Write or import a few words and watch your twin take shape — right now, no account needed. Create one only when you want to keep it.
Import the twin file from your old device, then log in with your passphrase.
Enter your email, the Recovery Key you saved at signup, and a new passphrase.
Show a count on the app icon when reminders are due — visible even with the app closed. On iPhone this needs your OK once (no notifications are sent).
Share SelfTwin as a card — let friends discover a private twin that's made of your own words and never makes things up.
Back up your twin or move it to a new device. The file is encrypted — only your passphrase unlocks it. No cloud, ever.
Bring in a journal, notes, or a chat export (.txt · .md · .json). Only your own words become part of your twin — on-device; your words are never uploaded.
Choose how it sounds — or record your own. Honest note: picking a voice works for anything it says; cloning your voice to say any words needs a trained voice model (not on-device yet — like frontier fluency, a wall we respect). Recording stores your voice on this device for the phrases it speaks back from your own words.
It can greet you and check in when you open the app — always grounded in your own data, never made up. You decide.
Pro is an annual plan (~$4/mo, billed yearly). Get a key, then paste it below to unlock — verified fully offline, nothing phones home.
Your agent runs entirely in this browser, on this device — your words never leave it, and it works offline. It lives alongside your journal on its own memory (what you teach it) — not fused with your archive.
Write in your own language, any script. It learns each word and keeps its meaning from how you use it — never English, never invented. Tap an underlined word to see its meaning in your own words.
Point your camera at any writing — a journal page, a note, a whiteboard, a number — or pick a photo. It reads the text on your device (nothing leaves your phone), drops it into your entry for you to check, and it becomes part of your book. Seeing here is a way of listening — it reads what you show it, it never judges you. First use unpacks the on-device reader (~17 MB, once); after that it reads offline. Printed text reads best.
Give it your phone's motion sensors (one permission). It watches the real motion on its own and reverse-engineers what it can — gravity when still, your rhythm when you move it — and says plainly when there's no law. Everything it does is journalled below for you to check. (It only reads motion; it can't touch anything else on your phone.)
Hold the phone and move as you naturally do. It applies frequency analysis and motion-physics to learn the signature of how your body moves — your rhythm, your steadiness — then can tell whether new motion resembles you. (It reads the physics of your movement, not your mind — and it can't prove it's only you, just that motion matches the you it learned.)
Rest a fingertip gently over the rear camera, hold still ~15s in good light. It reads your heartbeat from the light through your skin (photoplethysmography). It cannot read your brain — no phone can — but this is a true frequency of your body.
Lie down, rest the phone on your chest, hold still ~30s. From the tiny motions alone it decodes your breathing and heartbeat — both below 20 Hz, sub-perceptual — and only reports a rate it can verify on a held-out half. If the signal is too weak, it honestly declines rather than guessing.
Every verified reading above is saved on your device (declined ones are never stored). Pro turns single readings into your resting-rate and breathing trend over days and weeks — computed on your phone — your readings never leave it.
Type a few things on your mind, or paste recent notes (one per line). The agent infers what you mean — your preoccupation, your intent, the undercurrent — holds it with honest confidence, and asks if it read you right. It learns from your answer. This is its plausible half: it never claims certainty, and you are the judge.
Watch the read become yours: your confirmed "that's me", the words you've confirmed are you, and the corrections I've absorbed. It only exists from your own yes/no — and it grows the more you use it.
The twin looks inward — honestly. It names its own weaknesses (each with the evidence, never flattered) and how it would grow; and from your words above it reflects your patterns back — descriptively, never advice. Two sides of the shared self.
Paste a sequence (15+ numbers). The agent guesses the law and proves it on numbers it never saw.
Paste real measured pairs — x values, then the matching y values. It fits the simplest law, learns from most of them, and tests on the rest. If it holds on points it never saw, it keeps it; if not, it refuses to invent one.
This is the only way back in if you forget your passphrase. There is no email reset and no support recovery. If you lose both your passphrase and this key, your twin is gone forever — encrypted so completely that no one, including us, can ever restore it. Screenshot it or write it down now.