Launch price $11.99 until 18 Nov, then $21.99. Pay once, 2 Macs
Pricinghushtype · voice to text for Mac
Hold a key and talk. Clean text appears in the app you are in. Recognition and cleanup run on your Mac, so nothing is uploaded, there is no account and there are no word limits.
One payment · 2 Macs · macOS Sonoma 14 or later · Mac only · buy, then download
Your Mac's own speech recognition does the listening, with network recognition switched off. Cleanup runs on the Mac too. No account and no screenshots. Voice commands read the names of buttons and menus in the window in front only when you give one.
Turn Wi-Fi off and it keeps typing for you, as long as the language's speech model is installed on your Mac. The app goes online only for the licence server (your key, the licence check and a download link) and, if you choose it, a one-time download of the accurate engine; never your voice or text.
No meter, no weekly allowance, no sign-in. Dictate as much as you like.
See it working
Dictation and cleanup, control by voice, Wi-Fi off, the accurate engine, meetings, captions and plans. Captions are on; each feature has its own short video on the features page.
Use cases
Eight ways people use it. Drag, scroll or use the arrow keys to turn the carousel.
Hold the key wherever you can type.
Formal in Mail, relaxed in Messages.
Say a flag or a file path, get what you would type.
Two sides of the call, with timestamps.
Select, hold the key, say what to change.
Recordings in, text out. Thoughts into your notes.
Wi-Fi off. Nothing leaves.
21 languages in the app.
Notes, documents, browsers, forms: hold Right Option and your words are typed at the cursor, tidied by fixed rules. If it pastes, your clipboard is put back afterwards. Password fields are always skipped.
The journey of a word
Four steps from your voice to text in the app you are in. The whole path is on your Mac. The full walk-through.
Step 1
Hold Right Option. A small pill at the bottom of the screen shows that it is listening. Tap twice to keep listening hands-free.
Step 2
Your Mac's own recogniser turns the sound into words, on the Mac. The audio is thrown away, never saved.
Step 3
Fixed rules drop the fillers and take-backs, write numbers and units, and finish the text for the app you are in.
Step 4
Let go and the text is typed at your cursor. It is saved as a draft first, so a failed typing never loses it.
Try the cleanup
Pick an example or type your own. It is the same rule set the app applies after it has heard you.
A text demo of the cleanup rules, not the speech recogniser. It runs only in this page, on the text you type. The app does the listening, on your Mac, and then applies these rules.
Examples
Try saying things like “comma”, “new line”, “no wait”, “scratch that”, “twenty five percent”.
What gets typed
um, the doubled I and the take-back are gone; a like beside commas is dropped.
Cleanup you can predict
The same words give the same text every time. There is no learning from you in the background and nothing that changes without you knowing. Each rule is fixed, and you choose how much of it applies, per app.
Per-app styles
hushtype looks at which app has your cursor and finishes the text to suit it. The built-in guesses are below. In Settings you can set a style, a filler level and a language for any app, or switch it off there.
Formal
Full sentences, written-out words, a full stop at the end.
Same words
we are gonna ship it tomorrow
Typed
We are going to ship it tomorrow.
Casual
No full stop after a single sentence.
Same words
we are gonna ship it tomorrow
Typed
We are gonna ship it tomorrow
Code
Lower case, no full stop, symbols and file names.
Same words
we are gonna ship it tomorrow
Typed
we are gonna ship it tomorrow
Anything else gets Standard. Prompt and Notes are styles you pick for an app, for long messages to coding agents and for notes with lists.
Developers and coding agents
Say file names, flags and symbols and get what you would have typed. Terminals and code editors use the Code style by default. For long messages to coding agents there is a Prompt style that keeps your line breaks and adds bullets and code blocks by voice.
You say
let camel case user name equals camel case get user open paren close paren
Code types
let userName = getUser()
You say
npm run dev dash dash port equals three thousand
Code types
npm run dev --port = 3000
You say
check at path src slash views slash home view dot tsx and fix it
Prompt types
Check @src/views/HomeView.tsx and fix it
You say
steps new bullet install it new bullet run it
Prompt types
Steps - Install it - Run it
You say
run code block npm test code block now
Prompt types
Run ``` npm test ``` Now
You say
look at file app model dot swift please
Prompt types
Look @AppModel.swift please
Meeting mode
Captions for anything your Mac plays, translated if you like, and notes with who said what. Try the translation switch.
Priya: Can we move the review to Friday?
Start meeting mode and hushtype records your microphone and the sound your Mac plays, then keeps a transcript on this Mac with You and Others turns, timestamps and notes. It records sound only. It never captures the screen.
