Auto mode: dictation that works out what it is
By Ryleigh Newman · Published · Updated
Auto mode is the YapToText setting that formats each dictation by itself: something shaped like a letter comes out as an email, something short or casual stays exactly as spoken, and everything else gets a careful pass that fixes what was misheard and rewrites nothing. It is on by default, with switches on the Home page, the Modes page and in Settings, General, Auto mode.
I dictate everything, and I got tired of picking a mode before every dictation, so Auto picks for me. The checks and the cleanup model it can call both run on your Mac, so nothing leaves the machine to decide.
This app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.
On this page: 10 sections
What each dictation turns into
Each dictation gets one of four treatments:
- A letter. A dictation shaped like one, opening with a greeting such as “Dear Jessica” and closing with a sign-off such as “Best regards”, becomes an email: the body only, no Subject line, addressed to the person you named, and signed with your name if you gave the app one.
- Short or casual chat. Six words or fewer, or a short message with casual words like gonna or lol, stays exactly as spoken.
- Everything else. A careful pass that fixes misheard words, punctuation and capitalization, and keeps your wording, tone, slang and sentence structure.
- A code editor or terminal. When the words alone do not decide, dictating into one leans toward Code Comment.
Auto never turns your speech into a bulleted list or a headed note on its own. Say “as a bullet list” at the end of the dictation, or pick Note mode, when you want one.
How it decides
Auto reads four signals, cheapest first:
- A spoken instruction at the end. Something like “make that formal” is taken off the text and followed, whatever else Auto decides.
- The shape of the words. A greeting and a sign-off mean an email. A handful of words, or a short casual message, means chat. Ordinary prose gets the careful pass.
- The app you are dictating into. When the shape leaves it open, Mail, Outlook, Spark, Airmail and Superhuman lean email. Messages, Discord, Slack, WhatsApp, Telegram, Messenger and Teams lean chat. Xcode, VS Code, Cursor, JetBrains IDEs, Sublime Text, Zed, iTerm2, Terminal, Hyper and WezTerm lean Code Comment. Notes apps get no lean.
- A one-word check by the cleanup model. Only when the shape and the app still leave it unclear.
Most dictations are sorted by their words alone, so the model is asked to classify only when the words leave it unclear. History records the choice: click Show pipeline details on any dictation and the Auto detected line says what it picked.
Steer it with a few words at the end
End a dictation with an instruction and Auto follows it instead of typing it.
Say: End any dictation with this while Auto mode is on.
make that formalWhat happens: The phrase is not typed. Auto mode rewrites the dictation in a formal, professional tone.
- Make it formal. Or make it professional: a formal, professional tone.
- Make it friendly. A warm, friendly tone.
- Make it casual. A relaxed, casual tone.
- Make it concise. Or make it shorter: as short as it can be without losing meaning.
- As a bullet list. A bulleted list, the one way to get bullets from Auto without picking Note.
- As an email, as a note, as a message. A complete email, a tidy note, or a short casual message.
That, this and it all work in place of each other, “more formal” works too, and a “please” on the end is fine. An instruction makes Auto rewrite even a message it would otherwise have left as spoken. The instruction goes at the very end, and it is never typed.
Greetings, your name, and your language
Three rules decide how an email opens, how it is signed, and what language comes out:
- Greetings. An email gets a greeting only when you name or clearly imply a recipient, like “hi Jessica”.
- Your name. Emails are signed with the name in Settings, Advanced, Your name. Auto asks for it when it is empty, in a Sign emails as field on the Home page and during setup. With no name, there is no sign-off, and never a placeholder like [Your Name].
- Your language. The rewrite stays in the language you spoke. Auto does not translate.
If Whisper misspells your name at the end, YapToText writes it the way you gave it, as long as it sounds the same. That is how “Thank you, Riley” comes out as “Thank you, Ryleigh” for me. Give the app your name once, and emails sign with it from then on.
