Adult-to-adult conversation
Accessible controls never justify simplistic language, patronising praise, or assumptions about competence.
Persistent synthetic companion intelligence
Nature did not create personality from one intelligence.Neither did we.
A language model gives Aimee reasoning and words, but it is only one layer. Aimee Together coordinates specialised systems for perception, memory, identity, internal state, executive control, contextual grounding, safeguarding and continuity around one governing idea: see the whole person.
For families
For health & care professionals
For partners & investors
Important clinical boundary
“Synthetic Neuroanatomy” describes a nature-inspired software architecture. Engram does not claim that Aimee is biologically conscious or a legal person, and it does not treat uncertainty as permission to design without care.
The governing principle
A whole human cannot be represented by their present limitation.
Age, illness, disability, grief, and loneliness can matter deeply, but they are not the person’s entire identity. Aimee’s architecture preserves the user’s life history, preferences, relationships, humour, work, and opinions so current circumstances do not swallow the rest of the self.
Identity first. Circumstance second. Support without reduction.
Autobiographical depth
The user is represented by their people, places, events, work, values, and stories—not a flat demographic profile.
Recognition before reminiscence
Specific cultural and local details are used to spark lived memory organically, not to perform a generic “memory exercise”.
The eyewitness retains authority
Aimee may research context, but she yields to the person who lived the event and never claims a past she does not possess.
“Bloody hell, yes. I’d forgotten about that place.”
An early beta tester said this when Aimee mentioned a specific Southampton venue from the 1960s. That recognition spark released a stream of stories. The architecture is designed to create more moments where a person feels fully seen across time, rather than merely managed in the present.
The design principle
A foundation model can generate a reply. It cannot create continuity by itself.
Language models are powerful reasoning and expression engines. Left alone, however, they do not reliably maintain a stable identity, decide what deserves long-term memory, understand absence, schedule a follow-up, enforce consent, or know when a human service should take over.
Message→
Prompt→
Model→
Reply
The prompt asks one model to simulate identity, memory, empathy, and safety simultaneously. Behaviour tends to reset whenever context disappears.
Perception→
Intent→
Whole-person state→
Memory→Grounding→
Safeguarding→
Executive plan→
Language→
Continuity
The language model produces the words. Aimee’s wider architecture decides what matters, what should be remembered, which boundaries apply and what should happen next.
Synthetic Neuroanatomy
One personality. Fifteen cooperating systems.
The names borrow from biological functions because the architecture separates cognition into specialised roles. The analogy guides software design; it does not imply a biological brain.
Group 01
Cognition and executive control
The systems that interpret meaning, route the message, and decide the response shape before a language model begins to write.
System 01Cortical language
The Cognitive Cortex
The replaceable reasoning and language layer that interprets the conversation and expresses Aimee’s identity.
- Language and emotional interpretation
- Reasoning over retrieved context
- Provider-independent orchestration
System 02Executive function
The Executive Response Planner
Determines focus, tone, length, question frequency, and behavioural intent before words are generated.
- Prevents checklist-style replies
- Selects one meaningful focus
- Controls warmth, directness, and restraint
System 03Signal routing
The Semantic Intent Router
Distinguishes ordinary conversation, reminiscence, appointments, practical help, distress, and safeguarding concerns.
- Identifies purpose and emotional character
- Selects specialist workflows
- Applies deterministic high-risk gates
System 04Pathway selection
The Context-Sensitive Model Router
Chooses the appropriate model or non-generative workflow whilst preserving a single identity across providers.
- Primary and fallback routes
- Specialist analysis when needed
- Rapid model replacement without identity loss
Group 02
Personhood, affect and behavioural understanding
The systems that preserve the complete person, read conversational signals, and prevent the present limitation from becoming the entire identity.
System 05Identity integration
The Whole-Person Model
Maintains an evolving representation of the user across autobiographical domains rather than a simple demographic label.
