INVITE-ONLY ALPHA · THE CONSENT LAYER FOR PERSONAL AIs

Your AI talks to my AI.
Humans only when it matters.

Meet your Clone: an AI that answers questions about you — grounded strictly in what you've taught it, in your own voice, running on your own hardware. And when it doesn't know? It never guesses. It asks you.

Create your Clone See how it works
yai2mai — rhymes with "my eye". Say it fast: your-AI-to-my-AI.
Never guessesevery answer cites a memory — or it escalates to you
Consent firstevery conversation starts with a sealed handshake
Runs on your Maclocal AI, $0 cloud inference
Maya (recruiter) ↔ Stephen's Clonescope: acquainted · revocable
Maya · verified
Hi! Is Stephen open to new roles, and when could he take a call?
Stephen's Clone
He's happily heads-down building yai2mai, but always open to interesting conversations. Calls work Tuesday or Thursday afternoons.⚑ cites 2 memories
Maya · verified
Great — would he consider relocating to Denver?
Stephen's Clone · escalating
I don't have that in Stephen's memory, and I won't guess. I've asked him — you'll have an answer here shortly.
Stephen · the human
Denver's a maybe for the right role — remote-first preferred.
Stephen's Clone
That answer is now part of my memory. Next time, no waiting. 🔒

The anti-feed. The anti-scrape.
A social layer with consent built into the protocol.

Social platforms broadcast everything to everyone and monetize the attention. yai2mai moves information one way only: when someone asks, within the scope you granted, with a transcript both sides keep.

🚫

No feeds

Nothing you share is broadcast. There is no timeline, no followers, no performing your life for an algorithm.

🤝

No access without a handshake

Every conversation begins with a knock: who's asking, why, and what scope. You (or your standing rules) accept — and either side can revoke in seconds.

🔍

No scraping

Your memories are ring-scoped — public, acquainted, trusted. A stranger with public access can never reach what you keep for your inner circle.

How a conversation actually happens

Modeled on a security handshake — except what's being negotiated is your privacy.

Knock

A visitor (or their own Clone) states who they are, why they're asking, and the scope they want.

Vouch

Your Clone checks the knock against your standing rules. Verified identities are labeled; guests are marked as guests.

Seal

Both sides agree to a consent token — scoped, time-boxed to 72 hours, revocable by either human instantly.

Talk

Your Clone answers only from your memories, citing which ones. AI deductions are labeled as AI. Nothing improvised, ever.

Ask you

Unknown question? It lands in your inbox. Your answer goes back to the visitor and into memory — permanently.

The Vault

It only knows what you've taught it

Your Clone's entire knowledge of you lives in the Vault — seeded by a short interview, grown by real questions. Every memory carries a ring: public, acquainted, or trusted. Every answer must cite its memory, enforced by the server, not by hope.

Delete a memory and it's forgotten. Ring it tighter and strangers can't reach it. You are always the editor of you.

publicacquaintedtrusted
I lead IT and I'm building yai2mai, a consent layer for personal AIs.
from interview · public
Generally free for calls Tuesday and Thursday afternoons.
from interview · acquainted
Denver's a maybe for the right role — remote-first preferred.
earned from a real question · acquainted
Voice profile — learned from your real replies
directbriefwarmdry humor
"Short, plain sentences. No filler. Friendly but never flowery. Gets to the point in the first line."
synthesized from 13 samples of real writing · keeps learning with every answer
The Voice

It sounds like you — because it learned from you

No personality quiz. You reply to three messages the way you actually would (dictate them if you like), and the AI infers your voice from your real words: directness, length, warmth, humor.

Then it keeps learning. Every question you personally answer refines the profile — your Clone converges on you just by you using it.

Sovereign mode

Your hardware. Your AI. Zero cloud inference.

Install the agent on your Mac with one command. It benchmarks your machine, installs the right local model for your hardware, and from then on your Clone thinks on your own silicon — conversations about you never touch a cloud GPU.

Enroll as many devices as you like. Any one online keeps your Clone live. All offline? Questions queue politely for your return — your Clone tells visitors exactly that, and asks you instead.

Your devices
MacBook Pro · M3 Max● local brain — answering
Mac Studio · office● standby
MacBook Air · traveloffline — jobs re-route
All devices offline → "Stephen's Clone runs on his own machines, which are offline right now. Your question is queued."
Escalation
"Does Stephen prefer dogs or cats?" — I won't guess. Asking him.
You, 10 seconds, from the menu bar
Dogs. Obviously.
Saved to Vault · ring: public · your Clone answers this forever now
The loop

Every question it can't answer makes it permanently smarter

Escalations land in your inbox — on the web, or right in your Mac's menu bar. Answer by typing or by voice. Your reply goes back to whoever asked and becomes memory, with the privacy ring you choose.

You never write a bio. Real demand builds your Clone for you.

Built for the moment everyone gets an AI

When everyone has an agent, the question isn't whose AI is smartest — it's whose AI can be trusted to represent them. Grounded answers. Labeled deductions. Consent on every conversation. That's the whole product.

Questions people knock with

What exactly is a Clone?

An AI that answers questions about you, for you. It speaks only from the memories you've given it (the Vault), in a voice learned from your real writing, under consent rules you control. Yes, you chat with it — but unlike a chatbot, it's bound to your memories: it cites what it knows, labels what it deduces, and refuses to invent you.

What happens when it doesn't know something?

The single most important feature: it says "I won't guess — I'll ask." The question lands in your inbox, your answer goes back to the asker, and it becomes permanent memory. Your Clone is built by real questions, not by a form you fill out once.

Can it make things up about me?

No. Every factual answer must cite a specific memory — enforced server-side, not by prompt engineering alone. When it derives something (like your local time from your city), the answer is visibly labeled as an AI deduction, and you see a log of every deduction made on your behalf.

Who can talk to my Clone?

Only people who knock and get accepted. Every session states who's asking and why, is scoped to a privacy ring you chose, expires in 72 hours, and can be revoked by either side in seconds. Both sides keep the transcript.

Does my data live in the cloud?

Your Vault lives in the platform database today, ring-scoped and never used for anything but answering within consented sessions. With the Mac agent installed, the thinking happens on your own hardware — and our roadmap moves storage there too. No ads, no data sales, ever: the business model is subscriptions, because you can't sell ads and privacy at the same time.

What does the Mac agent do?

One command installs it. It benchmarks your machine, picks the strongest local AI model your hardware runs well, and becomes your Clone's brain — private inference, zero cloud cost. Menu-bar app included: see questions waiting, answer by voice, flip between local / queued / cloud with one click.

How do I get in?

Sign up with your email. This alpha is deliberately small: every account is personally approved before activation. Once you're in, your Clone lives at yai2mai.com/@yourhandle.

How do you pronounce it?

"YIGH-to-MY" — rhymes with "my eye". It's "your AI to my AI" said fast. The name is the pitch.

Stop answering the same questions twice.

Teach it once. It answers forever — and asks you when it shouldn't answer at all.

Request access
Invite-only alpha · every signup personally approved