Document 1 of 3 · Product

Zarela — Product Definition

What Zarela is, who it is for, the problem it solves, its pillars, its surfaces, and what is deliberately in and out of scope. The shared reference for product, design, business, and leadership.

Audience: CEO · Product · Design · Business Status: Draft v0 Read before: Technical & Delivery

Purpose

Define the Zarela product clearly enough that any stakeholder shares the same mental model of what we are building and why.

Audience

Readable by non-technical stakeholders: CEO, product, design, and business teams.

Scope note

Product-focused. No low-level infrastructure detail — that lives in the Technical Architecture.

01Executive Summary top

Zarela is a chat-first and voice-first AI companion for autistic adults (18+) with borderline intellectual functioning. Rather than being another app to learn, Zarela joins the family's shared chat like a new, trusted member — a calm, patient presence beside the people the person already knows. The person can type or speak in their own language from day one.

The product rests on three pillars: companionship (a companion that never tires and gently reduces isolation), care (consent-first features that look out for the person through the day), and learning (everyday tasks turned into calm, guided steps that build lasting skills). A defining idea is fading prompts: once a skill is learned, the reminders quietly recede while watchful care continues. The watchful care never switches off; only unnecessary prompting fades.

The emotional core is the future-care anxiety of aging parents and caregivers — “what happens to my child when I'm gone?” Zarela is positioned as part of the answer: building independence and continuing to support the person. It is always a complement to human caregivers and therapy, never a replacement. Because the people it serves are vulnerable, Zarela is a YMYL (Your Money or Your Life) product where safety, consent, privacy (GDPR), and trust are first-class concerns.

The one-paragraph version

Zarela is a calm, adult, multilingual AI companion that lives inside a familiar chat, supports an autistic adult's companionship, daily care, and learning, keeps a trusted family circle gently in the loop with consent, and grows the person's independence over time. It works fully on its own; a rehabilitation center can optionally be connected on top.

02What is Zarela? top

Zarela is a chat-first and voice-first AI companion for autistic adults (18+) with borderline intellectual functioning. Instead of a separate app the person must learn, Zarela appears as a new member in the family's shared chat — a calm presence in the chat list alongside trusted people. The person can type or speak in their own language, and a system-wide communication layer lets every participant use their own language (family, caregivers, and optionally a center's staff); the language complexity is handled behind the scenes.

The three pillars

Companionship

A patient, always-available companion that never tires and gently reduces isolation.

Care

Consent-first features that look out for the person through the day — distress, scams, location risk, missed routines, escalation.

Learning

Everyday tasks turned into calm, guided steps that build real, lasting skills at the person's own pace, with fading prompts.

What problem does it solve?

03Target Users and Stakeholders top

StakeholderWho they areWhat they need from Zarela
The person (“loved one”)An autistic adult (18+) with borderline intellectual functioning. Uses a managed account.A simple, familiar, calm companion; help understanding the world; safety without feeling watched; growing independence.
Responsible Support PersonA parent, primary family supporter, or legal/responsible support figure who creates the initial support environment.Confidence the person is supported and safe; a way to set up and gently oversee without controlling forever.
Family membersRelatives in the trusted circle.To stay connected and in the loop, with consent.
CaregiversAdded for ongoing support, routines, companionship, or escalation. A role, not a default separate app.Permission-scoped access to help with daily support and escalation.
Emergency contactsPeople surfaced in limited, time-critical flows.To be reachable when escalation requires it.
Rehabilitation center staff (optional)A professional caregiver on a center-owned account, connected only by family consent.To observe, teach, and guide the AI for a case, with continuity across staff changes.
Zarela company / AdminInternal operators, content, support, and moderation roles.To run, moderate, and support the platform safely — an internal surface, not a normal user app.

The managed-account model

The person never self-registers. A Responsible Support Person creates the initial support environment so every person has a responsible adult and a trusted circle around them. This is an early-stage setup model, not permanent control — and it sharply lowers the risk of impostors entering the user pool, which matters greatly for the future Match-maker feature.

