How the assistant learns to write as you

The assistant does not have a house style. It has yours, and it gets it from the only honest source there is: the mail you have already sent. It reads how you open, how you sign off, how formal you are, the phrases that keep appearing, how long your sentences run, and builds a writing profile plus a bank of your own examples to work from.

The first pass happens shortly after you connect an inbox and takes a few minutes. You do not have to start it. After that it re-reads your recent sent mail about once a day to stay current with how you are writing now, rather than how you wrote in March.

What this means for your mail

To learn from your writing, the assistant has to read it, and the reading is done by the AI. So the sent mail it learns from is sent through our relay to the model, the same way a draft or a summary is. The profile that comes back lives on your computer with your mail. We keep neither: our records of a request are who made it, when, which feature, and what it cost us.

This runs in the background from the moment you install, because an assistant that cannot write like you until you have found a settings page is an assistant nobody ever turns on. If you would rather it waited to be asked, the switch is in the assistant’s settings and turning it off stops it. The same screen tells you when it last learned and from how much mail, so it is automatic without being secret.

How it gets better

In the most direct way possible. Every time it drafts something and you change it before sending, the difference between its version and yours is a lesson, and it folds those lessons back in. If its drafts start drifting away from you, it notices and re-learns from your recent corrections rather than waiting for the next scheduled pass.

You can also teach it directly. Open the assistant and edit the profile by hand: correct the greeting it thinks you use, add a rule, or tell it something about your business it should always remember. Typed rules outrank anything it inferred.

Fact Guard, before every draft reaches you

Every figure, price, quantity, date, phone number, link and email address in a draft is compared against the real conversation and what you have taught it. Anything that cannot be accounted for is flagged in the draft, before sending, every time.

This is a check run by code, not a promise made by the AI, and it exists because an invented number in a live negotiation is the one mistake this product refuses to make. It is also why your agent stops rather than answers when a buyer asks your price.

What it will never do

Send anything. Not a draft, not a follow up, not a reply. Everything it writes waits for you to read it and say yes, and there is no setting that changes that.

Next: Building your first agent