Memory can tell you what happened. Continuity is what lets what happened still matter.
One idea runs through all six: an AI that you or your customers keep. It’s built to hold on to what matters over months of conversation and across model changes — and to be changed by what you’ve been through together, not just reminded of it.
For companionship
What you get: a companion that grows with you over time, not one that resets.
If you’ve seen 50 First Dates, you know the setup: the same first day, every day.
Most AI companions live there. You can tell one your whole story, and next week it’s a stranger being polite. If it does remember, it usually remembers the way a file remembers: it can quote you back, but nothing that happened has changed it.
LSF is built so a companion is still the one you know next month and next year, and after the model underneath is replaced — without the details that matter being summarized away — and so the time you’ve spent together has changed who’s talking to you.
For caregiving
What you get: the same companion, with the accountability to be trusted where the stakes are higher.
Where the stakes are higher, “it remembers” isn’t enough. It has to be trusted — and it has to be understandable.
A caregiving companion sits with someone for months: routines, family, what a good day looks like, what a hard one looks like. Today, a model update can quietly reset the character that person had come to know, and nobody can say what changed.
LSF is built so the companion stays the same someone over months of conversation and across model changes, without the details that matter being summarized away — grown by what you’ve been through together. And it’s built so what changed her is inspectable: you can see what changed and why. Not because a rule says so today. Because the first time something goes wrong after an update, the operator will wish they had it.
Not a medical, clinical, or regulated-care claim.
For education
What you get: a tutor that grows with the learner and can show what changed.
A tutor who has known a learner for three years knows more than the answer key.
Learning is cumulative, and so is the relationship that carries it: what a student struggled with last spring, what finally made it click, how they respond to being pushed. Most AI tutors reset that with every update, and the ones that remember tend to remember the way a file remembers.
LSF is designed for the years-with-a-learner relationship: the same tutor across terms and across model changes, changed by what it has learned about this student. Teachers and parents can inspect what changed along the way — it isn’t locked inside the tutor.
For embedded products — devices, cars, robots
What you get: a continuing character your customers keep, whichever model you ship next.
Your product ships once. The model underneath it changes every few months.
If your device, car, or robot has a character customers talk to, every model update is a quiet risk: the voice they know comes back a little different, and there’s no way to say why. Swapping models is a fact of life; it isn’t the hard part. The hard part is keeping the someone your customer has been living with — and letting the time they’ve spent together count.
LSF is designed to be licensed into hardware and software you already sell. It’s built so the character carries on across sessions, devices, and changes of the underlying model, with what changed it open to inspection. Designed to run local-first; cloud optional.
For research
What you get: an AI identity you can study for years — and defend in a methods section.
Long-running studies need the same identity at the end as at the start.
If you study how people and AI relate over time, the model swap under your subject is a confound: the identity you were measuring quietly becomes a different one, and the record of what changed usually doesn’t exist. Continuity is a research condition before it’s a product feature.
LSF is designed to give a study a continuing AI identity that stays the same someone over months of conversation and across model changes, without the details that matter being summarized away — grown by what it has been through with your participants, with a record of what changed and why that your team can inspect.
Bounded longitudinal evidence from our own long-running work exists, as internal analysis; we share it under NDA, by request, in stages.
For role-play and gaming
What you get: characters players keep, across the titles you ship next.
Players remember the betrayal in chapter three. The character should be different for it.
Persistent characters are the promise of the medium — the companion across three titles, the rival who holds a grudge, the world that knows what you did. Today that persistence tends to break at the model swap: same name, same lines, someone else underneath. And a character that only remembers is a lore file with a voice.
LSF is designed so a character stays the same someone over months of play and across titles and model changes, without the moments that matter being summarized away — and is grown by what the player has done with it. Not just remembered. Changed by it.
