Writing

Write the body of an email with AI from prompt, an instruction or a few rough notes, in the style of the mail this workspace has sent. Give threadId to write a reply: the messages of that thread are read as context. Give draft, the body written so far, to have prompt change it instead, withSubject true to get a subject too, and send, the send options as they stand (even empty), to let prompt change the recipients, the send time, tracking, the signature or the language; times are read in timeZone, or the time zone of the account. Give from, the address it goes out as, and send.fromOptions to let prompt move it to another of those addresses, and send.encrypt when every recipient can receive encrypted mail, to let prompt turn end-to-end encryption on or off. templates true lets prompt switch the email to one of the workspace's published templates, and files true lets it attach files the workspace holds. Returns the text first, then any subject, send changes, template and files to attach, in the shape sendEmail takes; it saves no draft and sends nothing, and it spends one AI action.

Send a new email. To answer an email, use replyToEmail instead: it works out the recipients, the subject, the threading and the quoted history from the message being answered. from is the address it goes out as, and you must name it: there is no default sender, so a send with from left out is refused rather than given an address you did not choose. The one exception is a draftId with from left out, which keeps the address that draft was written as, and is refused when the draft holds none. A draftId of a saved forward sends that forward: the forwarded email and its attachments go below your message, so the message holds only the words the user adds and never a copy of the original. What you may choose: as the owner, or with a role that reaches every address, any address on a verified domain this workspace owns; otherwise only the addresses you were given to send as. whoAmI says which applies, and listWorkspaceMembers names the addresses and whole domains each person reaches. Set translateTo to write it in the recipient's language: the subject and the body are translated before it goes and what you wrote is kept below the translation, unless includeOriginal is false. NOBODY SEES THAT TRANSLATION FIRST. This sends immediately, so call previewTranslation and show your user the result if they should read it before it leaves. A language this server cannot place is refused rather than sent untranslated. Translating costs three model calls: the body, the subject and one to work out what you wrote in. It translates only. It does not proofread the message and it never answers it. attachments takes file ids from uploadFile or listFiles, replyTo sets where answers go, signature false leaves the address's signature off, signatureId picks one of its signatures instead of the default, and scheduledAt sends it later instead of now.

Answers "what would this actually say in their language, before I send it". Returns the translated subject and the exact body sendEmail would put on the wire, so the person you are working for can read it, and change it, first. Nothing is sent and nothing is stored. To send what they approved, pass the subject and body straight back to sendEmail with translateTo LEFT OUT. Their edits included, that is then word for word what leaves; passing translateTo again would translate a second time and the wording would move. It translates and nothing else: it does not proofread, it does not improve the writing, and it never answers the message. Costs three model calls, the same as sending with translateTo, so preview once rather than per revision.

Every language translateTo accepts, so you never have to guess a code. Pass search to filter by a fragment of a code, an English name or the language's own name. translateTo takes any of those three forms, so this is a way to check yourself rather than a step you must take first. Costs nothing. It is a table, not a model call.

Reply to an email inside its conversation. Use this, never sendEmail, whenever you are asked to reply to, answer or respond to an email. You give the thread, the words to add and whether it is reply or reply all; the server works out the rest from the message being answered, the same way the mail client does: who it goes to, the Re: subject, the In-Reply-To and References headers that keep it in the thread, the quoted history below your text, and the address it goes out from, which is the address of this workspace that received the message. There is no default sender, so a message none of your addresses received is refused unless you name from. Reply (the default) goes to the sender, or to their Reply-To, even when that is a colleague writing from another address of this workspace. For a message sent from here that no other address of this workspace received, it goes to the people it was sent to. Reply all also sends to everyone else on its To and Cc. Your other addresses are always left out and nobody on its Bcc is ever included. It answers the latest message of the thread unless you pass messageId. The message is plain text and is sent exactly as written. When the user handed you a saved draft of this reply, pass its draftId so that draft is deleted once the reply is sent. This sends the reply; it does not save a draft.

Scores a message the way a receiving mailbox would, before it is sent: a spam score, a phishing score and an AI-writing score, each 0 to 100, with what raised each one. Call it on a draft before sendEmail or replyToEmail when the user wants it to land well, then fix what it names. Nothing is sent or stored, it costs nothing and it never calls a model. A low score is a good sign, not a delivery guarantee.

Rewrite a subject or the body of an email and get up to five different versions, as the rewrite menu of the composer does, in the style of the mail this workspace has sent. action is shorten, lengthen, rephrase, formal, casual or custom, and custom needs instruction to say what to change. target is body (the default) or subject. Give threadId when the text is a reply, so the rewrite fits the conversation. Nothing is saved, and it spends one AI action.

Suggest a short subject line for the body of an email, under 100 characters, in the style of the mail this workspace has sent. Nothing is saved, and it spends one AI action.

Up to 3 short replies to the latest message of an email thread, the ones the reading pane offers under it, each with a label of a few words and the reply itself, written in the language of that message and in the voice of this workspace. They are kept until another message arrives, so asking again is free. Start a reply from one with replyToEmail, as it is or adapted. A thread whose latest message needs no reply, because your own reply came last or it is a newsletter, an automated notice, spam or an encrypted message, has none, and so does every thread when the workspace turned replySuggestions off in its settings.

Find the sign-in or verification code in a message you pass in, with the reader that puts a code at the top of a message in Nextmail. Send the whole MIME source as raw, or the subject with html or text. It answers only when it is sure: exactly one code worded as one, so two different codes, or a number that could be an order, a booking or a promotion, give no code. It reads only what you send and cannot tell whether the sender is genuine. For mail already in the mailbox, getThread prints the code itself.