Offline voice-to-text for any desktop app, fully on-device.
Three steps, one hotkey. Hold, speak, release, and your words land where your cursor already is.
Press and hold Ctrl + Shift + Space from any window. Text is injected at your cursor system-wide, from your editor to your browser.
Talk at your own pace while a small overlay shows it is listening. Whisper runs on your CPU, so your voice audio never leaves your machine.
Release the hotkey and the text appears at your cursor in any app, cleaned and capitalized, ready to send.
Hold one hotkey, speak, and clean text lands at your cursor. No window switch, no copy and paste.
One hotkey, every app. Default is Ctrl Shift Space, and you can change it.
Offline voice-to-text for any desktop app, fully on-device.
Speech recognition runs entirely on-device with no internet connection required.
Native keyboard injection places text at your cursor in any app, with no plugins.
Send transcribed speech to your own AI provider and type the response in place.
Words like um and uh are stripped and sentences are capitalized automatically.
Pick Formal, Professional, Neutral, Casual, or Friendly, applied locally.
Choose a model from ultra-light to maximum accuracy to match your hardware.
Set a language or use auto-detect across Whisper multilingual models, all offline.
A second hotkey routes your voice through your own AI provider, then types the answer where you are working. You bring the key. Your key and your prompts never touch our servers.
Hold the agent hotkey and talk. Whisper transcribes on your device.
The text goes straight to your provider with your key. Not through us.
The answer lands in your active app, ready to send or edit.
Bring your own key. Works with the providers you already use.
Your API key stays on your device, and your prompts go directly from AirTypes to your provider. We never see your key, your prompts, or the responses. Questions? [email protected].
How it worksAn honest look at how AirTypes lines up against other dictation apps. Competitor prices are approximate and may change.
| Feature | AirTypes | Superwhisper | Wispr Flow | Whisper Flow |
|---|---|---|---|---|
| Runs 100% offline | ||||
| Linux | ||||
| macOS | ||||
| Windows | in development | |||
| Voice to your own AI (BYOK) | ||||
| Audio stays on device | ||||
| Price | $3.99/mo | approx $849 lifetime | approx $12-15/mo | approx $8-10/mo |
AirTypes
Superwhisper
Wispr Flow
Whisper Flow
Speaking runs around 130 words a minute, typing around 40. AirTypes gives that difference back in any app you already use.
Dictate commit messages, code comments, and pull request notes without leaving the editor.
VS Code, Slack, TerminalDraft long passages by voice, then let local cleanup handle filler and grammar.
Google Docs, Notion, WordAnswer tickets and chats faster by speaking replies that land at your cursor.
Gmail, Zendesk, IntercomCapture notes and prompts by voice and route them through your own AI with My Agent.
Obsidian, ChatGPT, ZoteroA few words from people using AirTypes in their everyday work.
“The privacy angle sold me. I work with sensitive client data, and AirTypes processes everything locally. I verified it by running with my network disabled. Still works.”
“I switched from another tool because it sent my audio to the cloud. AirTypes does everything offline. The tone selector keeps my output consistent without editing.”
“I write code comments, PR descriptions, and Slack messages all day. AirTypes saves me at least an hour. I hold the hotkey, say what I mean, and it shows up where I need it.”
“I run Linux full-time and there is nothing else like this on the platform. It works in my terminal, and the hotkey is system-wide. Voice typing that actually works on Linux.”
“The My Agent feature is genuinely useful. I say ‘Email, tell the board we hit our latest milestone’ and it drafts a polished email right into my compose window.”
“I take notes during user interviews and AirTypes keeps up with natural speech. It works in Notion, Google Docs, Figma. I can’t go back to typing notes.”
“Long writing days were hard on my wrists. Now I speak most of my first drafts and my hands feel better. AirTypes cleans them up with the right tone.”
“English is my second language and AirTypes still understands me. At $3.99 a month it is cheaper than what I used before and fits how I actually work.”
Transcription runs on your own hardware. Here is exactly what happens to your audio, your keys, and the little we keep to run your account.