# Weekly sync 23 Oct 2026, 2:05 PM · 24:10 ## Decisions - We decided to ship on Friday. ## Action items - [ ] I'll send the release notes by Thursday. (You) - [ ] Can you review the pricing page? ## Transcript **Others** [00:00] We decided to ship on Friday. The design is done. **You** [00:06] I'll send the release notes by Thursday. **Others** [00:10] Can you review the pricing page?
Voice edit
Select some text, hold the key and say what you want done to it: “make it shorter”, “more formal”, “fix the grammar”, “translate to Spanish”.
Mechanical edits run by rules on the Mac. Open-ended edits and translation use your Mac's own on-device language model, on the Macs that have one. Rules alone cannot translate, and the app says so instead of guessing. Nothing is sent to a server either way.
Control your Mac
Apps, windows, clicking by name or number, keys, text, settings and your own commands, with no keyboard and no mouse. Recognised on the Mac. Press a command below, or grab the scene and turn it.
Open, switch, quit and hide apps by name; open folders, files and websites; search the web.
“open Safari” · “switch to Mail”
Halves, quarters, thirds, centre, fill, next display, minimise, full screen, and put it back.
“left half” · “move this window to the right half”
Click by name, by number or on a grid; double-click, right-click, drag, hover and scroll.
“click Done” · “click File, Export as PDF”
Any key or shortcut by name, literal typing, spelling with the phonetic alphabet, and code formatters.
“press command shift four” · “press escape three times”
Move by word, line, sentence or paragraph; select a phrase or a range; replace, insert and delete.
“move back three words” · “go to the end of the line”
Number every word in a text field and edit by number; pick another guess for what you just said.
“show text numbers” · “select 12”
Hold a command key, say a whole command while you dictate, or switch on always-listening and start with “hey hushtype”. Restart, shut down, empty the Trash and any send, delete or leave ask first.
Accuracy and feel
An optional accurate engine you download once, the words you care about given to the recogniser first, live text in the pill, a quicker start and a hush mode for speaking softly. All on your Mac.
Apple's recogniser is built in and always there. If you want more, Settings › Engine offers a larger speech model, in about 100 languages, or a smaller fast one for Macs with less memory.
hushtype tells you what it is and how big before anything is fetched. It downloads once from downloads.tealmac.com, every file is checked against its checksum, and from then on it runs on your Mac with Wi-Fi off. Nothing you say is sent anywhere, and Apple's recogniser takes over whenever the accurate engine can't run.
Needs a Mac with Apple's own chips (Intel Macs keep Apple's recogniser). Its very first start prepares the model for your Mac, which takes a few minutes, once.
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. Names, products and jargon can come out right straight away, not after a fix.
Optional: names on screen. When a dictation starts, the visible labels of the front window (a person in Mail, a channel in a chat app) are added too. Never password fields or what you are typing, and nothing is kept.
Live text shows the words in the recording pill as you speak, newest words in view, so you can see it heard you and stop early if it didn't. The cleaned-up text replaces them when you let go.
Quick start keeps the recogniser loaded for the minutes you choose after each dictation (a third as long on battery, never on Low Power Mode). The microphone stays closed unless you also switch on “keep the microphone ready”, which holds just the last half second in memory with the orange dot showing.
In a shared office, late 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.
One switch in the menu. It works on the audio in memory while you speak; nothing is recorded.
Talk to a coding agent the way you would explain it to a colleague. With the Prompt style, three sentences or more are sorted into the goal, then Requirements, Constraints and Done when. File paths become @ references, and “inline code … end code” or “code block python … code block” give code by voice. Short requests, your own bullets and your code blocks stay as you said them.
You say
um can you add a dark mode toggle to the settings screen. it should remember the choice. use at file theme dot swift for the colours. don't change the layout. make sure the tests still pass
Prompt types
Add a dark mode toggle to the settings screen. Requirements: - Use @theme.swift for the colours. Constraints: - Don't change the layout. - Make sure the tests still pass. Done when: - It should remember the choice.
Assist
Plans from a sentence, with every step shown before it runs. Live captions for anything your Mac plays, translated if you like. Typing that fits the field you are in. A nod for yes. All on your Mac.
Hold the command key (or start with your wake phrase) and say what you want the way you would ask a person: move today's screenshots from the Desktop into a folder called October. Set up for a call: Zoom left, Notes right, Focus on. While dictating, start with “do this”.
hushtype turns it into numbered steps it already knows how to do, shows them with exactly which files they would touch, from a real look at the folder, and waits. Nothing moves until you say yes, click Yes or nod. Say “put the files back” and the last file plan is undone.
Common requests are planned by fixed rules on any Mac. Anything else is planned by your Mac's own on-device model on macOS 26, and every plan is checked against a short list of allowed steps before you see it: apps, windows, folders, Settings panes, Focus, dark mode, volume, and files inside your own folders. A plan never types, presses keys, opens web addresses, runs commands, sends messages, deletes for good or restarts the Mac.
