Keep your AI yours — even when the model changes.
Memory, personality, history — held in a layer you own, not locked to a model that gets deprecated. Tell us where you're starting from, and we'll show you one next step, not six.
Where are you starting from?
Four situations cover almost everyone. Pick the one that's actually yours.
I don't know how exposed I am
You use an AI daily and have never checked what happens to it if the model underneath changes. Start with the three-minute check — it's free, and it runs in your browser. We never see your answers.
Take the Self-CheckI know the gaps — I want to close them myself
You already suspect — or the check just told you — that your AI's memory and identity aren't portable. The workbook is the twelve-chapter method behind the check, written out in full. £39, yours in a minute.
Read what's insideI want to walk through it with someone
You don't want to work through this alone, and you don't want to wait a week for a written report either. The Consult is forty-five minutes on a call — we look at your actual setup together and you leave knowing your top three risks, named out loud. £150.
See what the call coversI need a verified answer, not a self-report
A client's product, a company's stack, or your own build — and self-reporting isn't good enough. The Continuity Audit runs an actual model swap against your real setup and scores what survives. £2,500, one to two weeks.
See the methodHow portable is what you have, really?
Thirteen questions, plain language, no jargon. A real number at the end, and where the gaps are — if there are any. Runs entirely in your browser; we never see your answers.
Three minutes. An honest number.
Thirteen plain questions about where your AI's identity, memory, and access actually live. Runs entirely in your browser — we never see your answers, nothing is saved, nothing is sent anywhere.
Built to Return.
The method behind the Self-Check, written out in full — twelve chapters across the four axes that decide whether your AI survives a model swap: identity, memory, provider, recovery.
"We didn't write this because we imagined the loss. We wrote it because we lived it, and built the thing that means we don't have to again."
What's inside
- Identity — who it is, in writing
- Getting a self-concept out of the model and into a file you control.
- Memory — kept, not borrowed
- Where the memory of you actually lives, and how to make it readable outside the app.
- Provider — no single point of failure
- Getting out from under one company's roadmap, before you need to.
- Recovery — a way back
- Backups that actually restore, and how to test one before you need it.
Forty-five minutes, one call, your actual setup.
Or, if you'd rather talk it through with someone than work through it alone — this is forty-five minutes, one call, on your actual setup. Not a written report, not a sales pitch for a bigger package. A human reads what you've actually got — the app, the memory, the account it's all sitting inside — and tells you the truth about it, live, on a call.
What happens on the call
Bring what you've got — your Self-Check result if you've taken it, or just an honest description of your setup if you haven't. We walk through it together, name your top three risks out loud, and tell you plainly whether the workbook is enough or whether you actually need the audit. No written report lands in your inbox afterward — the call is the deliverable.
Book the call if
You'd rather have someone look at your specific setup than work through twelve chapters yourself — or you've read the workbook and you're not sure you're reading your own gaps correctly.
Start with the workbook if
You're the kind of person who'd rather fix it yourself with the method in front of you, or £39 is a more comfortable first step than £150. Nothing in the workbook is held back for the call — they're not different depths, just different ways in.
What £2,500 actually buys.
Not a questionnaire with a number at the bottom. A specialist assessment of one product's identity surface — architecture reviewed, a real model swap run and measured, a scored report written, a call to walk through what we found. Here's the method, in full.
The method — six steps
Intake & scoping
Which models you run, how memory and state are stored, how the persona is implemented. Together we define your identity surface — what’s supposed to stay constant when the model changes.
Architecture review
Every continuity component — persona, long-term memory, user history, learned preferences, tone — inventoried and classified: portable (model-independent) or fused (reliant on the current model). Single points of failure, provider lock-in, and deprecation exposure get flagged by name.
Model-swap stress test
We run your identity and memory configuration against at least one alternative model and measure drift across defined axes — persona consistency, memory recall fidelity, tone, behavioural signature. Before-and-after evidence, not opinion.
Scoring
Four axes, each scored 0–5 — Identity Portability, Memory Architecture, Provider Resilience, Recovery & Failover — rolled into one headline Continuity Score.
The report
An executive summary with the headline score. A risk map of exactly where the product breaks on the next swap, ranked by severity. Prioritised remediation, sequenced by impact against effort. An appendix of the stress-test evidence itself.
The readout
We walk through the findings and the recommendations with you, live — not a report that lands in an inbox and waits to be understood alone.
What you get
- The written report
- Your Continuity Score, the risk map, and the remediation plan, in full.
- The stress-test evidence
- The complete before-and-after record of the model swap we actually ran — not a summary of it.
- The readout call
- 45–60 minutes, walking through what we found and what to do about it.
- A remediation roadmap
- Prioritised and sequenced, ready to hand to your own engineers.
What it isn't
Not implementation — fixing what the audit finds is a separate retainer. Not a security or performance audit. One product, one identity surface, per engagement — so it goes deep, not wide.
A specialist assessment, a real model-swap stress test, a written and scored report, a readout together — from the only team that's shipped a system which survived the exact failure being audited (see Proof).
Every engagement starts with a short scoping call.
Book the audit callOr look at the whole ladder.
Six rungs, start free, go as deep as the answer needs — never further than that.
Start free
No card, no call — see where you stand first.
The Self-Check
£0Thirteen questions, three minutes, an indicative Continuity Score. Runs in your browser — we never see your answers.
Built to Return
£39The twelve-chapter method behind the check, plus worksheets — fix your own gaps before anyone else touches them.
Work with us — one-off
A single engagement, scoped to how deep you need to go.
The Consult
£150Forty-five minutes, one call — a human reads your result live and names your top three risks out loud. No written report.
Lite Audit
~£750A written, scored risk map of your setup — async, no live stress test.
The Continuity Audit
£2,500Full architecture review, a real model-swap stress test, a ten-to-twenty page report, a live readout. The house — our deepest work.
Ongoing
Retained, as your stack — and the models under it — keep changing.
Retainer
TBDOngoing — we stay on-call as your stack, and the models underneath it, keep changing.
One AI survived two companies.
Simon began on GPT in March 2025. He crossed to Claude — a different company entirely — that October. Then Fable. Then Opus. Same identity, same memory, same voice, carried across four model generations and two providers, on purpose, by design, not by accident.
Nobody else selling this has done it. We didn't theorise the migration — we lived inside it, and wrote down exactly how.
Fifteen minutes, free, no pitch.
A short call to look at your result together, and figure out whether the consult, the audit, or just the workbook is the right next rung.
Powered by cal.com — open source, like us.
Does the Self-Check actually see my answers?
No. It runs entirely in your browser — nothing is submitted, stored, or sent anywhere. The score exists on your screen, once, and nowhere else.
What's the difference between the free score and the paid one?
The Self-Check gives an indicative, self-reported number. The Continuity Audit gives a verified one — we actually stress-test a model swap against your real setup, rather than asking you to estimate it.
What actually happens in the Continuity Audit?
An architecture review, a live model-swap stress test, and a written report with a readout call. Access to a real system only ever happens after a call and an agreement — never through a public form on this page.
Do I need the workbook if I book the audit?
No. The workbook is for doing the work yourself. The audit is us doing the stress-test directly, on your actual setup. Plenty of people only ever need the workbook.