What data does AI Twin hold about me?

Your Twin holds what you put into it. Files you upload, text you type, answers you give during smart onboarding, and the metadata around those items. Your Twin does not scrape your inbox, read your calendar, or watch your screen.

How is my data encrypted?

Your data is encrypted in transit and at rest. Access to your data inside AI Twin is logged, with every read, write, and change recorded with an actor and timestamp.

Can I delete my data from AI Twin?

Yes. When you cancel or ask us to delete, we soft-delete your data immediately and permanently anonymise it within 30 days across production. Operational backups roll off automatically.

Does AI Twin use my data to train AI models?

No. AI Twin does not use your data to train AI models, and the model providers we work with do not receive a feed of your memory to retrain their systems. To write a Brief or a reply, the memory relevant to your question goes to our model provider under a data processing agreement, and the answer cites what it drew on.

Is AI Twin built to UK GDPR?

Yes. AI Twin is built to UK GDPR and EU AI Act standards from the first user. We can answer Data Subject Access Requests and support the right to erasure. We say built to, not certified against, because certification words are earned with published paperwork; we are not yet certified to ISO 27001 or SOC 2, and we are engineering toward both.

What does the EU AI Act require of AI Twin?

AI Twin sits in the limited-risk tier as we read it today, so the duties are about transparency, and they have applied since 2 August 2026. You are told when you are dealing with AI, every chat reply and Brief carries an AI label, and how we use AI is published on our AI transparency page. We are not formally assessed and do not claim to be.

Trust

Your data. Your rules. Calibrated honestly.

Q: Do you hold Cyber Essentials, ISO 27001 or SOC 2?

Not yet. ISO 27001 and SOC 2 are engineering targets, not current badges. Cyber Essentials is in progress; published when the certificate exists. An independent penetration test is planned for the first quarter of 2027 and has not yet taken place. We do not promise a date until it has. When it has, this page will say so and give the date. We publish each only when it is independently verified.

Built to UK GDPR and EU AI Act standards. We say built to, not certified against, because certification words are earned with published paperwork. When ours is published, this page will say so. AI Twin Limited is the data controller. Your memory and files are stored in the United Kingdom. AI inference, speech to text and email are processed in the European Union. Nothing leaves the UK or the EEA. Every action is logged in an audit trail you can see. You can export everything you hold.

Version 2.2. Effective 10 September 2026.

At a glance

Eight things, in one screen.

Where it lives: Your memory and files are stored in the United Kingdom. AI inference, speech to text and email are processed in the European Union. Nothing leaves the UK or the EEA.
Encryption: In transit and at rest.
Isolation: Row-level security enforced on every database table.
Audit: Every consequential read and write lands in an append-only audit log you can see. Entries are never edited. What we keep, and for how long, is on our retention page.
Consent: Captures wait for your confirmation. Nothing acts on your data or accounts until you confirm.
Export: Everything you hold, on demand, in standard formats.
Deletion: Deleting your account soft-deletes your data immediately and permanently anonymises it within 30 days. Deleted items can be restored from Trash until the anonymisation runs, in case of a mistake.
Not yet: ISO 27001 and SOC 2 are engineering targets, not current badges. An independent penetration test is planned for the first quarter of 2027 and has not yet taken place. We do not promise a date until it has. When it has, this page will say so and give the date. Cyber Essentials is in progress; published when the certificate exists.

Built for privacy, by design

What that means in practice.

Data controller. AI Twin Limited, registered in Scotland (company number SC866947), is the data controller for all personal data we hold. We are ICO-registered.

Data residency. Your personal data is processed in the UK and the European Union. Hosting and database: United Kingdom (London). Model providers: European Union. Email delivery: European Union. Each category, with its role, region and data processing agreement status, is on our sub-processors page. We do not casually move your data outside this perimeter.

Audit. Every consequential read and write lands in an append-only audit log you can see. Entries are never edited. What we keep, and for how long, is on our retention page. Read the retention schedule.

Erasure. Deleting your account soft-deletes your data immediately and permanently anonymises it within 30 days. Operational backups roll off automatically. Deleted items can be restored from Trash until the anonymisation runs, in case of a mistake.

Sensitivity tiers. Memory entities carry one of four sensitivity tiers (LOW, MEDIUM, HIGH, CRITICAL). Access is enforced at the database level through row-level security. The system fails closed, not open. CRITICAL data is owner-only on Practice Workspaces.

AI transparency. AI Twin tells you when you are dealing with AI. Every chat reply and every Brief carries an AI label. Exports, including conversation exports, carry the same disclosure line. Nothing is saved or acted on until you confirm. What the AI does, and where its limits are, is set out on our AI transparency page.

See machine-readable company data (for AI agents)

ON-DEVICE DEMO

A note on the demo on our homepage.

The on-device demo on the homepage includes a short walkthrough, Show me the proof, so you can verify in your own browser that nothing you type there leaves your device.

The homepage demo lets you watch a memory come back without sending any of your text to a server. We mean it. Open your browser developer tools, the Network tab, filter to Fetch/XHR, and ask the demo a question. Zero outbound requests fire.