04Product Principles top

These principles are binding. Every surface and feature decision should be checkable against them.

Chat-first

The experience lives inside a familiar chat. Zarela is a contact, not a dashboard. Intelligence stays behind the scenes; the surface stays calm and predictable.

Voice-first from day one

Voice is a core accessibility feature, not a later enhancement. Many autistic adults find typing slow, tiring, or impossible, so the person can speak naturally and hear a reply.

Adult tone, never childish

The voice is calm, warm, respectful, and adult. Normal adult curiosity — relationships, attraction, sexuality, emotions — is handled safely and without shame, never infantilized.

Consent-first

No monitored capability or peer connection runs without a confirmed, current consent record. Consent is versioned and revocable.

Not a surveillance tool

The person should feel supported, not watched or controlled. Family answers come from consent-bound summaries, never unrestricted access to private chats.

Autonomy-building, not permanent control

Progressive Autonomy means building independence over time. Reminders and restrictions fade as readiness grows; only watchful safety support remains.

05Product Surfaces / Applications top

Zarela is delivered as separate apps/surfaces over one shared identity. Caregiver is a role inside the Family & Support App, not a separate default app. The center surface is optional.

SurfaceWho uses itRole in the product
Person AppThe autistic adult (managed account)The companion side. Extremely simple and familiar; chat-first and voice-first.
Family & Support AppResponsible Support Person, family, caregivers (per permission), emergency contacts, optional center staff (by consent)The support side. Chat-first; heavy controls live inside categorized Settings.
Specialist / Center App OPTIONALA center's professional caregiver, on a center-owned accountOptional professional-support layer, active only when a family connects a center by consent.
Admin PanelInternal company roles (Super Admin, Operator, Content, Marketing, Support, custom)Internal company surface for operations, content, support, and Room/moderation management. Not a normal user app.
Website / Marketing SurfacePublic visitors, prospective families, rehabilitation centersDiscovery, SEO/AEO content, and waitlist capture. Separate from the core app experience.

06Person App Definition top

The companion side of Zarela. Its guiding design principle is extremely simple and familiar — it should feel like a chat app and a feed the person already knows, so the training cost is near zero. It must not feel like a care dashboard, clinical tool, learning platform, or admin surface.

Four areas only

The Person App has exactly four simple areas: Chat, Today, Timeline, and Help. There is deliberately no separate “Dictionary” tab — explanation is part of Zarela's character, not a tool.

Chat

Voice (required from day one)

Today

A simple “what matters today” area — not a complex calendar. Contains medication reminders, appointments, center visits, daily routines, simple step-by-step tasks, and gentle wellbeing check-ins, with simple actions like Done, Remind me later, Help me do this, Ask my caregiver. Zarela can guide a task step by step at first; as the person becomes more independent, reminders become quieter and less frequent while safety checking continues in the background (fading prompts).

Timeline (safe interest feed)

A safer, family-trusted alternative to open algorithmic feeds: a curated, interest-based feed, not an open social network. Content is based on the person's interests with Responsible Support Person / permitted caregiver safety settings and safety filtering, in age-appropriate adult framing. If a piece of content raises a question or strong emotion, the person can ask Zarela to explain it calmly. It deliberately has no open DMs with strangers, public comments, uncontrolled user search, addictive explore engine, uncontrolled external links, ads, or any direct path for strangers to reach the person.

Help

A very simple Help area, always one tap away (“I need help”, “I feel unsafe”, “I am lost”, “Call/message my caregiver”, “Ask Zarela what to do”). Depending on consent and settings, Zarela can guide the person calmly, suggest contacting a caregiver, notify the trusted circle, share location, and escalate distress or safety alerts. Critical feel: this must not feel like surveillance — the person should feel supported, not watched.

