Here is the whole product in one example. You hold the hotkey in your email client and say this, the way people actually talk:
"ok so um the meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again"
"The meeting is moving to tomorrow at 10:00. The client is not available. Send him the invoice again."
In Plain mode, the same sentence comes out as "The meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again." Plain keeps your words, drops the leading "um", capitalizes the first letter and adds a period, all on your own CPU with nothing uploaded. Polished sends the recording to AirTypes and returns the clean version. You pick the mode on the Home screen and can switch any time.
Install AirTypes on macOS or Linux
Go to airtypes.com/download. The page detects your system and offers the right file.
macOS
- Pick Apple Silicon (M1 or later) or Universal (Intel and Apple Silicon). Not sure? Apple menu > About This Mac: a Chip line means Apple Silicon, a Processor line (Intel) means Universal. AirTypes needs macOS 12 or later.
- Open the .dmg and drag AirTypes to Applications.
- AirTypes is not yet notarized by Apple, so macOS may block the first launch. Open it once, close the warning, then go to System Settings > Privacy & Security, scroll down and click Open Anyway next to the AirTypes line. You only do this once.
AirTypes lives in the menu bar. You do not need to keep its window open for the hotkey to work.
Linux
- .deb for Ubuntu, Debian, Mint and Pop!_OS.
- .rpm for Fedora, RHEL, openSUSE and CentOS.
- .AppImage for any x86_64 distro, with nothing to install: make it executable and run it.
Install the .deb or .rpm from a terminal with the command on the download page: it also installs libxdo, which AirTypes needs to start and the package does not pull in by itself (the fix if you already installed it). Linux needs Ubuntu 22.04, Debian 12, Mint 21, Fedora 36 or newer. The hotkey works in an X11 session; on Wayland it may not respond, and the Wayland fix shows how to switch.
On Linux, AirTypes sits in the system tray. Both Plain tiers run on a 4 GB machine with no GPU. For distro details, see the Linux voice dictation guide; for Mac specifics, the macOS dictation guide.
First launch: account, voice model, permissions
The first time you open AirTypes, it walks you through three short steps.
-
Create a free account
Use an email and a password, or Continue with Google. Already signed in on airtypes.com? Pick Sign in with a code from airtypes.com in the app, choose Sign in to the desktop app on the website, and paste the one-time code it shows. No card. Confirm your email within 7 days, or the account pauses until you click the link. The account is what lets you add Polished credit later; Plain stays free with no time limit.
-
Pick a Plain voice model
AirTypes downloads one Whisper model to your computer. It downloads once and then works offline. Not sure? Start with Crisp, the one the app suggests, and change later (see tiers below).
-
Grant permissions on macOS
Microphone, so AirTypes can hear you while the key is held, and Accessibility, so it can type your words into other apps. macOS asks once. Without Accessibility, AirTypes can hear you but cannot type, and the Home screen tells you so with a button that opens System Settings.
Setup ends with a short guide: where to talk, the two modes, and whether you want to hold the key or press it once. The key already works while you read, so you can try a sentence on the spot. You can reopen the guide any time from Settings > Help.
The hotkey: hold, speak, let go
The default hotkey is Ctrl + Shift + Space. On a Mac that is the Control key, not Command. It works from any application, wherever your text cursor is.
- Click into any text field: an email, a Slack message, a code editor, a terminal, a form in your browser.
- Hold the hotkey. You hear a short beep when recording starts (turn sounds off in Settings if you prefer silence).
- Speak in full sentences at your normal pace.
- Let go. AirTypes transcribes and types the text at your cursor.
Hold or toggle
In Settings > Controls > Recording Mode you choose between Hold to Record (the default: hold while speaking, release to stop) and Press to Toggle (press once to start, press again to stop). Hold suits short messages. Toggle is easier for a long paragraph or a first draft, because you do not have to keep three keys pressed.
Changing the key
Click the hotkey pill under Settings > Controls > Hotkey and press the combination you want, then save. Avoid combinations your system already uses, such as Cmd + Space for Spotlight on macOS. The app suggests Cmd + Option + Space on a Mac and Alt + ` on Linux as alternatives. On a Mac, Cmd + Option + Space opens a Finder search window by default, so turn that off under System Settings > Keyboard > Keyboard Shortcuts > Spotlight before you use it.
When the text appears
Nothing is typed while you are still speaking. AirTypes listens for as long as the key is held (or until you press it again in toggle mode), then transcribes the whole recording and types it at your cursor once you let go. This works the same way in Plain and Polished, so you can pause to think mid-sentence without half-finished words landing in your document.
Plain vs Polished, and how to switch
AirTypes has two modes. You switch between them on the Home screen, and the choice applies to every app.
| Plain | Polished | |
|---|---|---|
| Price | Free, no time limit | Prepaid credit, $5 for 600 minutes of speech |
| Where speech is transcribed | On your CPU, offline | On AirTypes servers, by a large speech model |
| Audio uploaded | 0 bytes | The recording is sent to AirTypes |
| Internet | Not needed, works in airplane mode | Needed |
| Languages | English | The language you speak |
| Cleanup | Light: leading um/uh, first capital, final period | Removes filler words and false starts, fixes grammar and punctuation |
| Hardware | Runs on 4 GB RAM, no GPU | Nothing heavy runs locally |
When to use Plain
Plain is the right choice when the exact words matter, when the content is private, or when you are offline: notes, code comments, client details, anything under an NDA, dictating on a plane. It is also the default for people who simply want fast, free typing by voice and do not mind cleaning up an "uh" now and then.
