The Creomind Manifesto

Human beings leave enormous amounts of evidence of what happened to them, and very little structured evidence of how they understood it.

We inherit photographs without the reasoning behind the decisions in them. We inherit letters without the argument the writer was having with themselves. We inherit a voice on an answering machine and no way to ask it anything.

Every generation has closed part of this gap with the tools it had. Portraits preserved faces. Photography preserved appearance. Audio preserved voice. Video preserved moments.

We are now beginning to preserve something harder: the patterns behind how an individual understood the world.

Not consciousness. Not resurrection.

A deliberately created record of a human perspective, structured deeply enough that parts of it can be explored long after the original conversation ends.

That distinction is not modesty. It is the whole design. A record built from real human evidence can be checked, questioned and traced. A claim to have captured a mind cannot.

The richest model of a person is the one they deliberately help create.

Creomind does not assemble people from scraped traces. We sit with someone while they are here to decide what matters, and we ask the questions that nobody thinks to ask until it is late: what changed your mind, what did you stop believing, who were you before you became my mother.

Consent is not a compliance step at the start of the work. It is the reason the work is worth anything.

A person is not a dataset, and they are not a summary either.

Identity is layered and contextual. It holds contradictions that a well behaved system would quietly average away. A record worth keeping has to make room for the opinions someone defended, the ones they softened, and the years in between.

So we keep the source and the inference apart. What someone said is evidence. What a model produces is inference. Both are useful. Confusing them is the failure mode we design against.

Presence on the surface. Provenance underneath.

A conversation should feel like a conversation. Someone answering in the first person, in a voice you recognise, is not a trick as long as you are never in doubt about what you are talking to. Generated speech stays marked as generated, in the interface and in the export, and the evidence behind it stays one step away.

A model should also know what it does not know. Where the evidence runs out, the honest answer is to say so before answering, and to name a speculative answer as speculative.

This is a long project.

Creomind builds slowly, for a small number of families at a time, with a horizon measured in generations rather than releases. The archives we help create are intended to outlast the software they were built on, which means open formats, portability and a refusal to make continuity depend on our own survival.

We would rather take six months over something than ship it in six weeks and call it a person.

Some people find this uncomfortable, and we take that seriously.

The objection is usually that this denies death, or interrupts grief. The people who have already lost someone rarely ask whether it should exist. They ask why the question they wanted to ask was never recorded.

Grief does not need to be denied to be made bearable, and memory does not need to be silenced to be honoured. There is a version of this technology that exploits longing, and a version that respects it. We have chosen the second, and the principles page exists so that choice can be checked rather than believed.

The mission is simple, even when the execution is not.

Help people leave a faithful record of their own perspective, made with their consent, structured well enough that the people who come after can still learn something from it.

There are questions you have not thought to ask yet. The archive should still be able to meet them.