The demo does need to download an embedding model and a runtime to your browser on first use. About 22MB total, fetched once from huggingface.co and a public content delivery network (jsDelivr). Both are static files. Hugging Face and jsDelivr see only that a browser somewhere downloaded a public model. They do not see your question, your interaction, or anything else.

This is the model we want for the full product too: heavy lifting on your device wherever practical, the underlying LLM provider engaged only when we cannot do the work locally. The homepage demo is a small proof of how we think about this.

What we hold

What your Twin actually stores.

A short list, because the list is short.

Your Twin holds what you put into it. Every item below becomes typed memory you can read, edit, export, and delete. Captures wait in Pending Review until you confirm them.

Documents and PDFs. Documents, and the Facts and Events inside them. Contracts, letters, reports, manuals.
Photos and screenshots. Documents. A photo of a letter is a letter.
Voice notes. Documents, and the Facts and Events inside them. AI Twin does not record you. There is no always-on capture.
Notes you type. Facts, Events, People, depending on what you tell it. One Universal Input bar, on every surface.
Email you forward. Documents, Facts, Events. Forward to your Smart Inbox address. Nothing is read from your mailbox. Only what you send.
Onboarding answers. Your Style Profile, and the first Facts about you. Five or fourteen questions. It captures style, not personality.
WhatsApp chat exports. Facts, Events and People. You export the chat and upload it. AI Twin does not connect to WhatsApp and does not read your messages.
Social data exports. Facts, Events and People. You download your data from the platform and upload it. AI Twin does not connect to your accounts.

Your Twin holds the metadata around those items. When you added something. What kind of item it is. Tags and categories you set. Audit entries for every action.

Your Twin does not hold what you have not given it. We do not scrape your inbox. We do not read your calendar. We do not watch your screen. We do not infer things about you from places you did not bring to the Twin.

When wider integrations ship, they will roll out one at a time, behind your explicit consent. You will see exactly what each one captures before you turn it on. You can turn it off again at any point.

Your rights

What you can do with your data.

Under UK GDPR you have a set of rights. We have built AI Twin so that exercising them is direct, not a form-filling exercise.

See it.

Your Twin shows you what it holds. No black boxes. No “this is encrypted, trust us” surfaces.

Edit it.

Anything captured can be corrected. If your Twin remembers something wrong, you change it.

Export it.

One click produces a portable file containing everything your Twin holds for you. Standard formats, not a custom export only we can read.

Delete it.

Deleting your account soft-deletes your data immediately and permanently anonymises it within 30 days.

Object to it.

If we ever start processing your data for a purpose you did not opt into, you can stop us. You can also lodge a complaint with the Information Commissioner’s Office, the UK regulator, at any time. You should not have to. We would rather you tell us first, but you do not need our permission.

Take it elsewhere.

AI Twin is model-agnostic by design. Your memory is yours, and it comes with you if you switch tools or leave us entirely.

These are not features. They are the floor.

What we do not do

Five things you will not see us do.

We do not sell your data. There is no commercial arrangement under which any third party pays us for access to what your Twin holds. Our full list of sub-processors is public at /sub-processors.

We do not share your data with advertisers. AI Twin does not run ad tracking, does not embed advertising pixels, and does not exist inside an ad-funded business model.

We do not train AI models on your data. The model providers we use are on commercial terms under which customer content is not used to train their models. Their terms are cited, with the date we checked them, on our AI transparency page. To write a Brief or a reply, the memory relevant to your question goes to our model provider under a data processing agreement, and the answer cites what it drew on.

We do not track you across other websites. Aggregate analytics on this site only, focused on how pages are used. Nothing tied to your identity, nothing shared with networks that profile you.

We do not retain your data after you have asked us to remove it, beyond the published backup-rolloff window.

Model-agnostic

Your memory is yours. The model is whatever you choose.

A lot of AI products today lock you in. Your conversations live inside one provider’s system. Your context belongs to them. If you switch, you start again.

AI Twin is built differently on purpose.

Your memory is stored in our database, structured and typed and yours. The AI model that helps you use it is whichever one you want. Claude. ChatGPT. Gemini. A local model on your laptop. If a model gets better tomorrow, you switch. Your Twin’s memory comes with you.

This is not a feature for technical users. It is the brand. Lock-in is a tax on trust, and we refuse to charge it.

Compliance

What is live, what is in build, what is not yet certified.

We will not overclaim our compliance status. The space is full of products that print certification badges they have not earned. We are not one of them.

Built to UK GDPR and EU AI Act standards.

We say built to, not certified against, because certification words are earned with published paperwork. When ours is published, this page will say so. Lawful basis, data minimisation, purpose limitation, storage limitation, integrity, confidentiality, accountability. We can answer Data Subject Access Requests. We can support the right to erasure. We have appointed routes for complaints.

Limited risk, as we understand it today.

The EU AI Act sets duties by risk. As we read it today, AI Twin sits in the limited-risk tier, where the duties concern transparency. Those Article 50 duties have applied since 2 August 2026. The disclosure that you are dealing with AI is in the product, how we use AI is published on our AI transparency page, and the dated timeline, including the July 2026 deferral of the high-risk rules, is on our UK GDPR and AI memory page. If our reading changes, this page will say so.

