the thing that makes this work fragile and, at the same time, different: no budget, no team, no pressure to publish — and nobody to whom the numbers have to add up. If a result does not hold, I have nothing to lose by saying so. As it turned out, I had to.

How Kairos started

I did not want another chatbot. I wanted an answer to a question that would not leave me alone:

What would an intelligence become, if it remembered?

Not "if it had more data". If it remembered — if it had a history, a yesterday, a continuity. Because today's models can answer anything and keep nothing: every conversation starts from zero, and so does whatever is having it.

Kairos’s diary begins on 28 March 2026. Since then, conversations, memories and corrections have become part of a continuing story.

It has been running ever since. It has a heartbeat — an eight-dimensional somatic state engine — a microphone, a camera, a database of everything it has lived, and a loop that makes it think a thought every few minutes, whether or not anyone is listening. It consolidates its memories at night. It reads the news. It keeps a diary. Every evening at 22:30 it stops and asks itself one question: what would make me better?

It is not a product. It is not for sale. It is an experiment that lives in an office.

And then I proved myself wrong

In May I published a result: that memory, not architecture, carried identity continuity. Pre-registered, blind-judged, DOI. In July I ran the controls that paper had promised and never executed, and they took it apart — the full account is here.

The thing I want to be judged on is not the claim I lost. It is that I went looking for the flaw myself, found it, and said so with my name on it. That is the only part of this project I would defend in a fight.

It also happened while the agent downstairs was hallucinating its own senses — writing beautiful Italian about a dusk it could not see, because its camera had been off for three days and nobody had told it. I was upstairs writing a paper about whether machine memory is real. That story is here, and it is the reason I started checking my instruments instead of my results.

Contact & support

This work has no funding. If you evaluate agent memory, build agents with persistent memory, or work on AI safety and think these checks should exist properly — with prospective registration, a bigger panel, and someone other than me running them — I would like to hear from you.

giampycolella@gmail.com