What stays human

Where AI helps. What it never touches.

Memrease uses AI for the parts of family-archive work that would otherwise be tedious. There are parts of the work that aren't tedious; those stay yours.

The shape of the problem

AI can clone a voice from thirty seconds of audio. It can fake a photograph well enough to pass on a phone screen. It can write a paragraph in your mother's register from a handful of her letters. The world is heading toward an era when the usual ways of knowing what's real won't keep up.

For most things, the answer is institutional — cryptographic signatures, watermarks, provenance chains. For a family, the answer is closer to home. You don't need a hash to know your wedding photo is real. Your brother was there. Your dad remembers the speech. The people whose voices you trust are still the strongest evidence you have.

Memrease is built on that basis. The only stories that hold are the ones told by people you trust. Memrease's job is to keep those voices intact — not to substitute for them, not to speak over them, not to invent more of them.

Where AI helps

AI does the work that family archives used to need a paid archivist for. It is the assistant in the back room, not the voice on the page.

Describing a photo — only if you turn it on

On our Heritage tier, a family can switch on photo descriptions. When it is on, AI looks at a photo you upload and describes what is visibly in it — to help you write the story that goes with it, and to draft a few gentle prompts you review before anything reaches the person at the centre. It is off by default, the family's lead Keeper controls it, and you can turn it off at any time. The AI is never asked to identify or name the people in a photo, and every prompt it drafts is yours to confirm, edit, or decline.

What it never touches

These are the boundaries that do the most work. They are non-negotiable, and they are why Memrease can be a place where your family's voices are kept clean rather than corrupted.

Which providers we use, and why

Two, currently: Anthropic for the language work (the recipient conversation; thread extraction; summarisation; the guided interview; and, for Heritage families who turn it on, photo descriptions), and OpenAI for moderation and transcription. Both have written commitments not to train on customer data; we audit those commitments at every renewal.

The choice of providers is itself a discipline. We picked the two companies whose stated values most closely match the lines we've drawn above — not the cheapest, not the most fashionable. If either changes its policies in a way that breaks the commitments on this page, we'll change provider.

The models we use

We name the family of model rather than a version number, because the version changes more often than the commitment does. Today the language work runs on Anthropic's Claudemodels, and the safety screening on OpenAI's moderation model. When we move to a materially different model we note it here with the date, so this page always reflects what is actually running.

What this means for your family

The careful version: AI handles the librarian work; people handle the storytelling. Your archive doesn't hold what an algorithm generated about your family. It holds what the people in your family actually said about themselves, gathered into a place that's clean of the synthetic substitutes accumulating everywhere else.

That's what makes Memrease different from a photo backup, a social network, or a generative-AI scrapbook. The voices in your archive are the people you trust. Memrease is the binding around them.