What the careful pass will not do
When Auto picks the careful pass, its instructions tell the cleanup model to repair the transcript and leave your voice alone:
- No rephrasing. It does not formalize, soften, shorten or expand anything, and keeps blunt or profane words as you said them.
- No dropped words. It never deletes a sentence or a clause, asides included, and never replaces anything with an ellipsis.
- No invented details. Spoken numbers become digits, without a unit or a time of day you did not say.
- No notes. It outputs only the corrected text, with no comments about what it changed.
The app checks the result too. If it reads like an assistant's reply, grows well past what you said or repeats the reference text, it is thrown away and the raw transcript is typed instead. Auto fixes the transcript, not the person who said it.
Turn it on or off, or skip it once
Auto is easy to switch off, and easier still to skip for a single dictation:
- The switches. Home page, After transcription card, Auto mode. The top row of the Modes page. Settings, General, Auto mode, Adapt to what you're saying, and the setup screen.
- The mode lists. While Auto is on, it is number 1 in the Mode picker in the menu bar and on the panel. Picking another mode in the menu bar turns Auto off, and Auto then leaves those lists until you switch it back on, on the Home page or in the Auto row at the top of the Modes page.
- Just this once. While you talk, press 2 to 9 to use that mode for this dictation. Press 1 to bring Auto back. The next dictation uses Auto again, and YapToText's keys lists the rest.
- The master switch. Auto needs Post-transcription analysis, on the Home page. With that off, every dictation is typed exactly as spoken.
A number key overrides Auto for one dictation without turning it off. The modes guide explains what each mode Auto can pick does.
Limits
- A cleanup model. Anything beyond keeping words as spoken needs one. The Homebrew build ships without models, so download Phi-3.5 Mini Instruct on the AI Models page. The models guide has the steps.
- Very short emails. Six words or fewer stay as spoken, even to Mail. End with as an email to force the format.
- Fixed app lists. Only the apps named above lean one way. Every other app gets no lean.
- Per-app modes. While Auto is on, it decides how each dictation is written, so a mode you gave an app shapes the writing only when Auto is off.
- No lists by itself. Bullets come only when you ask for them.
If Auto picks wrong
- Say it at the end. Finish with as an email, make it formal or another instruction above.
- Press a number. While you talk, pick the mode for this one dictation.
- Fill in your name. Settings, Advanced, Your name, so emails sign correctly.
- Check what it saw. History, Show pipeline details, Auto detected.
- Turn it off. Switch Auto off and pick a mode, or give each app its own mode.
If it keeps getting one kind of dictation wrong, open an issue on GitHub with what you said and what came out.
The short version
- Letters. Become emails, signed with your name.
- Chat. Stays exactly as you said it.
- Everything else. Gets misheard words fixed and nothing rewritten.
- You stay in charge. Say make that formal or as a bullet list at the end, or press a number.
Talk first, and let Auto work out what it is. For AI dictation in general, without the cloud, AI dictation on a Mac has the bigger picture. Every issue on GitHub gets read.
This app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.
Related
Questions
Does YapToText format emails automatically?
Yes, with Auto mode on, which is the default. A dictation shaped like a letter, with a greeting and a sign-off, becomes an email body addressed to the person you named and signed with the name in Settings, Advanced, Your name.
Why did Auto mode not clean up my casual message?
Short and casual messages stay exactly as spoken on purpose. End with make that formal, or press the number for Clean Up while you talk, to have it rewritten.
Can Auto mode make a bulleted list?
Yes, when you ask. End the dictation with as a bullet list. Auto never turns speech into a list on its own.
Where does Auto mode get my name for the sign-off?
From Settings, Advanced, Your name, which Auto asks for when it is empty. With no name set, it leaves the sign-off out and never writes a placeholder like [Your Name].
Does Auto mode send my dictation anywhere to decide?
No. The checks and the cleanup model run on your Mac. The only network use is a model download you start yourself.