hushtype · questions

Questions, answered plainly.

Permissions, offline use, languages, limits, and a straight list of what is still rough. If yours is not here, email us.

Buying and installing

How much is it?: Paid once, in US dollars: $11.99 at launch until 18 November 2026, 18:30 UTC, then $21.99. One licence works on 2 Macs. There is no subscription, no account and no word limit.
How do I get it?: Buy first, then download. Checkout takes you to a thank-you page that shows your licence key with a Copy button and a big Download button. The same key arrives in your order email. Any time later, paste it into "Bought already? Enter your licence key" on the install page (or press Download in the app's Settings › Licence) for a fresh download link; each link works for 10 minutes, and you can ask as often as you need. There is no public download: the disk image is only handed out for a key that is paid and not refunded.
How do I unlock it after installing?: Open hushtype, click its icon in the menu bar, paste your licence key under "Bought already? Enter your licence key" and click Activate. It unlocks on that Mac straight away. The app checks the licence again now and then; offline it keeps working until the check is due, then asks to go online once.
Can I hear it before I buy?: A web page cannot run the on-device recogniser, and sending your voice to a server would break the promise, so there is no voice demo here. The cleanup demo on the home page runs the same rules on text you type. If hushtype does not work for you, ask for a refund within 7 days (see refunds).
How many Macs?: One licence works on 2 Macs. Enter the same key on the second one. Moving to a new Mac? In Settings › Licence on the old one, click Sign out of this Mac to free its place, then activate the key on the new one. Lost the old Mac? Email hello@tealmac.com.
What does it need?: macOS Sonoma 14 or later. It is a Mac app only: there is no Windows, Linux, iPhone or web version.
Why does macOS say it was not opened?: hushtype is not notarised by Apple yet, so macOS asks you to allow it once: click Done, then open System Settings › Privacy & Security and click Open Anyway. After that it opens like any other app. The install guide has the steps with pictures.
What is the refund policy?: 7 days from purchase, if hushtype has a verified problem on your Mac that we cannot fix. A full refund turns the licence key off. See the refunds page.

Permissions

Why does it ask for the Microphone?: To hear you while you hold the dictation or command key, and while a meeting is being recorded. If you switch on always-listening for voice commands, the microphone stays open until you switch it off, and macOS shows its orange dot. Otherwise the microphone is off.
Why Speech Recognition?: It lets your Mac's own recogniser turn what you said into text. Network recognition is switched off, so this happens on the Mac.
Why Accessibility?: So the app can notice your dictation key and type into the app you are using. The same permission lets voice commands press keys, click buttons and menu items by name, move windows and select text. It never types into password fields.
Why would it ask to control System Events or Finder?: Only when you first say restart, shut down, log out, dark mode or empty the Trash. macOS asks once whether the app may ask System Events or Finder to do that. Nothing else is controlled this way.
Does it need Input Monitoring?: hushtype watches a modifier key (Right Option by default). Whether macOS also asks for Input Monitoring for that on your version is something we have not confirmed. If hushtype is listed under System Settings › Privacy & Security › Input Monitoring and dictation does not start, turn it on.
Why Screen Recording, for meeting notes?: macOS files the capture of system sound under Screen Recording. Meeting mode needs it to hear the other side of a call. The app receives sound only and never captures the picture. If you never use meeting mode, you never need to allow it.
Reminders and Calendar?: Only if you press the follow-up buttons after a meeting. Then macOS asks once for access so action items can become reminders and timed sentences can become calendar events.

Offline and privacy

Is anything sent to a server?: Not your voice and not your text. Recognition and cleanup run on your Mac. The app's network calls go to our licence server (activating your key, re-checking the licence now and then, signing out of a Mac and asking for a download link) and, only if you ask for the accurate engine, one download of its model files from downloads.tealmac.com. The privacy page lists exactly what each one sends.
Does it work with Wi-Fi off?: Yes, as long as your Mac has the on-device speech model for your language (System Settings › Keyboard › Dictation). If it does not, the app tells you; it does not fall back to a server. Some features use system models, such as the language model behind Polished and voice edit; macOS may need to download those once, and that download is the system's, not the app's.
Does it save my audio?: No. Audio is turned into text and thrown away. A test scans the source to make sure nothing writes it to disk. A side effect: if recognition fails, the audio cannot be replayed, and the app says so.
Does it read my screen?: It takes no screenshots and never captures the picture. When you give a voice command such as click Done or show numbers, it reads the names and positions of the buttons and menus in the window in front through Accessibility, at that moment, and keeps none of it. Meeting mode hears sound only.
What does it do with my clipboard?: To type, it can paste. That briefly puts your text on the clipboard and then puts back what was there before. If you prefer, it can insert through Accessibility instead.
Can I delete everything?: Yes. One click exports everything and one click deletes history, meetings, stats and transcripts. You can also auto-delete history after 1, 7, 30 or 90 days, or keep no history for chosen apps. Your licence stays.

