You're about to hand an assistant your passwords, your inbox, your books. Let's talk about the scary part.
That worry is the right instinct, and you shouldn't hire anyone, human or not, from someone who waves it off. So here's exactly how your assistant is kept on a leash: what it can touch, what it can never do, and who answers for it. In plain English.
The block lives in the code itself. It can line a payment up, but money only moves after your yes. Deleting for good, it can't do at all.
Your assistant works for you alone. Nothing it learns is shared with any other customer. It's in the policy, in writing.
Ask "what did you do today" and get the full list, in plain English, any time.
No ticket queue, no offshore support maze. Something feels off, you call David.
Every layer, in plain English.
The plain version up top. The details underneath for whoever in your life reads the fine print.
It only gets the keys you grant, and you hold the off switch
Your assistant works inside the accounts you choose to hand it, and nothing else. It can't wander. And stopping it never requires a ticket: say stop, and it stands down. Want an account cut off at the root? Password logins die the moment you change that password, and connected sign-ins get revoked while you watch. The keys were always yours.
- access granted account by account, agreed with you at setup
- scoped to the narrowest permission each program offers
- a stop text freezes work on the spot; the human-verified shutdown is finished the same day
- password logins: change the password and that account is cut off instantly
- sign-in connections (Google, Microsoft and similar): a password change alone doesn't always kill an active connection, so David revokes it and confirms it's dead
Passwords live in a vault, not in the AI
Your logins are stored in a professional password vault. The assistant borrows a key at the moment it works. It doesn't memorize your passwords, and they never sit in its head.
- dedicated encrypted vault, separate per customer
- credentials are never pasted into AI prompts or its memory
- vault access is itself logged
Big moves stop and wait for you
Spending money, deleting things, anything you'd wince at: the assistant physically can't do it alone. It stops, asks you, and waits. That rule lives in the code, where the AI can't talk its way around it.
- moving money without your OK: blocked in code, no exceptions
- permanent deletes: blocked in code outright, and no yes unlocks them
- the approval comes to your phone, and only you can give it
Everything it does is on the record
Every email sent, every call made, every file touched: written down where you can read it. No mystery moves. If you ever wonder what it did while you were on a roof, ask.
- a full action log per assistant, readable by you any time
- reviewed by a human, not just stored
- retention spelled out in the privacy policy
The whole crew can use it. The whole crew isn't the boss.
Your team can all talk to it, but it doesn't treat everyone the same. At setup, you decide who can ask for what: who sees the money, who books jobs, who it takes orders from. The sensitive stuff goes only to the people you name.
- per-person rules, written into the setup with you
- sensitive topics like payroll and the owner's inbox limited to named people
- requests outside someone's lane get declined and flagged to you
Smooth talkers get nowhere much
Your assistant answers phones and reads email, so strangers talk to it all day. The design assumes some of them will try something: a caller can't change its rules, because the rules don't live in the conversation. It can't hand over what it was never given. And the big-move stops hold no matter how convincing anyone sounds.
- what callers and emails say is treated as a request, never as a new rule
- a stranger's claims get logged and checked, not written over your records
- odd or pushy requests get flagged to a human
- and per the straight talk below: nobody honest claims immunity, so the layers above assume tricks will sometimes land
Your business never mixes with anyone else's
Your assistant is set up for you, works for you, and learns you alone. Nothing crosses over to another customer. Not your data, not your prices, not your habits. Ever.
- separate setup, accounts, phone number, and memory per customer
- nothing you share is used to train anyone else's assistant
- the commitment is written into our Privacy Policy
The heavy lifting runs on certified rails
The AI itself, the phone lines, the hosting: all run on major providers that pass the audited security certifications big companies demand, the same rails half the internet runs on. We don't home-brew the hard parts.
- AI, telephony, and hosting vendors hold SOC 2 security audits
- accounts protected with two-factor sign-in
- the full vendor list is public in our Privacy Policy
Now compare it to the sticky note.
Here's the part nobody says out loud: at most small businesses, "security" today is one shared password, written on a note by the register. An assistant done right doesn't lower your guard. It raises it.
Straight talk: what nobody can promise
No honest company will promise you an AI can never be fooled or never make a mistake. Anyone who promises that is selling something. What we promise instead: your assistant holds the fewest keys it needs, the dangerous moves are physically blocked without your OK, everything it does is on the record, and a real person reads that record. And when it does get something wrong, it's fixed the same day by the person who built it, and the fix sticks. That's more protection than most businesses have ever had.
Still worried? Good.
That's the instinct of someone who built something worth protecting. Bring the hard questions. David answers them himself, no script.
Book a call15 minutes · your okay before big moves · one human accountable
David builds each assistant himself, a few at a time. Spots open monthly.