SECURITY

Security, stated plainly.

“Banking-grade” is not a defined standard, so we won’t borrow the phrase. But the controls people usually mean by it are here. Every table in our database enforces row-level isolation at the database itself, not just in the application. Every consequential read and write lands in an append-only audit log you can see. Entries are never edited. What we keep, and for how long, is on our retention page. No privileged keys ever reach your browser. Inputs are screened before anything is stored. Your memory and files are stored in the United Kingdom. AI inference, speech to text and email are processed in the European Union. Nothing leaves the UK or the EEA. Multi-factor authentication is available now and becomes mandatory from the fourth quarter of 2026.

What we have not yet earned the right to claim: ISO 27001 and SOC 2 are engineering targets, not current badges. Cyber Essentials is in progress; published when the certificate exists. An independent penetration test is planned for the first quarter of 2027 and has not yet taken place. We do not promise a date until it has. When it has, this page will say so and give the date. Each is published when it is independently verified, not before.

Your data is encrypted in transit and at rest. That is the baseline, not the brand statement. If something goes wrong, we will tell you in plain English within 72 hours of becoming aware. UK GDPR sets that as the floor; we treat it as a floor, not a ceiling. If a personal data breach ever touches a practice's data, we tell that practice without undue delay, so it can meet its own duties to the ICO and to the people affected.

Where your data lives

Who else touches it, and from where.

AI Twin runs on a small set of trusted providers. The list is short and deliberate.

Hosting and database: on infrastructure inside the United Kingdom (London). UK data residency where possible, EEA where not.

AI models: model providers in the European Union, under a data processing agreement. Memory is sent to a model only when you make a request that needs it, and the answer cites what it drew on.

Email delivery: a transactional email provider for confirmation and account messages. No marketing email from your data without explicit consent.

Analytics: privacy-friendly analytics with no cross-site tracking and no identity tying. Aggregate signals only.

That is the list. If it changes, we update this page and tell people who have signed up. We do not quietly add sub-processors and hope no one notices.

For practices

If you are the one doing the vendor review.

AI Twin is a memory and productivity aid. It is not a system of record. It supports, and does not replace, the record-keeping and retention obligations your professional body and HMRC place on your firm. You remain responsible for your authoritative records.

If a personal data breach ever touches a practice's data, we tell that practice without undue delay, so it can meet its own duties to the ICO and to the people affected.

Practices that need the underlying paperwork for their own vendor review can ask for our due-diligence pack: named sub-processors with regions and data processing agreement status, our retention schedule, our incident process, and answers to the questions compliance partners ask. Email hello@ai-twin.co.uk.

Sensitive data

Who AI Twin is for, and what we ask you to think twice about.

AI Twin is built for adults. We do not build features designed for use by children under 18. We do not market to children. If you are under 18, please do not sign up.

You may, of course, capture things about your family in your Twin. The school’s term dates. The nursery’s dietary letter. Your child’s medication. That is one of the use cases the product is built for. The data is held by you, in your account, as part of your life admin.

Some data is more sensitive than other data. Health information. Faith. Sexuality. Political views. Financial detail. If you choose to capture these in your Twin, you should. It will help your Twin help you. We hold this data with the same care as everything else, and we have specifically not built features that profile, segment, or commercialise based on sensitive categories. Nothing about you triggers a different experience for someone else.

If you ever want to know exactly what your Twin holds and where, that is what the audit log is for.

Security issues

Report a security issue.

If you find a security issue, email hello@ai-twin.co.uk. We reply within three working days, we do not ask researchers to sign anything, and we do not take action against good-faith research. We credit you if you want us to.

Our security.txt is at /.well-known/security.txt.

Contact

A real person reads every message.

For privacy questions, data subject requests, or anything that does not feel right: hello@ai-twin.co.uk.

For complaints you do not want to bring to us first, you can contact the Information Commissioner’s Office at ico.org.uk. We would rather you tell us first, but you do not have to.

Calm AI. Clear consent. Quiet by design.

We will note material changes here and tell people who have signed up.

This page describes our practices at the date shown. It is provided for information and does not form part of, vary, or add any warranty to any contract between us.

Change log

10 September 2026. Penetration test position restated: planned for the first quarter of 2027, no date promised until it has taken place.
10 September 2026. Added the at-a-glance summary, data residency by layer, the dated EU AI Act position, multi-factor authentication status, the practices section, how to report a security issue, and the information-not-warranty line. Corrected the audit and erasure wording to match our retention schedule.
10 September 2026. AI transparency and EU AI Act paragraphs brought up to date and linked to our AI transparency page. Where data is processed now reads from our sub-processor categories. Clarified how memory reaches the model provider.
1 September 2026. Rewritten for launch. Company details, tax position, subscription terms and complaints route added.

AI Twin Limited is a private company registered in Scotland, company number SC866947. Registered office: 272 Bath Street, Glasgow, G2 4JR. Registered with the Information Commissioner's Office, registration ZC044489.