Fully offline and private
Whisper runs on your CPU, so your voice audio never leaves your machine. Recordings are transcribed on-device and are never uploaded to us or anyone else.
My Agent, your own AI key
A second hotkey routes your speech through your own AI provider with BYOK. When you route voice through OpenAI, Claude, Gemini, Groq, Ollama, your API key and prompts go straight to them. They never pass through AirTypes.
We store only your account basics
Your email and a count of how much you have used, enough to run your subscription. We never store your audio or transcripts, and we never sell your data.
No tracking, no ads
There are no analytics SDKs watching what you type and no advertising profiles.
The product is paid, so you are the customer, not the inventory.
Questions about your data go straight to [email protected].
Choose the Whisper tier that fits your speed and accuracy needs, from near-instant to maximum quality. Transcription runs entirely offline.
~40 MB
Fastest results on any device, great for quick notes.
~75 MB
Balanced speed and accuracy for everyday dictation.
~245 MB
High accuracy on machines with 4 GB RAM or more.
~770 MB
Professional accuracy for longer, detailed writing.
~875 MB
Near-maximum accuracy with fast turnaround.
~1.6 GB
Maximum accuracy for technical content, needs 12 GB RAM.
Try every feature for 7 days, no credit card. Then $3.99 a month, or cancel before the trial ends and pay nothing.
Full access to everything. No credit card to start.
Sign up, download, and start dictating in minutes.
Everything in the trial, kept on. Billed after your 7 days, cancel anytime.
Billed monthly after the trial. Cancel before it ends to pay nothing.
Native downloads run fully offline. Windows is in active development, and subscribers get it at launch with no extra charge.
Apple Silicon and Intel, fully offline transcription.
macOS 12 or later, 4 GB RAM
Ubuntu, Fedora, Arch, and most Debian and RPM distributions.
x86_64, 4 GB RAM, no GPU required
In active development. Subscribers get it at launch with no extra charge.
Windows 10 or later (planned)
7-day free trial / Free signup in seconds / $3.99 per month after
Answers on offline dictation, My Agent, privacy, and pricing.
AirTypes is a 100% offline voice-to-text desktop app for Linux and macOS today, with Windows in development. You hold a global hotkey (default: Ctrl+Shift+Space), speak naturally, then release. AirTypes uses Whisper AI running entirely on your CPU to transcribe your speech and type the result directly at your cursor, in any app, with no internet required.
Yes, 100% offline. All speech recognition happens locally on your device. Your audio never leaves your machine. AirTypes works in airplane mode, on air-gapped computers, in hospitals, government offices, and any environment where internet access is restricted or privacy is essential.
AirTypes works with every desktop application: VS Code, Slack, Gmail, Google Docs, Microsoft Word, Notion, Discord, Terminal, ChatGPT in your browser, and any other app. It injects text at your cursor system-wide using native OS keyboard injection, so there are no app-specific limitations.
Yes. With My Agent and a Prompt profile (or your own instructions), AirTypes sends your transcribed speech to your AI using your API key (BYOK). The model returns polished, structured text you can use with ChatGPT, Claude, Gemini, or any OpenAI-compatible provider. Whisper transcribes offline on your device; the AI call goes straight from your machine to your provider, nothing through AirTypes servers.
AirTypes differs in three key ways. Fully offline: no audio ever sent to any server (most competitors are cloud-based). Native Linux and macOS: live today, Windows in development (many competitors are macOS-only). My Agent (BYOK): a separate hotkey routes voice through your AI, with optional profiles (for example Prompt, Email) via the first words you say. At $3.99/month, it is also more affordable than most alternatives.
AirTypes uses OpenAI Whisper AI, among the most accurate offline speech recognition models available. Accuracy scales with model size: the base model is excellent for everyday use, while Large V3 Turbo and Large V3 deliver near-human-level transcription accuracy for professional and technical content.
Yes. AirTypes automatically strips filler words (um, uh, like, you know, so), fixes grammar, capitalizes sentences, and applies your chosen writing tone. Choose from Formal, Professional, Neutral, Casual, or Friendly, all processed locally with no internet.
