AI Twin: A Typed, Audited Memory Layer for Privacy-Conscious Professionals

The dominant model for AI memory today is: the tool keeps a log of your conversations and uses it to improve future responses. The provider controls what's stored, how long it's kept, whether it's used for training, and whether you can fully delete it. For casual use, that trade-off is largely invisible. For a GP managing patient notes, a solicitor tracking case context, or a consultant who regularly asks AI for help with client-sensitive decisions, it's a fundamental problem. AI Twin is built on a different premise. The platform stores a typed, audited, user-controlled memory — not a conversation log — that sits above any AI model you choose to use. Ask a question via the AI Twin interface, forward an email, capture a voice memo, or paste a document: the system classifies each input into a typed category (Person, Document, Fact, Event, or Episode), assigns a sensitivity tier, and adds it to an append-only audit trail. When you query your memory later, you get the structured record — not a hallucinated reconstruction from a year-old chat. The positioning is deliberate: 'The product is the memory underneath: typed, audited, yours. Chat is one door in, not the whole product.' That distinction matters for anyone who has been frustrated by AI tools that forget critical context between sessions, mix up client details, or surface information you didn't intend to include in a given query. It also answers the objection at source: this is not a smarter note-taking app. It is a structured memory layer that AI queries against — with access controls and an audit trail the note-taking category has never needed to build.

Universal Input accepts five entry types: text, forwarded emails, voice memos, slash commands, and URLs. Each input gets classified and routed to the appropriate memory category. A forwarded email receipt becomes a Fact. A meeting invitation becomes an Event. A voice memo summarizing a client conversation gets structured as an Episode. The typing is intentional — it means that when you query 'what medications am I on,' you get the structured medication record, not a match from an old chat message where medications were mentioned in passing. Retrieval is direct: query your AI Twin memory with a natural language question or a typed filter, receive a structured Brief in return, and use that Brief as context in whichever AI assistant you are running — Claude, ChatGPT, Gemini, or a local model. The memory does the organizational work so the AI does less guesswork. Hybrid Retrieval combines typed filters with natural language search across the full memory layer. The Smart Inbox turns email forwarding into an active capture mechanism — a forwarded receipt, invitation, or document is converted directly into typed memory without manual classification. Smart Tasks add scheduled intent: a Daily Brief (scheduled at 7:30 am) surfaces upcoming events and pending memory items with a confirmation step before anything runs. The confirmation model matters — nothing executes without user sign-off, consistent with the broader audit-and-consent architecture.

Most AI tools add compliance language after the core architecture is fixed — a GDPR statement on a US-hosted product whose data processing was never designed for European regulatory requirements. AI Twin's compliance posture is the opposite: UK/EU data controls and privacy are structural invariants, not settings you configure. Data is stored in the UK (London). AI inference and email processing run within the EU. Encryption covers data in transit and at rest. The platform's 'we do not train models on your data — full stop' commitment is contractually enforced through its model providers (Anthropic and Google Vertex AI), not simply a marketing statement. Sensitivity tiers — LOW, MEDIUM, HIGH, CRITICAL — are enforced via row-level security at the database level, not at the interface. When you mark a memory entry CRITICAL, that classification gates access before any query reaches it. The append-only audit log is directly addressable: every meaningful access event is logged in a trail that no one — including AI Twin — can edit or remove. The product is registered with the ICO (ZC044489). It is built to UK GDPR and EU AI Act standards; it is not yet ISO 27001 or SOC 2 certified (those are listed as engineering targets). For regulated professionals who have avoided AI adoption because they couldn't answer 'where is this data, who can see it, and can I delete it,' the ICO registration, RLS enforcement, and append-only audit log provide meaningful structural assurance today.

Beyond individual use, AI Twin supports a Practice workspace model for professional teams. The owner's Personal subscription (£18/month) is required, and the Practice add-on sits at an additional £25/month per workspace — making the base for a solo Practice owner £43/month. A solo professional who does not need workspace sharing pays £18/month. Team members join at £9/month each with configurable read-only or read-write access. Up to five Practice workspaces can exist per account owner. The permission model extends to sensitivity tiers: shared workspace members do not automatically inherit access to HIGH or CRITICAL-tier memories. A practice owner can share client intake notes (LOW/MEDIUM) with an assistant while keeping privileged legal or medical records (HIGH/CRITICAL) inaccessible to the shared workspace. This is the permission granularity regulated teams need before routing anything through a shared AI layer. Worked pricing examples from the product site: Practice with 3 team members runs £70/month; Practice with 5 team members runs £88/month. Annual plans reduce the effective rate by two months. When the 7-day trial ends, memory is preserved for 30 days — access resumes the moment you subscribe, so there is no pressure to commit before you have fully evaluated the workflow.

The clearest use case is a professional in a regulated industry — a solicitor tracking client context across multiple matters, a GP managing patient records, an accountant with multi-client financial context, a therapist keeping session notes — who has already decided they want AI assistance but cannot use tools that don't provide auditability and data residency controls. AI Twin is the answer to 'I want to use AI for client work, but I need to be able to explain what it knows and demonstrate I can delete it.' Individuals managing complex personal life context — household logistics, caregiving information for a parent, health records for a family — are the second strong fit. Domain-specific workspace bundles (Household, Caregiver, Career, Tax/Finance, Travel, Business) configure contextually relevant memory categories, and the typed structure means a health query returns actual health records rather than a mix of health-adjacent chat history. Founders and consultants who regularly context-switch across clients, projects, or relationships — and want AI assistance without context bleed between different engagements — are well-served by the workspace separation model. Where the product is less suited: if you want a fully ambient AI that passively observes your devices or screen in real time, that is a different category. AI Twin's capture model is intentional — you forward, paste, speak, or type inputs deliberately. That architectural choice is the compliance guarantee; passive capture and GDPR-grade audit control are fundamentally in tension.