Zarela as friend & explainer

This replaces any separate dictionary feature. If the person doesn't understand a word, a message, a joke, a video, a feeling, or a confusing situation, they simply ask their friend Zarela. It feels conversational, never like using a tool. Zarela can explain difficult words, messages from others, jokes and social cues, online safety risks (OTP codes, gift cards, suspicious links, money requests, scams), emotions, and topics like boundaries, privacy, consent, friendship, relationships, and personal safety. Because the person is an adult, Zarela does not infantilize normal adult curiosity, and handles sensitive topics safely, respectfully, and without shame — gently encouraging involvement of the trusted circle when a situation suggests risk.

Progressive Autonomy

An extension of fading prompts: Zarela can gradually reduce communication restrictions as the person gains readiness. The goal is independence, not permanent control; safety support stays active while only unnecessary control fades. Indicative levels:

LevelMeaning
L0Zarela + Responsible Support Person only.
L1Initial contact list created by the Responsible Support Person.
L2Suggested contacts/friends, with support-person approval.
L3Limited friend requests by the person.
L4Freer social use with safety support.
L5More independent use with quiet background support.

Levels are indicative and subject to safety review; the exact unlock criteria are an open decision.

What must not be in the Person App

Early-stage, open user search and open social discovery are off — but can unlock gradually via Progressive Autonomy. The person should never have to manage the safety system.

07Family & Support App Definition top

Chat-first and simple. Its visible surfaces are Chat, Rooms, Match-maker, Overview/Today, and Settings — five surfaces only. Most care, permission, consent, autonomy, routine, caregiver, and center controls live inside categorized Settings so the app never feels like an admin dashboard.

Who uses it & setup roles

Chat (the main surface)

Its own familiar Chat area, as simple as the Person App's. Zarela is always pinned as a family assistant. The family can chat with Zarela directly, with the loved one and trusted circle according to permissions, and with caregivers and permitted contacts. Family-to-family chat from the Match-maker opens inside this Chat surface; important alerts can appear as cards inside Chat. Zarela answers family questions only within consent and permitted summaries (e.g. “How was he today?”, “Why was the food reminder missed?”, “How should I message him?”).

Not a surveillance tool

Zarela answers from consent-bound summaries, never unrestricted private chat access. The app must not become a way to watch the loved one.

Rooms

Official Zarela-managed community spaces for families and support people — community-building, discussion, online events, education, and location/language/topic-based support (by country, city, language, topic, or event type).

Hard rule — Rooms are company-created and company-governed

Families can join or leave official Rooms, but they cannot create Rooms or define Room policies.

Families canFamilies cannot
Browse official Rooms; join (if policy allows); leave; participate per policy; attend official events; receive announcements; report messages or members.Create/edit Rooms; define Room policies; assign moderators; create official events; turn private groups into Rooms; use Rooms to bypass Match-maker; directly access vulnerable adults through Rooms.

Only Zarela Company / Admin can create, edit, archive, pause, or close Rooms, define policies, assign moderators, schedule official events, control visibility, classify by country/city/language/topic, and enforce moderation.

Match-maker

Safe, family-mediated discovery of compatible peers (layered bios, city-level location, both families accept, support-person approval). It is intentionally top-level because it is high-value for families. It must not be available directly to the person in early stages; the person does not browse, search, or initiate. (Full flow appears under Conversation Types.)

Overview / Today

A family-facing summary surface: how the loved one is doing within consent-bound summaries, upcoming routines and reminders, and important alerts surfaced as cards. It reflects, it does not expose raw private chat.

