Memory you can edit
Every belief as a card with provenance. Because you told me on Aug 3. Tap to see the message. Tap to forget, and it is gone.
Prefersaisle seats, morning flights
Sourceyou said so, Aug 3
edit · forget
Inviting design partners now
The assistant that texts first, and is right.
A chatbot answers questions. An assistant notices things. Eaiden lives in your messages, reads your email and calendar, watches for what matters, and texts you before you knew to ask. Every unprompted message has to be worth the interruption.
Tuesday 8:41
Check in for your Denver flight opens tomorrow at 9. Want me to handle it? Aisle seat like last time.
yes. and can you find me a flight back thursday night
Found the 7pm and the 9pm.
1) 7pm, aisle seats, $340
2) 9pm, $280
Reply 1 or 2, or tell me something else.
1
Booked the 7pm, aisle, seat 14C. Confirmation is in your email and filed in the app. $340 on the Amex, under your travel cap.
1considered reply, even when you send four texts in a row
90%floor on proactive precision. A watcher below it stops interrupting and moves to the morning brief
0settings the model can touch. Money and autonomy live in the app, not the prompt
How it works, in three beats
Everything important can be done by texting. The app explains the thread; it never competes with it.
-
It lives in your messages
The app thread today; SMS and Telegram are built and switch on as those rails are provisioned. No new app to remember to open. When you text, the first reply lands in tens of seconds, the way texting a competent friend feels. Short messages, matched to how you write, willing to disagree.
-
It notices things
Email, calendar, and what matters. A flight with check in tomorrow. A subscription that started charging you. The email that has waited three days for a reply. Each ping draws from a hard interruption budget, so silence is the default and a text is a claim that it was worth it.
-
It asks before it acts
Autonomy is earned one rung at a time, per capability. Rung 0 reads and briefs. Rung 1 drafts and asks, showing you the exact email or purchase before anything leaves. Rung 2 acts and reports. Rung 3 runs the routines you named. You move any capability up or down, and the model cannot.
- Rung 0Read and briefSummaries, briefings, watching. No writes anywhere.
- Rung 1Draft and askComposes the artifact, shows it verbatim, waits for your reply.
- Rung 2Act and reportRoutine sends and bookings within a named budget. Anything unusual becomes a question.
- Rung 3Standing autonomyAlways check me in. Each routine carries its own cap and monthly summary.
- NeverHard gatesNew payees, anything over your cap, deleting data, changing the ladder itself.
The glass box
The app is not a second chat window. It is everything the assistant knows, every action it took, every automation, every pending decision. Visible and editable.
An activity feed of everything it did
What woke it, what it read, what it did, what it cost. The feed renders exactly what the agent read. What you see is what it saw.
Read 3 emails from United0.4c
Read 2 calendar events for Thursday2.1c
Asked you: 7pm or 9pmheld
Automations with kill switches
Standing watchers and schedules as cards. What fires it, what it does, last run, next run, and a switch that stops it for good.
Morning briefdaily 7:30
Watch for flight changeson email
Price drop on the sofahourly
Decisions as texts
One question, numbered options with consequences, enough context to decide without opening anything. Spending money always waits for you. Nothing is lost if a text scrolls away; the same queue lives in the app.
17pm, aisle seats, $340
29pm, $280
blocking · waits for you
Pricing
These are the prices the product will launch at. Nothing can be bought yet: design partners run on the free tier while the paid tiers are built. One standing rule: we never retroactively paywall a habit.
Free
$0
- Reactive chat
- The daily brief
- One watcher
- One taste of rung 1: a single held draft flow
- A monthly pool of 60 full runs, the pool every account runs on today
Lite models only, no browser. It demonstrates proactivity rather than subsidizing it.
Planned
Pro
$29 / mo
- Full proactivity
- Integrations and memory
- Browser actions with your own logins later
- A monthly pool of 400 full runs, with a heads up at 80 percent
At the cap it finishes the task in flight, then asks once: continue at cost, or hold deep work until the reset. Never a dead stop mid task.
Join the waitlistLater
Concierge
$149 / mo
- Heavy automation
- Priority execution
- A larger pool
- Human fallback for the physical world
Task scoped operators at cost. Operators never see your chat history or memory.
Privacy is architecture, not prompt text
The prompt will leak. Assume it already has. So the walls are built where a prompt cannot reach them.
Your data never trains anything
Email, calendar, and connected accounts are read for the task and retained no longer than the task needs. Not used for training, ever.
Delete everything in one action
One control in the app removes memory, history, artifacts, and tokens. Deletion is real deletion.
The model never controls money or autonomy
Spend caps, the trust ladder, and payees are set only in the app and enforced at the tool boundary. A fully compromised model still cannot move them.
Get your number
Design partners first, then the public waitlist. Every friend who joins with your link moves you up three places. The plan is that every user gets an assistant with a phone number, so forwarding that number is the demo.
You are in line.
Share this link. Each friend who joins with it moves you up three places.
You were already on the list, and we kept your spot. Your share link was shown once, when you first joined.