Accuracy and feel

Is there a more accurate engine?: Yes, as an option. In Settings › Engine you can download a larger speech model once (the accurate model, in about 100 languages, or a smaller fast model for Macs with less memory). Before anything is fetched, hushtype shows what it is, how big it is and that it happens once. After that it runs on your Mac with Wi-Fi off. Apple's recogniser stays the default and takes over whenever the accurate engine can't run, so a dictation is never lost to it. Meetings and hands-free listening always use Apple's recogniser.
Does the accurate engine send anything?: No. The one download fetches the model's files from downloads.tealmac.com and checks each file against its published checksum; it sends nothing about you or what you say. After that the engine reads the audio on your Mac only. You can delete the model in Settings › Engine at any time.
Which Macs can run the accurate engine?: Macs with Apple's own chips, not Intel Macs (those keep Apple's recogniser). The accurate model wants 8 GB of memory and is more comfortable with 16 GB of memory; the fast model needs less. When it starts for the very first time, macOS prepares the model for your Mac. On the accurate model that takes several minutes; Apple's recogniser is used until it is ready, and later starts take a few seconds.
Which words does it listen for?: Before you speak, the recogniser is given your dictionary, your snippet names and the vocabulary packs you switched on for the app in front, most personal first, up to a hundred. If you switch on names on screen, the visible labels of the front window are added, never password fields or the text you are typing, and nothing is kept.
What is hush mode?: For dictating while speaking softly: in a shared office, at night, next to someone asleep. Quiet speech is raised to a steady level before the recogniser hears it, the room is faded down between words, and the pill shows a meter made for a whisper so you can see you are heard. Switch it on in the menu or in Settings › Engine. It works on the audio in memory; nothing is recorded.
What does live text show?: The words heard so far, in the recording pill while you speak, with the newest words in view. When you let go, the cleaned-up text is typed and the pill clears. With the accurate engine the words so far are re-read about once a second; switch live text off in Settings › Engine if you prefer a quiet pill.
Does quick start keep the microphone on?: No. Quick start keeps the recogniser (and the accurate model) loaded for a while after each dictation, so the next one starts sooner; the microphone stays closed. A separate option, off by default, keeps the microphone ready too: then it stays open for the same minutes with the orange dot showing, and only the last half second is held in memory, so a word said as you press the key isn't cut. Neither runs on Low Power Mode, and both last a third as long on battery.
What does the Prompt style do with a long request?: Three sentences or more are sorted into the goal first, then Requirements, Constraints ("don't", "make sure", "never") and Done when ("it should", "so that"). File paths you say become @ references, and "inline code … end code" and "code block python … code block" give code by voice. Your own bullets and code blocks are kept as they are, and short requests stay as said. On macOS 26 with Apple Intelligence you can also have the on-device language model tidy the wording.

Using it

Does it work in every app?: Pasting works in most apps. Games and remote desktops tend to ignore pasted text, and some apps, several built on Electron, do not take Accessibility insertion. It always skips password fields. You can name apps where it should never type.
Which languages?: Dictation listens in the languages your Mac has installed for on-device recognition, one at a time; set it per app. Cleanup rules (fillers, spoken punctuation, take-backs) exist for English, German, French, Spanish, Portuguese, Hindi. The Portuguese and Hindi rules still need a native check. The app itself is translated into 21 languages.
What is English only?: Smart formatting (emails, web addresses, phone numbers, dates, dollars and cents), lists and formatting by voice, voice commands, and numbers, units and developer symbols. The recogniser and the other cleanup rules follow the language lists above.
Are the translations good?: They are machine-made and not yet reviewed by native speakers, so some wording will be off. Text that lives in the core rules, such as rule reasons, licence errors, style names and transcript speaker labels, and the permission prompts, is English only. Newer screens are English only until the catalog is rebuilt.
Are there limits?: No word limit, no meter and no weekly allowance. The Mac's recogniser can stop on long audio, so the app restarts it as it goes; very long meetings and files are where a gap is most likely. Tell us if you find one.
How accurate is it?: Accuracy is whatever the recogniser you chose gives: Apple's on-device recogniser, or the optional accurate engine you can download once. Cleanup fixes fillers and formatting, not words heard wrongly. We publish no accuracy figures and make no claim about it. Your dictionary is given to the recogniser before you speak, and Teach it helps with a mistake that repeats.
How fast is it?: The Stats window shows your own timings: the median and the slowest 1 in 20, from key release to typed text, over your last 200 dictations. We publish no speed numbers, because yours depend on your Mac.
What do Polished and voice edit need?: Mechanical edits run by rules on any Mac. Open-ended edits, a grammar pass in Polished mode and translation use your Mac's own on-device language model, on the Macs that have one. Where there is none, rules do what they can and the app says what it cannot do.
The Mac already has dictation. Why this?: The Mac's dictation types what it hears. This adds the cleanup rules, a style per app, symbols for code and coding agents, a dictionary and templates, history you control, meeting transcripts and file transcription. Whether that is worth it is for you to decide; the refund window is there for that.

