Speak. It types.
Local AI voice dictation for Windows and Mac. Your audio never leaves your machine. Press F8, talk, done.
14 days free with just an email, then $9 / month.
- Runs on your machine
- Works offline
- Detects your language
- Uses an NVIDIA GPU on Windows, Apple Silicon on a Mac
Press F8. Talk. Press F8.
The bar shows it is listening. The words appear where your cursor was.
-
F8
Start
The bar turns on. Talk at your normal pace. On Auto it hears English or Spanish; any other language, pick it in Settings.
-
F8
Stop
It transcribes on your computer. Your audio never leaves it.
-
→
Typed
The text lands in the app you were in: an email, a chat, a terminal, a coding agent. It stays on the clipboard too.
What it does
-
Fully local
Your voice never leaves your machine. Transcription makes no network call, and nothing you say is stored in a cloud.
-
GPU-accelerated
Whisper on an NVIDIA GPU on Windows or on Apple Silicon on a Mac, and on the CPU otherwise. Four models, Base to Large v3.
-
Works everywhere
Discord, Slack, VS Code, browsers, terminal, any native app.
-
English or Spanish, detected
Auto chooses between English and Spanish. For another language, pick one of twelve in Settings.
-
CPU fallback
Without an NVIDIA GPU or Apple Silicon it runs on the CPU, more slowly.
-
AI polish (opt-in)
Optional post-transcription cleanup via local Ollama (default) or OpenAI. Cleans filler, fixes punctuation, respects your custom dictionary.
-
Searchable history
Local, searchable log of every transcription. Copy or delete any entry; nothing is synced.
-
Dictionary and snippets
Teach Whisper your proper nouns and brand names. Map short triggers to longer expansions. Your signatures, boilerplate, code templates — dictated in one word.
Dictate prompts to your coding agent.
Using Cursor, VS Code, Claude Code, or Codex? Stop typing context. Press F8, describe the change, press F8.
-
Cursor / VS Code
Voice-prompt your agent
Click into Cursor's chat. Press F8. Say what you want changed. Press F8. Agent gets a precise, detailed prompt without breaking your thought.
"Refactor this function to handle null edge cases and add a unit test for the happy path and the null case."
-
Claude Code
Think out loud, commit fast
Running Claude Code in the terminal? Press F8 between agent steps. Describe your intent, requirements, context.
"Add error handling for the database timeout case and log the error with the request ID."
-
Any app · Any context
Documentation without dread
Dictate commit messages, PR descriptions, Notion docs, Slack updates. WhisperType works in every text field, not only in an IDE.
"Fix: resolve null pointer exception when user session expires during file upload."
Your audio goes nowhere.
Most dictation apps send your voice to their servers. Wispr Flow's own privacy page says its transcription always happens in the cloud. WhisperType runs the model on your machine, and the transcription happens there. No audio and no transcript is sent to us.
- Other apps Your voice Internet Their servers Text back to you
- WhisperType Your voice Your computer Text to your cursor
What stays on your machine
- No audio transmitted. Your voice is processed on your machine. It never touches our servers or anyone else's.
- Transcripts stay on your machine. Your history is saved locally, where you can search it and choose how long it is kept. It is not sent to us, unless you pick “Full — include transcripts” for usage statistics.
- Statistics without your words, by default. Settings → Privacy → Usage statistics has three levels. Off sends nothing. Anonymous stats (no text), the default, sends counters: timings, model, detected language, word and character counts, success or failure. Full — include transcripts also sends the text of each dictation, and is on only if you pick it. Audio is never sent. Every field.
What we collect
- Account email
- to start your trial and deliver your subscription license.
- Anonymous crash reports
- opt-in only, no audio, no transcripts — just stack traces
- Anonymous usage statistics
- on by default: speed, model, language, word counts — no text; off in Settings
- Your voice recordings
- Never.
- Your transcripts
- Never, unless you pick “Full — include transcripts” for usage statistics in Settings.
