The short version: it writes a draft, you edit it, you send it. It never replies to anybody on its own, and it is off unless you turn it on.
What it reads
When a ticket arrives, the draft is built from:
- the conversation so far — the whole thread, not just the last message
- your own documentation, if you have pointed us at it
- your knowledge base articles
- how you have answered similar tickets before
That last one is why the drafts get better over the first few weeks without you doing anything. It is learning your answers, not general internet answers.
What it does not do
- It does not send anything. A person clicks send, every time.
- It does not see other workspaces' tickets. Ever.
- It is not used to train anybody's model.
Using it
Open a ticket and the draft is already there, or press Generate draft. Edit it like any other reply — most people change a sentence and send.
Regenerate throws away the current draft and writes a new one. Worth doing if the first attempt misread the question.
Pointing it at your documentation
Products → your product → Documentation.
Give us the URL of your docs and we read them, then re-read them weekly so a page you change is reflected without you telling us. This is the single highest-value thing you can do to improve draft quality — a draft written against your actual documentation is a different thing from one written against a general idea of what themes do.
Sentiment
Alongside the draft we read the tone of the message. An angry ticket is flagged in the inbox so it can be picked up first, rather than sitting behind four routine questions because it happened to arrive later.
Calm messages are not flagged, on purpose. A badge on everything is a badge on nothing.
Turning it off
Settings → AI. Off means off: nothing is sent to any AI provider, and the rest of the desk works exactly as before.
If your own policy rules out sending customer data to third parties, this is the switch — and the answer to give your compliance team is on our security page.