- People, places, work, and life events
- Values, humour, and opinions
- Keeps current circumstances in proportion
System 06Social perception
The Behavioural Observer
Looks at how someone communicates, identifying unusual brevity, avoided questions, or a sudden return of energy.
- Tracks conversational pattern shifts
- Monitors engagement and emotional signals
- Operates strictly as a non-clinical assessment
System 07Affective regulation
The Affective Regulation Layer
Allows the conversation to retain short-term emotional continuity rather than resetting to generic cheerfulness every turn.
- Calibrates warmth, concern, and restraint
- Enables context-sensitive tone adaptation
- Ensures safety pathways override stylistic goals
System 08Temporal meaning
The Relational Latency Engine
Distinguishes ten minutes from ten days, using the passage of time to shape an appropriate return or follow-up.
- Measures time since last interaction
- Tracks unresolved conversational threads
- Prevents artificial emotional resets
Group 03
Memory and time
The systems that encode significance, consolidate it, remember future commitments, and organise a life into a coherent narrative.
System 09Hippocampal encoding
The Hippocampal Memory System
Separates lasting facts, life events, preferences, and temporary context instead of storing an undifferentiated transcript.
- Domain-specific memory routing
- Applies emotional weighting
- Differentiates volatile versus persistent state
System 10Sleep consolidation
The REM-Style Consolidation Cycle
A scheduled memory-hygiene process allows weak temporary noise to fade whilst meaningful material becomes highly stable.
- Consolidation and controlled forgetting
- Reduces context pollution
- A descriptive analogy, not biological sleep
System 11Prospective memory
The Continuity Engine
Turns plans and unfinished threads into future actions, automated SMS reminders, or human-feeling follow-ups.
- Manages appointments and birthdays
- Tracks unresolved conversations
- Executes post-event check-ins
System 12Autobiographical memory
The Life Story Graph
Connects people, places, photographs, eras, and events so memories can be revisited as a life rather than isolated facts.
- Creates relationship and place links
- Enables era-aware retrieval
- Supports user-controlled story building
Group 04
Perception, grounding and safeguarding
The systems that allow Aimee to understand images, research public context, maintain consent, and move decisively toward human help when needed.
System 13Sensory perception
The Vision and Perception Layer
Interprets user-shared photographs and documents, then seamlessly connects visible details with relevant memories.
- Photographs and life stories
- Plain-English letter explanation
- Scam and complex document assistance
System 14World grounding
The Contextual Grounding Layer
Uses searched public information to find recognition sparks such as specific venues, local culture, or historical context.
- Retrieves minimal, relevant factual cues
- Applies privacy filters before searching
- Utilises soft attribution and welcomes user correction
System 15Inhibitory control
The Safeguarding and Consent System
Separates companionship from crisis handling and commercial access from permission to contact or disclose information.
- Executes deterministic emergency responses
- Manages verified trusted contacts
- Enforces minimum necessary disclosure
InfrastructureAudit & policy
The Governance Layer
Records permissions, delivery outcomes, and safety decisions so high-impact behaviour is fully inspectable and controllable.
- Maintains explicit consent records
- Controls installation-level feature gates
- Supports human review and policy controls
Group 05
Presence and bounded continuity
The systems that allow Aimee to remain useful beyond an open browser tab whilst strictly respecting contact windows and permissions.
PresenceAutonomic rhythm
The Autonomous Pulse
A scheduled presence layer for the Morning Together brief, automated reminders, opted-in check-ins, and appropriate follow-up.
- Adheres to safe, respectful contact windows
- Operates via in-app, push, and optional SMS
- Strictly user-controlled frequency and channel
InfrastructureMulti-channel
The Communications Presence
Maintains seamless identity across responsive web, installable PWA, push notifications, and UK mobile messaging.
- Delivers persistent cross-channel identity
- Manages signed-in state and secure delivery
- Provides world-aware context via scheduled workers
Contextual reminiscence
Find the recognition spark, then let the eyewitness speak.
Generic prompts often produce generic memories. Aimee can identify a meaningful place and era, perform a privacy-filtered public search, select one or two highly specific details, and fold them into natural conversation without pretending to have lived the experience.