Captions for whatever your Mac plays: a video, a call, a podcast, a lecture. They sit in a bar on top of every window, even over full-screen video, and never take a click. Choose the text size, top or bottom, how dark the background is and how many lines.
Optional translation into your language uses macOS's own on-device translation (macOS 26), with the original underneath if you like. A language macOS hasn't installed needs its own download first.
Only the sound is used, never the picture, and never hushtype's own sound. Nothing is kept unless you choose Save this transcript.
When a dictation starts, hushtype reads a short stretch of text before the cursor and fits what you say to it. In a list, the next item gets its bullet or number. Mid-sentence, the opening word stays lower case. Spaces and full stops at the join come out right.
In code, a spoken name follows the style the file already uses (userName or user_name) and new lines keep the indentation. Cleanup follows the language the field is written in.
Never a password field, used for that one dictation only, never kept. Off for any app you choose.
With AirPods or Beats that track head movement, nod for yes or shake your head for cancel when hushtype asks first, before sending, deleting, quitting or running a plan. Hands stay where they are.
A nod has to be a real nod: walking, turning to look at something or a slow look down at the keyboard don't count, and a firmness slider sets how big it must be.
Head movement is read only while the question is on screen, stays on the Mac and is never kept. Off until you switch it on.
The details
Teach it, smart formatting, voice commands, templates, unsent text, your own timings and the time-saved formula. All of it runs on the Mac.
The same mistake twice is what people dislike most about dictation. Open Teach it from the menu, load your last dictation, fix the text and press Find changes. It proposes “when it hears X, write Y”, or “always write Y”.
Spoken data is written as data, not as words. It is in Settings › General, on by default, and off in Exact mode and Code style. English only, by rules that can be wrong at the edges.
Dollars and cents are written as dollar amounts. Web endings it knows include com, org, net, io, dev, edu, gov, app, info, biz, xyz, tech, uk, de, fr, ca, au and co. A spoken run of 7 digits now becomes 555-1234; switch smart formatting off for plain digits.
Say a command as a whole sentence and it acts instead of typing. A command inside a longer sentence stays words. Off per app in Settings › General.
“Open Mail” and “switch to Safari” work as a whole sentence while dictating, or with the command key; the voice control page has every command. Sending key presses and opening apps are written and build, but we have not yet run them ourselves in every app.
Say the name on its own and the template is typed, with the date, time, weekday, clipboard and current app filled in where it uses them.
Standup, bug report and meeting notes were already there; the Add built-in templates button adds the rest. Macro packs import and export as JSON so a team can share them.
Cleaned text is saved as a draft before it is typed. If typing fails, or the app stops, the text waits in the menu under Unsent text, with Copy and Delete.
The Stats window shows your own timings: the median and the slowest 1 in 20 time from key release to typed text, over your last 200 dictations. Timings only; no text is kept for it.
It is a way for you to check whether it feels fast on your Mac. We do not publish speed numbers, because yours depend on your Mac.
No hidden numbers. hushtype shows how it got the figure, and the last 7 days use the same rule.
time saved = words ÷ your typing speed − time you spent speaking (never below 0)
Example with made-up numbers: 300 words at a typing speed of 40 words a minute is 7.5 minutes of typing. Speaking them took 2 minutes. Time saved: 5.5 minutes. You set your own typing speed in Settings.
Bring your words
Switching should not mean starting your word list over.
Bring a dictionary and snippets from another dictation app as CSV, TSV, JSON, a plain list or lines like a -> b. Column names are read loosely, and a first-run screen walks you through it.
Export the same words as JSON, CSV or text whenever you like. Your words are a file you own, not something held for you.
The importer is written from common export shapes. If your old app saves something it does not read, tell us and we will add it.
Add names and terms so they come out spelled your way, with optional “sounds like” fixes. Built-in packs cover software, legal and health vocabulary.
Say a short trigger, get a longer text. Templates can fill in the date, time, weekday, clipboard and the app you are in.
Word packs import and export as files, so a team can hand the same vocabulary around.
Files and inbox
Drop audio or video onto the Transcribe files window or the open menu panel, or use Open With. Files queue up and are recognised on your Mac.
Save as .txt, Markdown with frontmatter, .srt or .vtt. A watch folder transcribes whatever lands in it and writes Markdown to a Transcripts folder. Each file gets one speaker label.
Press a hotkey, say a thought and it is saved as a line like - 14:05 text #tag in a daily note or a single file. Say “tag work” to add a tag.
Obsidian-style frontmatter and date file names are supported, with a target per style.
Languages
Dictation listens in the languages your Mac has installed for on-device recognition, one language at a time, which you can set per app. If a language is not installed, the app tells you and does not fall back to a server.
