What Frank does with a message
Every message that arrives gets exactly one of four outcomes. Nothing falls between them.
Escalate
Important, sensitive, or uncertain. It goes to you, flagged, and Frank does not touch it.
Draft
A reply is written in your voice and waits in your queue. You read it, edit it if you want, and send it.
Organize
Labelled, filed, or routed. No reply needed, and none is written.
Auto-send
Sent on its own — but only inside a narrow situation you turned on yourself, and only after twenty-nine separate checks pass.
The part that makes it different
Plenty of software will write an email for you. That part is commodity now. The question nobody answers well is who decides whether it should be sent at all.
Frank does not ask an AI model whether it is safe to reply. That decision is made by a fixed, deterministic set of rules — before the model sees your mail, and again after it writes.
Mail touching legal matters, contracts, disputes, HR, compensation, payments, invoices, banking, tax, insurance, health, credentials, security, regulated advice or complaints is never sent to a model provider at all. It goes straight to you. That is a list, not a confidence score, and there is no setting that turns it off.
Automatic sending is off when you sign up and stays off until you enable it, one narrow situation at a time. Every automatic send has a two-minute undo window, and there is a kill switch in the header of every page that stops all automation across your account instantly.
How Frank learns your voice
When you connect a mailbox, Frank reads your sent mail once and measures how you actually write — your typical reply length, how often you use contractions, how you open and close a message, the phrases you reuse.
What it keeps is the measurements, not the mail. The sent messages used to build the profile are not retained afterwards. Your profile is stored against your mailbox alone, is never shared with another customer, and is never pooled with anyone else's writing.
Your mail is not used to train any model. Not ours, not anyone's. That is a contractual commitment under Google and Microsoft policy, and it is set out in full in our privacy policy.
What access Frank asks for
The narrowest access that makes the product work, requested in stages.
| Permission | What it is used for | When it is requested |
|---|---|---|
gmail.readonly | Read incoming mail so Frank can decide what it is and draft a reply. Read sent mail once to learn your writing style. | At sign-up |
gmail.compose | Create draft replies in your mailbox for you to review. | At sign-up |
gmail.labels | Organize mail with labels. | At sign-up |
gmail.send | Send the specific replies you have authorised. | Only if you enable automatic replies. If you never do, Frank never holds this permission. |
For Microsoft 365 we request Mail.ReadWrite, User.Read and offline_access — plus Mail.Send only when you enable automatic replies. Your organisation's administrator must approve Frank before any mailbox can connect. Frank accesses only the mailbox of the person who signed in and cannot read other mailboxes in your organisation.
You can revoke access at any time, from your Frank settings or directly in your Google or Microsoft account. Disconnecting revokes our access, destroys the stored tokens immediately, and stops all processing.
Where Frank is right now
Frank is in active development and is not open for general sign-up. We are working with a small number of design partners and going through Google and Microsoft verification before anyone's real mail is connected.
If email is eating your week and you would be willing to spend fifteen minutes telling us how, we would genuinely like to hear from you — whether or not you ever become a customer.