AirTypes offers a 7-day free trial with full access to all features, no credit card and a quick free signup. After the trial, AirTypes Pro is $3.99/month. You can cancel anytime.
AirTypes offers 6 voice quality tiers, from ultra-light to maximum accuracy: Swift (~40 MB), under 1s transcription, any device; Crisp (~75 MB), balanced speed and accuracy; Clear (~245 MB), high accuracy, 4 GB RAM; Bright (~770 MB), professional accuracy, 6 GB RAM; Sharp (~875 MB), near-maximum accuracy, fast; Deep (~1.6 GB), maximum accuracy, 12 GB RAM.
Yes. AirTypes supports 100+ languages via Whisper AI's multilingual models. You can set your language explicitly or use auto-detect mode. Languages include Spanish, French, German, Portuguese, Hindi, Japanese, Chinese, Arabic, and many more, all processed offline.
My Agent is a second mode in AirTypes that routes your voice through an AI before it is typed at your cursor. Where normal transcription types what you say verbatim, My Agent sends your speech to an AI that shapes, rewrites, or acts on it, then types the AI's output directly in your active app. The flow is: hold the agent hotkey, speak, Whisper transcribes locally, text is sent to your chosen AI API, and the AI response is typed at your cursor. Everything from Whisper transcription to cursor injection is handled by AirTypes. Your AI call goes directly from your device to your provider, and AirTypes never sees the content.
Yes. AirTypes is currently the only polished, paid, actively-maintained offline voice dictation app for Linux. Superwhisper supports macOS and iOS only; Windows is listed as in progress on their roadmap with no Linux version planned. AirTypes uses the same Whisper AI models as Superwhisper, runs 100% offline on your CPU, and includes features Superwhisper lacks, like My Agent (BYOK AI routing). At $3.99/month it is also significantly cheaper than Superwhisper, which charges roughly an $849 lifetime fee (approximate).
The fundamental difference is privacy. Wispr Flow processes your audio on their cloud servers, so your voice leaves your device. AirTypes is 100% offline: your audio never leaves your machine. Other differences. Linux: AirTypes supports Linux, Wispr Flow does not. Price: AirTypes is $3.99/mo versus Wispr Flow at roughly $12-15/mo (approximate). AI routing: AirTypes My Agent uses your own API key (BYOK) with no data through AirTypes servers, while Wispr Flow routes through their cloud. If privacy and offline operation matter, AirTypes is the stronger choice.
Yes. AirTypes is live on Linux now. Supported distributions include Ubuntu 20.04, 22.04, and 24.04; Fedora 36+; Arch Linux and Manjaro; and most Debian-based and RPM-based distributions. It requires x86_64 architecture and 4 GB RAM minimum. No GPU required, Whisper runs on CPU. macOS is also available now (Apple Silicon and Intel). Windows is in development; subscribers get it at launch at the same price.
My Agent works with any OpenAI-compatible API endpoint. This includes OpenAI (GPT-4o, GPT-4 Turbo, GPT-3.5); Anthropic Claude via OpenRouter (claude-opus-4, claude-sonnet-4 and others); Google Gemini via OpenRouter or Gemini API; Groq for ultra-fast inference on Llama and Mistral; Ollama for 100% local use with no internet needed at all; and any OpenRouter model, 100+ models via one endpoint. You bring your own API key. AirTypes only stores it locally on your device, it is never sent to our servers.
No. AirTypes sees nothing. Your API key is stored locally on your device only, never uploaded to our servers. When My Agent runs, your text travels directly from AirTypes on your machine to your chosen AI provider. AirTypes acts as a local bridge, we never see the request, the response, or the content of your voice. This is architecturally enforced: the HTTP call to your AI provider is made by the AirTypes desktop app directly from your device, not via any AirTypes proxy or backend.
Linux and macOS are available now. Windows is in active development. Early subscribers who sign up today will get the Windows app the moment it ships, at the same $3.99/month price, with no extra charge.
Early access
Leave your email and we will let you know when the next AirTypes Pro build is ready, plus any early-access offer we open up. Transcription stays fully offline; nothing about your audio is shared.