Controlling the Mac by voice

How do I fix a word it got wrong?: Say correct that for a numbered list of the recogniser's other guesses for what you just said, plus a capitalised and a singular or plural version, then say choose 2 or click a line. Correct and a word does the same for one word. If nothing fits, choose Spell it and spell the word; the rest of the sentence comes back after it. The other guesses are kept in memory for your last dictation only and never written to disk.
Can I edit a word without moving the cursor to it?: Yes. Say show text numbers and every word you can see in the field gets a number. Then say select 12, delete 7, capitalise 3, replace 5 with a new word, insert a word after 5, or click 4 to put the cursor there. The numbers update after each change.
Does it work with VoiceOver?: Say turn on VoiceOver or turn off VoiceOver: the app presses Command F5, the switch macOS gives VoiceOver, after checking whether it is on already. VoiceOver's own commands stay VoiceOver's. The Accessibility Keyboard has no public switch, so that command opens its settings.
Can I use the whole Mac by voice?: That is the aim: apps, windows, clicking anything by name, number or grid, keys, typing and spelling, moving and selecting text, System Settings, notifications and your own commands, with no keyboard and no mouse. The voice control page lists every command with the words to say.
How do I give a command?: Three ways. Hold the command key you pick in Settings › Commands, say it and let go. Say a whole command while dictating with the dictation key; anything that is not a whole command is typed. Or switch on always-listening and start with your wake phrase.
Is it listening all the time?: Only if you switch always-listening on; it is off until you do. While it is on, the microphone stays open and macOS shows its orange dot. Speech is recognised on the Mac, only sentences that start with your wake phrase or come right after a command do anything, and nothing heard is kept. Say go to sleep to pause it.
What if it mishears and does something I did not want?: Restart, shut down, log out, emptying the Trash and any send, delete or leave always ask first. Quitting an app and sleeping the Mac ask by default. Undo that and put the window back take most other things back. A sentence that is not clearly a command is typed instead.
Does it work in every app?: Clicking by name and selecting text by what it says need an app that shares its buttons and text with macOS through Accessibility. Most Mac apps do; many web editors and some apps do not. There, show numbers, the grid and the keys still work.
Which languages do commands understand?: English for now.

Plans, captions, typing that fits, nod to confirm

What can a plan do?: Open, switch to, hide and quit apps; put windows left, right, full screen or side by side; open folders and System Settings panes; set Focus, dark mode and volume; and find files in your own folders by kind and date, then make a folder and move, copy, rename or move them to the Trash. A plan never types, presses keys, opens web addresses, runs commands or Shortcuts, sends messages, deletes for good or restarts the Mac: ask for one of those and it says so and does nothing.
How do I ask for a plan?: Hold the command key, or start with your wake phrase, and say what you want in your own words: move today's screenshots from the Desktop into a folder called October, or set up for a call: Zoom left, Notes right, Focus on. While dictating, start with do this. The panel lists the numbered steps and, for files, how many and which ones; nothing happens until you say yes or click Yes.
Does planning need the internet or macOS 26?: No internet: everything is planned on the Mac. Common requests (moving, copying, renaming or trashing files by kind and date, setting up apps and windows) are planned by fixed rules on any Mac. Anything else is planned by the Mac's own on-device model, which needs macOS 26 with Apple Intelligence on. Either way, the plan is checked against the list of allowed steps before you see it.
Can a plan delete my files?: No. The furthest it goes is the Trash, and it always asks first. It only works inside your own folders (Desktop, Documents, Downloads, Pictures, Music, Movies, iCloud Drive), never in Library, hidden folders or apps, and it never overwrites: a name that is taken gets a 2, 3 and so on. Say put the files back to undo the last file plan.
What do live captions capture?: The sound your Mac plays, the same way meeting mode does: no picture, and never hushtype's own sound. It is recognised on the Mac in the dictation language and shown in a bar on top of every window. macOS asks for Screen Recording because that is how it gates system sound.
Can captions be translated?: Yes, on macOS 26, with macOS's own on-device translation. If macOS hasn't installed a language yet, it needs its own download first (System Settings › General › Language & Region › Translation Languages); until then the captions show the original. You can show the original under the translation.
Are captions saved?: No. They live in memory while captions are on and are gone when you stop. Save this transcript, in the menu, writes a Markdown file with times, to a place you choose.
Does it read what I have typed?: Only a short stretch before the cursor in the field you are dictating into, read once when you start, to fit the new text to it: carry on a list, lower-case the opening word mid-sentence, fix spacing and full stops, keep code indentation and naming. Never a password field, never kept, never sent. Switch it off, or off for single apps, in Settings › Assist.
What is nod to confirm?: When hushtype asks before something risky (sending, deleting, quitting, running a plan), you can nod for yes or shake your head for cancel, with AirPods or Beats that track head movement. Head movement is read only while the question is on screen, stays on the Mac and is never kept. It is off until you switch it on, and macOS asks once for Motion access. Walking, turning to look at something or a slow look down don't count.
Can I start dictation from my AirPods?: Not yet. macOS gives apps no public way to read an AirPods press without taking over play and pause for every app, so we haven't built it.