Settings (category-based)

The heavy operational and configuration concepts live here, organized into categories so the app stays chat-first:

CategoryContains
A · Loved One ProfileLanguage, communication level, interests, sensitivities, calming strategies, overwhelm triggers, preferred output mode, independence level.
B · Trusted Circle & ContactsTrusted circle, family members, who can chat with the loved one, who receives updates, emergency contacts, add/remove contact permissions.
C · CaregiversAdd/remove caregiver, caregiver permissions, temporary access, schedule, routine access, alert access.
D · Routines & RemindersMedication, appointments, sleep, food/water, hygiene, therapy/learning tasks, reminder style, escalation if missed, fading prompts.
E · Safety & AlertsDistress, scam/money, location risk, unsafe messages, missed critical routine, unusual silence, escalation chain, push/SMS/call settings.
F · Progressive AutonomyCurrent level, unlocked/locked permissions, Zarela recommendations, approve/reject unlock, rollback if needed.
G · Consent & PrivacyWhat data is used, who sees what, voice/audio retention, location sharing, center sharing, match-maker visibility, data access/erasure.
H · Center ConnectionSelect center, connect/disconnect, assigned specialist, center access permissions, handover history, sensitive-data filtering.
I · Rooms Preferences (not room creation)Joined rooms, leave room, country/city preference, language preference, event notifications, report settings.
J · Match-maker PreferencesOpt-in/out, profile visibility, city-level location, visible fields, request permissions, pause discovery.

Trusted circle & caregiver management

The trusted circle is the private support network around one loved one. Caregiver management (add/remove, permissions, temporary access, schedules) lives inside Settings category C, scoped by permission so that adding a caregiver does not hand over account ownership.

Consent and privacy

Consent and privacy controls live in Settings category G: what data is used, who sees what, voice/audio retention, location sharing, center sharing, Match-maker visibility, and data access/erasure. Consent is the legitimacy layer for every care feature. (Enforcement mechanics are in the Technical Architecture.)

What this app must NOT become

  • Not a surveillance dashboard; not a place to read every private loved-one chat by default.
  • Not a replacement for the loved one's autonomy; not a place where caregivers automatically own the account.
  • Not an open social network; not a user-created community platform; not a way to bypass Match-maker.
  • Not a company admin panel for families; not a clinical dashboard.

08Specialist / Center App Definition OPTIONAL top

Not part of every family's default experience. It exists only when a family chooses to connect with a rehabilitation center and consents to professional access. The center adds professional continuity on top of the core product — it never gates the core.

Optional professional-support layer

Either a center recommends Zarela to a family, or a family selects a center later from inside the app — at onboarding or any time. Once the family consents, the center assigns a staff member to the case (family + person). The family may disconnect or change center per consent, policy, and audit rules.

Center-owned accounts

A center is an organization (tenant). Its specialist accounts are org-owned: the center owns the professional case relationship and staff-assignment records, while access to the loved one's personal context remains permissioned, purpose-limited, and filtered.

Staff assignment and reassignment

A center may manage multiple cases. If it changes the assigned person, it reassigns the case to new staff; the new staff member does not start from zero — the system preserves the previous helper/teacher/coach's relevant work history and continuity context. Every reassignment and access is recorded in the Audit Log.

What assigned staff can do

Center Zarela Assistant

The center has its own Zarela assistant that knows the permitted support context for a case. It helps staff ask questions during handover and continue support smoothly — answering from permitted support context, not unrestricted private history.

Sensitive-data filtering

The system detects sensitive/private topics. If a center or staff member requests data that should not be shared, the system does not provide it and shows a privacy-protection message instead. Access is org-scoped and purpose-limited; consent-backed and audited.

09Conversation Types top

One chat engine hosts five conversation types with different rules. The type determines whether Zarela joins, whether memory is written, and whether care/safety monitors — so ordinary chatting and a supervised, protected interaction coexist on the same product. In early stages the person does not freely search for or initiate peer connections; all matching is support-person-mediated.

TypeMembersZarela presentCare/SafetyNotes
Circle chatPerson + trusted circleYes — companionOnThe core experience.
Standard chatFamily/caregiver ↔ outside contactsNoOffOrdinary person-to-person chat for non-person personas. The person persona cannot do free out-of-circle chat.
Family-to-family chatTwo Responsible Support Persons (after both accept)NoOffOpened by the Match-maker once both families accept; opens before the peer chat.
Supervised peer connectionTwo matched autistic adultsYes — mediatorOn (heightened)Opens only after both Responsible Support Persons approve. Mediator + care active; easy block/report; escalation.
Official Community RoomFamilies, support persons, caregivers (per policy), invited professionals, Zarela moderatorsOptional — room assistantModeration + policyCompany-created only. Must not bypass Match-maker for direct peer connection.