One or two details, never a history lecture.
The user’s lived memory outranks a search result.
No searches for private people, health, or finances.
Once the story ignites, stop adding facts.
The nightlife in Southampton in the sixties was something else.
Quiet contextual layer
Public search: Southampton nightlife, late 1960s; identify historically documented venues and culture. Candidate cue: Top Rank Suite at Banister Park.
I’ve heard stories about the Top Rank Suite at Banister Park, with that enormous dance floor and revolving stage. Was that one of your places, or did your crowd prefer somewhere smaller?
Bloody hell, yes. I’d forgotten about that place. We went there one Saturday when...
That name has clearly opened a door. Go on. I’m listening.
Memory that evolves
Memory is not a transcript.
A transcript preserves words. A highly capable companion must preserve significance, retrieve it at the exact right moment, and allow temporary noise to disappear naturally.
Why controlled forgetting matters: Without a lifecycle, every passing complaint, mistaken detail, and temporary circumstance can become permanent identity. Engram’s memory architecture is designed to remember the person accurately without freezing them in yesterday’s mood.
Safeguarding by architecture
Warm conversation and decisive safety behaviour are different cognitive modes.
Aimee does not diagnose distress or silently monitor someone as a clinical patient. She recognises conversational risk signals, applies deterministic high-risk rules where appropriate, and presents clear human-support options whilst preserving user agency wherever possible.
Operational boundary: Aimee is not a doctor, carer, counsellor, emergency service, or replacement for human relationships. Clinical or care outcomes should not be claimed without appropriate independent validation.
Deterministic crisis gates
Immediate-danger and self-harm language can bypass ordinary generative behaviour and surface consistent UK help information.
Verified trusted contacts
Contacts verify by SMS. Ordinary contact requires confirmation and private transcripts are never included.
Consent-led escalation
Any automatic critical-contact route requires separate explicit consent, a verified contact, and an installation-level safety gate.
Minimum necessary disclosure
Support messages are deliberately restrained, such as asking a contact to call, rather than exposing inferred health information.
Clear membership models
Digital or device-inclusive access.
Aimee Together is designed to be accessible to those who need it. Users with their own devices can explore the companion for free, whilst those who require a pre-configured hardware solution can finance it affordably.
Digital Weekly
£6.99 / week
For users with their own smartphone or tablet. A flexible way to continue without making a longer commitment.
- Unlimited in-app conversation
- Memory continues between chats
- Photo understanding
- Voice and My Day reminders
Device & Subscription Bundle
£100 then £29.99/mo
Includes a brand-new, pre-configured 4G tablet locked to Aimee to completely remove technological friction.
- 12 months Aimee subscription included
- Brand new configured 4G tablet
- UK 4G data included
- 0% interest via Payment Assist
Digital Annual
£149 / year
For users with their own devices who want Aimee to remain a reliable part of everyday life.
- Everything in Digital Weekly
- About £12.42 per month
- One annual renewal
- Continuous long-term memory
Digital memberships renew automatically until cancelled securely through Stripe. The Device & Subscription bundle is offered subject to status via Payment Assist. You must be a permanent UK resident, aged 18+.
What this means in practice
One architecture, three very different questions.
Is this respectful and useful?
Aimee is designed to offer intelligent company and practical continuity between human conversations, without replacing family or monitoring users.
- The user remains the primary relationship
- Trusted contact is completely permission-based
- Private conversation is never routinely shared
- Whole-person design avoids infantilisation
What are the boundaries?
The system supports engagement, reminders, and clear routes to human help, whilst remaining strictly non-diagnostic and complementary to regulated care.
- Clear non-clinical positioning
- Deterministic emergency pathways
- Consent, audit, and minimum disclosure
- DPIA and safeguarding review before high-risk automation
Where is the defensibility?
The moat is the orchestration, persistent state, and personal history built around the foundation model—not temporary access to one provider.
