A mobile app that lets a person capture memories — stories, voice notes, journals — and chat with a Persona that speaks in their tone, remembers what they told it, and can be shared with family through a revocable invite link.
A chatbot without a history is generic. A chatbot trained on your public posts is a stranger with your haircut. Neither is what people actually want when they say they’d like to leave something behind.
The design brief was to build the opposite: a Persona seeded only by what the user chooses to tell it, tunable on day one, and shareable only by explicit invite.
Onboarding asks six deliberate questions. Every chat request ships those answers plus a window of the user’s own memories — so the Persona’s tone is set by minute one and its content is always grounded in something the user actually said.
Instead of exposing a prompt-engineering surface, onboarding asks the same questions a biographer would. Answers are stored alongside the account and prepended to every chat as lightweight system context. The Persona’s voice matches the first screen the user saw — not a generic template.
Settings can rewrite any answer later, and an optional long-form prompt field lets power users dictate a custom persona description.
Memories are first-class. Users add them by text, dictation (Whisper-grade transcription at 16 kHz mono), or imported journals. Each memory carries metadata — topic, date, priority — so the backend can surface the most relevant slice when a chat happens.
The client is offline-aware: NetInfo short-circuits any request when the device is offline, and AsyncStorage caches the last known state so the UI never crashes on a flight.
POST /chats/:id/message
{
"text": "What do you remember about grandma's house?",
"context": {
"persona": { reflect: "warmth", tone: "thoughtful", ... },
"memories": [
{ id: "m_01", topic: "childhood", date: "1994-07", text: "..." },
{ id: "m_12", topic: "family", date: "2001-12", text: "..." },
{ id: "m_23", topic: "places", date: "2015-04", text: "..." }
]
}
}
→ 200 OK reply grounded in user-owned facts
→ offline short-circuits before the request even leaves the device
Most AI products look like cold chrome. Clonara is the opposite: cream paper, warm brown ink, a single golden accent. Typography pairs a technical sans (Euclid Circular A) with an editorial serif (Canela Text) — utility for forms, intimacy for stories.
“The interesting work wasn’t the model. It was deciding which six questions the user should answer before the Persona ever replies — and trusting that those six answers are the product.”