The Match-maker flow (every step Responsible-Support-Person-controlled)

  1. Suggest — in the Family app, the AI shows a list of same-level peers it has matched.
  2. Browse — the family reviews a layered bio: name, photo, city-level location (never address), abilities, daily habits, sensitivities, self-reported challenges. Minimal fields while browsing; full detail only after both accept.
  3. Request → Accept — family A requests; family B accepts. Both Responsible Support Persons must accept.
  4. Family-to-family chat — the two Responsible Support Persons talk first (ordinary chat, no AI).
  5. Approve → Peer chat — only after both approve does the supervised peer connection open, with Zarela as mediator and care active.

Why this is the most sensitive feature

  • A browsable directory of vulnerable adults (photo, city, habits, challenges) is the single most sensitive surface in the system.
  • Mitigations: layered profiles, city-level location only, a separate explicit “may be listed” consent, opt-in default-off, audited and rate-limited browsing (no scraping).
  • No clinical data — only family self-reported descriptions; still high-sensitivity. A DPIA and autism-specialist policy review are required before building.
  • Mediation must be non-definitive (“they may have meant…”) and must never create a recursive loop between two users' Zarelas.

10Core Product Scope for First Release top

The first release is the direct-family core, working end-to-end with no center. It is production-quality for what it includes — not an MVP.

In the first release

  • Direct family onboarding (no center required)
  • Managed person account + trusted circle
  • Person ↔ Zarela chat, voice and text
  • Basic Family & Support chat
  • Consent foundation
  • Basic Help and escalation
  • Audit foundation
  • Companion & explainer behaviour
  • Today / reminders / routines (care basics)

Not in the first release

  • No Match-maker
  • No open social discovery
  • No full Timeline unless safety-reviewed
  • No Center layer required
  • No billing engine (commercial model open)
  • No cross-language family/teacher communication yet

The detailed phase-by-phase sequencing and acceptance criteria are in the Project Management Plan.

11Future Product Scope top

CapabilityWhat it addsStatus
Safe interest TimelineCurated, filtered, interest-based feed as a safer alternative to open feeds.FUTURE (safety-reviewed)
Global communicationEvery participant communicates in their own language — family-to-family, caregiver/coach, online teacher.FUTURE
Rooms / CommunityCompany-created community spaces and events; non-core, company-managed.FUTURE
Optional Center layerSpecialist app, case management, staff reassignment, Center Zarela Assistant.OPTIONAL
Match-makerFamily-mediated peer discovery and supervised peer connections.GATED (DPIA + specialist review)
Billing / subscriptionWhatever commercial model is later chosen.FUTURE (model open)

12Product Boundaries and Non-Goals top

13Open Product Decisions top

Unresolved or risky questions are listed here rather than silently resolved. These need a product owner and a decision.

Open decisions

  • Revenue / commercial model — intentionally open. Direct-to-family, center-funded, sponsorship, grants, and other models are all still possible; nothing is decided.
  • Support person's role during a peer chat — step back once it opens, or stay monitor/available throughout? Unresolved.
  • Center as its own app vs. a professional role in a shared surface — currently modelled as a center-owned professional-caregiver account; final form open.
  • Specialist / center reassignment scope — a reassigned staffer receives permitted continuity context and relevant work history, not unrestricted private chat history; exact boundary of “permitted continuity context” needs definition.
  • Progressive Autonomy unlock criteria — the L0–L5 levels are indicative; concrete, safety-reviewed criteria for each unlock are not yet defined.
  • Timeline launch gating — whether the Timeline ships in the first release depends on a safety review; default is to exclude it until reviewed.

Delivery-level decision gates (provider choices, DPIA timing, production hardening) are tracked in the Project Management Plan.