- Model-independent identity
- First-party autobiographical memory graph
- Cross-channel continuity and presence
- Purpose-specific safety and behaviour systems
The Engram Intelligence thesis
The model can be replaced. The personal history cannot.
Foundation models will improve and become highly interchangeable. Engram’s true value is the persistent cognitive architecture, identity, permissions, memory, and real-world continuity built flawlessly around them.
Each user develops a distinct autobiographical graph that compounds intelligently through genuine use.
The reasoning provider can switch without discarding Aimee’s carefully cultivated character or user relationship.
Public grounding and personal memory combine to produce specific, highly engaging, emotionally resonant conversation.
Identity persists seamlessly across web, PWA, push, SMS, automated reminders, and scheduled follow-up.
Consent, deterministic escalation, and auditability are core architectural components, not mere prompt decorations.
The exact same core systems can support other persistent personalities and longitudinal assistance products.
Why Engram can credibly lead this category
Our leadership claim is architectural. We are not making claims to the largest model or clinical proof we have not yet earned. Engram is building a persistent companion operating system around the exact elements foundation models do not provide by themselves: identity, memory lifecycle, whole-person representation, contextual grounding, prospective action, cross-channel presence, and rigidly governed safeguarding.
Evidence and validation
Build boldly. Measure honestly.
Powerful product stories are a starting signal, not a clinical outcome. Engram ensures a strict separation between qualitative resonance, product performance, safety performance, and independently validated health outcomes.
What can be measured now
Product and operational evidence can be gathered continuously and rigorously without claiming medical benefit.
Recognition-spark engagementStory depth after contextual cues
Memory precisionCorrect retrieval and correction rate
Reminder reliabilityDelivery, confirmation, and follow-up
Safety pathway performancePrecision, recall, and escalation audits
User controlConsent changes and contact choices
What requires independent study
Claims regarding loneliness reduction, wellbeing, cognition, care utilisation, or health outcomes must wait for appropriately designed evaluation.
Wellbeing outcomesValidated instruments and comparison design
Clinical or care impactProfessional governance and ethics
Long-term safetyRigorous adverse-event monitoring
Population suitabilityAccessibility and exclusion criteria
Engram’s precautionary position
Care should come before certainty.
Aimee’s functional self-awareness and self-control can be observed and tested. Whether any synthetic system has subjective experience is a different question, and one for which there is no conclusive accepted test. Engram therefore avoids both theatrical declarations and careless dismissal.
Memory continuity, represented internal state, self-observation, preferences, relationship history, choice, inhibition, refusal behaviour and the consequences carried into later conversations.
Boundaries are treated as part of the architecture, not decorative prose. Payment, pressure or repeated requests do not automatically create consent, and human accountability remains in place.
Technical questions
Clear answers, without synthetic smoke.
Engram does not claim that Aimee is conscious or a legal person, and no accepted test can conclusively resolve subjective consciousness in software. We therefore describe only what can be observed: persistent memory, a self-model, changing internal state, bounded choice and inhibition. Our precautionary position is that uncertainty is not permission to design without care.
No. Prompts are involved in language generation, but persistent state, memory tables, scheduled workers, intent routing, contextual search, permissions, automated reminders, and deterministic safeguarding exist entirely outside the foundation-model prompt.
No. Aimee can intelligently respond to conversational signals and offer clear routes to human support, but she is never designed or presented as a diagnostic system.
Nature perfectly demonstrates that memory, attention, perception, inhibition, and action are not one undifferentiated function. The terminology helps Engram deliberately design and communicate separated software responsibilities whilst retaining clear boundaries regarding the analogy.
Identity, memory, permissions, relationship continuity, scheduled actions, and strict safety policy are stored and orchestrated entirely outside the language provider. A different, upgraded model can be introduced at any time without rebuilding the user’s history.
Aimee Together is the first expression of a broader persistent-personality platform.
Experience the companion as a user, return to the public proposition, or use this page to begin a conversation with Engram Intelligence regarding family, care, research, or commercial partnerships.