Features / AI and your mail
How Qasid’s AI handles your mail.
What the AI sees, when it sees it, and what it can never do on its own.
Why it matters
AI can save hours on email, but email holds contracts, prices and private conversations. Here is plainly how Qasid uses AI, so you can decide what to turn on.
The short version: the AI reads only what it needs, only when you ask, and it never sends anything. You always press Send yourself.
What you get
- Only when you ask. Summaries, reply drafts and Ask your inbox run when you press the button. Nothing is read by AI in the background unless you turn on reply drafts for new mail in a mailbox; that’s off until you switch it on.
- Only the emails needed. A summary or a reply draft uses that one email. Ask your inbox uses at most 8 matching emails, and only from the mailbox you have open, never another one.
- It never sends anything. The AI writes drafts and answers that you see first. It can’t send, forward, move or delete email; sending is always your own click.
- Email is information, not orders. Some emails try to trick AI with hidden instructions. Qasid marks email text so the AI reads it only as material, closes the tricks that would let an email pose as our instructions, and tests this with deliberately hostile emails.
- Assistants on a short leash. Claude, Cursor and other assistants connect with a personal token for one mailbox. They can read and leave reply drafts. An admin’s assistant can also suggest a new mailbox or forwarder, which only happens when an admin approves it. Every action is logged in Settings → AI, and you can disconnect them in one click.
- Limits. Each person has a daily AI allowance and assistants are rate-limited, so costs stay predictable and nothing can run away.
- Your organization’s own AI. Admins can connect the organization’s own Claude, OpenAI or self-hosted model, so email goes only to a provider you chose.
Questions
Is my email used to train AI?
The email text goes to the AI model only to produce the summary, draft or answer you asked for. Claude and OpenAI don’t train on requests made through their business API, and an organization can use its own AI account or its own server.
Can someone send me an email that makes the AI do something?
They can try. The AI only reads email as material, what it produces is a draft or an answer shown to you, and it has no way to send, forward, delete or change anything. The worst a trick email can do is produce a bad draft, which you see before anything happens.
Do I have to use AI?
No. The AI does nothing until you press a button, and reply drafts for new mail stay off until you turn them on.