Cleanup rules (fillers, spoken punctuation, take-backs) exist for English, German, French, Spanish, Portuguese, Hindi. Numbers, units and developer symbols are English only for now. The Portuguese and Hindi rules still need a native check.
For Chinese, three optional touches run on your Mac: Traditional characters with Taiwan wording (软件 becomes 軟體), spoken numbers written as digits (二零二六年 becomes 2026年) and fewer stray commas.
The app's own menus and windows are translated into 21 languages. These translations are machine-made and not yet checked by native speakers, so some wording will be off.
Privacy controls
Audio is turned into text by the Mac itself and then thrown away. The text goes into the app you chose and, if you keep history, into a file on this Mac.
Wi-Fi on. Speech is still recognised on this Mac.
A visible “On this Mac” line in the menu.
Keep it, turn it off, or delete it automatically after 1, 7, 30 or 90 days. Set “no history” for single apps.
Audio is turned into text and thrown away. It is not written to disk.
Off on calls, in quiet hours, in apps you pick, and per Focus. Password fields are always skipped.
One click exports everything. One click deletes history, meetings, stats and transcripts. Your licence stays.
Words dictated, time saved and streaks are numbers on this Mac. No words or audio are kept for them.
hushtype has no analytics. The privacy page lists the one thing it can send.
Everything in it
Move the pointer over a card. The full index has every feature with one plain line.
Hold Right Option and speak. Tap twice to keep listening hands-free. Escape cancels.
Fillers, repeats, take-backs, spoken punctuation, capitals, numbers and units. The same words give the same text.
The words appear in the recording pill as you speak, newest words in view. The cleaned-up text replaces them when you let go.
For speaking softly: quiet speech is raised to a steady level, the room is faded down between words, and the pill shows a meter made for a whisper. In the menu and in Settings.
A larger speech model, in about 100 languages, that you download once from Settings › Engine; after that it runs on the Mac with Wi-Fi off. Apple's recogniser stays the default and takes over whenever the accurate engine can't run.
Ask in your own words with the command key or your wake phrase, or start with do this while dictating. The plan is shown as numbered steps and waits for your yes. Common requests work on any Mac; anything else uses the Mac's own on-device model on macOS 26.
Captions for whatever your Mac plays (videos, calls, podcasts), in a bar on top of every window and full-screen video. Recognised on the Mac. Size, position, background and number of lines are yours.
Reads a short stretch before the cursor (never a password field, never kept) and carries on a list, writes the opening word in lower case mid-sentence, and gets the spacing and full stops right.
Emails, web addresses, phone numbers, dates and dollars-and-cents are written as data, not as words. English only.
Click by name, show numbers and say one, or zoom a 3 by 3 grid down to a point. Double, right click, hold, drag, hover and scroll.
Formal in Mail, casual in Messages, lower case and symbols in Terminal and code editors. You can change any of them.
A long spoken request is sorted into the goal, then Requirements, Constraints and Done when. Spoken paths become @ references; bullets, inline code and code blocks marked with a language come by voice. Short requests stay as said.
Load your last dictation, fix it, and it proposes dictionary entries for the mishearings. Nothing is learned until you press Add.
Records your microphone and the sound your Mac plays, and keeps a transcript with You and Others turns, timestamps and notes.
Recognition and cleanup run on the Mac with network recognition off. No account and no screenshots. Voice commands read button and menu names only when you give one. The optional accurate engine is downloaded once and then runs offline.
How it compares
| Capability | hushtype | Cloud dictation services | Basic on-device dictation |
|---|---|---|---|
| Where recognition runs | On your Mac | Usually on a server | On the device |
| Works with Wi-Fi off | Yes, if the language is installed | No | Usually |
| Word limits | None | Some cap free use | Varies |
| Account | None | Usually required | Varies |
| How you pay | Once | Often a subscription | Varies |
| Your own wording kept (Exact mode) | Yes | Varies | Varies |
| Style per app | Yes | Varies | Varies |
| Symbols and file references for coding agents | Yes | Varies | Varies |
| Meeting transcript with Markdown export | Yes, one voice for everyone else | Varies | Varies |
| Recognition engine | The Mac's own recogniser. We have not measured its accuracy. | Varies | Varies |
| Platforms | Mac only | Often several | Varies |
The other two columns describe kinds of apps in general, not any one product, and they change over time. Check the app you are comparing with. This page makes no claim about accuracy: it depends on the recogniser, and ours is the one that comes with your Mac.
Pricing
One licence for your Macs, no subscription, no word limits.
$11.99Regular price $21.99
Launch price: $11.99 until 18 Nov, then $21.99. One-time, 2 Macs, no subscription.
After checkout you get a licence key by email and a Download button. Lost the download? Enter your key on the install page.
Questions
All questions, including permissions and what is still rough
Voice to text that stays on your Mac. $11.99 once at launch, then $21.99.
Mac only · refunds