When to use Polished
Polished earns its cost when the text goes to someone else and you want it to read well on the first try: client emails, support replies, a message to your manager. It also helps if English is not your first language, because it fixes grammar while keeping your meaning:
"yesterday I am going to client office and they are telling the payment will be late by two week, so we need to discuss about the budget again"
"Yesterday, I went to the client's office, and they said the payment will be late by two weeks. So, we need to discuss the budget again."
Polished aims to keep your meaning, numbers and names, and a guard backs that up: it keeps your numbers and catches rewrites that drift too far from what you said. If the cleanup drops a number you said or strays far from your length, AirTypes types your own words (the raw transcript) instead, and Home shows a short note saying it kept your words. It can also drop a greeting such as "hi" at the start of a message, so add one back if you need it. More detail on the Polished section of the homepage.
Start with Plain, free and offline
Install AirTypes, hold the hotkey, and type by voice in any app. Add Polished credit only if you want the cleanup.
Download for macOS and LinuxAdding Polished credit inside the app
Polished runs on prepaid credit. Your first minute is free, so you can try it before you pay. There is no subscription and nothing renews.
- On the Home screen, choose Polished. Once your free minute is used and you have no credit, it reads Polished · Add credit.
- Pick an amount: $5 (600 minutes of speech), $10 (1,200 minutes), one of the larger amounts, or Other to type any whole-dollar amount. Have a coupon or referral code? Add it under Have a code?: it adds extra credit, the price stays the same.
- Our checkout page opens in your browser. Pay there.
- Polished turns on the moment the payment goes through. You can keep dictating in Plain while you wait.
Credit is charged by the second, only while you hold the key, one price for transcription and cleanup together: $5 buys 600 minutes of talking, and a 30-second note costs under 1¢. Credit never expires. When it runs out, Polished pauses, your words are typed as you said them, and Plain keeps working for free. Full numbers are on the pricing section.
Your language, and Always translate to English
Plain is English only. Polished works in the language you speak. That includes English, Hindi, Bangla, Japanese, German and most other spoken languages, so try yours.
There is one setting for this, under Settings > Controls > Polished: Always translate to English. Off, Polished writes in the language you spoke. On, you speak your language and English is typed. Here is the same Bangla sentence both ways:
"কাল সকালে মিটিং টা হবে না, client এর flight delay হয়েছে, আমি নতুন time টা email করে দিচ্ছি"
"The meeting will not be tomorrow morning. The client's flight has been delayed. I am emailing the new time."
With the setting off, Polished keeps it in Bangla and tidies the punctuation: "কাল সকালে মিটিংটা হবে না। ক্লায়েন্টের ফ্লাইট ডিলে হয়েছে। আমি নতুন টাইমটা ইমেল করে দিচ্ছি।"
One thing to know: with Always translate to English on, the guard checks only that your numbers survive. A translation cannot be compared for length with what you said, so read names before you send.
Picking a Plain accuracy tier
Plain runs OpenAI's Whisper on your own CPU. You choose between two tiers under Settings > Voice model. Each model downloads once and stays on your computer.
| Tier | Whisper model | Download | Best for |
|---|---|---|---|
| Swift | tiny (English) | ~78 MB | Quick notes on any machine, near instant |
| Crisp | base (English) | ~148 MB | Everyday messages, fast with good accuracy. The tier we recommend |
Which should you pick? Start with Crisp: it is the tier the app suggests, and it runs on a 4 GB laptop with no GPU. Pick Swift when you want the fastest result on an older machine. When you need more accuracy than Crisp gives, for technical terms, accents or long documents, switch to Polished: a large speech model on the server transcribes the recording, so nothing heavy runs on your computer. The Whisper model sizes guide explains the trade-offs in depth.
How text reaches your app
Under Settings > Controls > Text Output you choose how AirTypes delivers the words:
Direct Paste (default)
Copies the text and pastes it for you with Ctrl + V (Cmd + V on macOS). The fastest option for long text, and it works almost everywhere.
Type at cursor
Types each character as if from a keyboard. Slower for long text but the most compatible, including terminals. Pick this if you dictate into a Linux terminal, where paste is often Ctrl + Shift + V instead.
Copy to Clipboard
Puts the text on your clipboard without pasting. Useful when you want to read it first and decide where it goes.
History and export
The History tab keeps every transcript, newest first, stored on this device only. You can:
- Search past transcripts.
- Copy any entry back to your clipboard, handy when you dictated into the wrong window.
- Export JSON to keep a backup or analyze your dictation.
- Clear everything when you want a fresh start.
Account: credit, charges and stats
The Account tab shows:
- Polished credit: your balance, and a list of every charge and top-up, each with one amount.
- Usage: transcriptions, words dictated, and an estimate of time saved compared with typing.
- Activity: a chart of your dictation by week, month or year.
- Support: open a ticket from inside the app if something is not working.
What AirTypes receives
Always: your account email and usage counts, such as how many words you dictated. In Plain, nothing else. Audio is transcribed on your CPU and 0 bytes of it leave your computer. In Polished, AirTypes also receives the recording: Polished sends the recording to AirTypes to transcribe and clean it up.
So the rule of thumb is simple. If a recording should never leave your machine, dictate it in Plain. For a longer discussion, read is voice dictation private?
Tips for daily use
- Speak the whole sentence, then edit once. Stopping to fix words mid-thought is what makes dictation feel slow.
- Start small. Chat replies and short emails first; long documents after a few days. The first-week dictation habit plan walks through it.
- Good audio beats a bigger model. A headset or a quiet room helps more than any setting.
- Mix the modes. Plain for notes, code and private text; Polished for messages other people read.
- Prompts count too. Speaking a prompt to ChatGPT or Claude gives the model more context than a terse typed one. See how to dictate AI prompts by voice.
- Read before you send. A two-second glance catches a misheard name.