What is rough or not finished

Has all of it been tried on real Macs?: The rules are covered by automated tests. The parts that need a Mac with permissions, such as the microphone, the key monitor, typing into other apps, meeting sound capture, Focus filters, Shortcuts, Reminders and Calendar, are written and build, but we have not yet run them ourselves on a Mac with every permission. Expect rough edges there, and tell us.
How far has the accurate engine been tried?: On a Mac with Apple's own chips we ran both models on recorded speech: they load from the downloaded folder, read a sentence with names in it correctly, and the download was tested against a server that drops the connection part way (it carries on where it stopped) and one that serves a damaged file (it is refused and deleted). Live text, quick start, keeping the microphone ready, hush mode on a real microphone and names on screen have not yet been tried by us with a real microphone and every permission.
How far have plans, captions, typing that fits and nod to confirm been tried?: Their rules are covered by automated tests: which files a plan picks and where they go, the checks on every plan, the caption lines and translation steps, the typing fixes, and what counts as a nod or a shake on recorded and made-up head movement. Running plans on real folders, captions on real sound, reading real fields and nodding with real AirPods have not yet been tried by us on a Mac with every permission. Expect rough edges there, and tell us.
What is still machine-translated?: The 21 interface languages, as above.
What are the limits of meeting mode?: Everyone else on the call is one voice, Others. Without headphones the microphone picks up the speakers, which shows as repeated lines, and these are not removed. Decisions and action items come from phrase rules, so they miss some and will tag some lines wrongly. Recording other people has consent rules that differ by place.
What are the limits of file transcription?: One speaker label per file. You cannot drop a file straight on the menu bar icon, because macOS menu bar items do not take drops; use the open menu panel, the app icon or Open With. Access to the Voice Memos folder may be refused by macOS, and then the app asks you to pick it.
What are the limits of Teach it?: It does not watch other apps. You load your last dictation, correct the text and press Find changes. Only mishearings of up to 3 words are proposed, not rewording, and nothing is learned until you press Add.
What are the limits for languages?: The recogniser listens in one language at a time, so speech in another language needs to be set per app or by hand. Cleanup rules switch between your languages sentence by sentence, by script and common words, but words from another language inside one sentence are not split.
Can it type in another language while I speak?: Not yet. The step is built, but the app has no translator wired in. Voice edit can translate selected text where your Mac has an on-device language model.
Has voice control been tried on real Macs?: Its rules are covered by automated tests and it builds, but we have not yet run it on a Mac with every permission against real apps: clicking, overlays, window moves, text selection, notifications and always-listening may have rough edges. Bluetooth, Focus and the Accessibility Keyboard have no public switch on macOS, so the app opens their settings instead. Numbers on words need fields that report where each word is drawn, which many web editors don't, and the correct that list depends on how many other guesses the recogniser on your Mac returns.
Anything else?: There is no global hotkey for read-back yet. Where a spoken file reference ends in Prompt style is a guess.

$11.99, once. 2 Macs.

Launch price: $11.99 until 18 Nov, then $21.99. One-time, 2 Macs, no subscription.

Buy, then download. No account, no subscription, no word limits.

Get hushtype $11.99 Bought already? Enter your key

Refunds · Support