Where each app sends your audio.
| WhisperType | Wispr Flow | Windows Built-in | |
|---|---|---|---|
| Audio stays local | Yes | No. Cloud only | No. Azure cloud |
| Platforms | Windows, macOS (beta) | Mac, Windows, iOS, Android | Windows |
| Works offline | Yes | No | No |
| Price | $9/mo · 14-day free trial | Free to start, then $15/mo ($12/mo annual) | Included with Windows |
| GPU acceleration | Yes. NVIDIA CUDA, Apple Metal | No. Cloud | No. Cloud |
Wispr Flow's privacy page says transcription always happens in the cloud. Microsoft says Windows voice typing (Win+H) uses online speech recognition powered by Azure and needs an internet connection. Checked on their own pages, 4 October 2026.
14 days free, then $9 / month.
Billed monthly · cancel anytime · 1 active machine
- NVIDIA GPU on Windows, Apple Silicon on a Mac
- Every model, Base to Large v3
- Detects your language
- All features, all future updates
- 14-day free trial — no card, just an email
No contracts, no annual lock-in. One price, every feature, every future update, and no credit card to start.
- What happens after the trial?
- The app locks until you subscribe. Subscriptions open in a few days. Nothing is charged until you subscribe.
- Can I cancel anytime?
- There is nothing to cancel yet: nobody is charged. Subscriptions open in a few days; the form above emails you the day they do.
- One machine or more?
- One active machine per subscription. Moving to a new laptop? Deactivate the old one, activate the new one.
Download WhisperType.
Download the installer for your platform. It's not yet code-signed, so Windows may show a SmartScreen warning ("More info" → "Run anyway") and macOS blocks the first launch until you choose "Open Anyway" in System Settings → Privacy & Security. Both are safe. Your 14-day free trial starts on first launch — no credit card, just an email.
How installing works on a Mac
-
1 · Download
Open the .dmg
Drag WhisperType into Applications. Then eject the disk.
-
2 · First open
macOS stops it once
On macOS 15 and later it says Apple could not verify WhisperType. Click Done, not Move to Trash. On macOS 14 or earlier you can instead right-click WhisperType in Applications and choose Open.
-
3 · Allow
Open Anyway
System Settings → Privacy & Security → scroll down → Open Anyway. Once for each version you download.
-
4 · Set up
Three permissions
Setup asks for Microphone to hear you, Accessibility to type the text and Input Monitoring to see the F8 key. Updating from 1.3 asks for the macOS permissions one last time; later updates keep them.
Not sure it will run? Check this computer. You can verify the SHA-256 hash in the changelog.
What you need to run it.
- OS
- Windows 10 or 11, 64-bit, Intel or AMD · Apple Silicon Mac: version 1.4.1, macOS 14 or later · Intel Mac: version 1.3.2, macOS 11 or later, slower (macOS is a public beta)
- GPU (optional)
- NVIDIA with the CUDA 12 libraries, or Apple Silicon
- No GPU?
- Runs on the CPU · slower
- Disk space
- 75 MB to 3 GB for the model you choose on first run
- macOS
- Public beta
Questions and answers
Does my audio go to the cloud?
No. Never. Your voice is processed on your machine by a Whisper model running locally. No audio and no text leave it. The app does go online for license checks, updates and usage statistics, and none of them carry your audio or your words.
Does it work without internet?
Yes, after the one-time model download on first run (75 MB to 3 GB, depending on the model you choose). Transcription itself is always local — no network call is made while you're recording or typing. WhisperType is a subscription, so your license is checked periodically when you're online, with a short grace period if you're not, but that check never touches your audio or your text.
What GPU do I need?
On Windows: an NVIDIA card with CUDA support; AMD and Intel GPUs are not used. On a Mac: Apple Silicon, through Metal. No GPU? WhisperType runs on the CPU — same features, just slower.
How is this different from Windows Speech Recognition (Win+H)?
Three things: (1) Windows Voice Typing uses Microsoft's cloud — your audio goes to Azure. WhisperType stays local. (2) WhisperType works offline. Win+H needs an internet connection. (3) You choose the Whisper model, from Base to Large v3.
Is there a Mac version?
Yes, macOS is a public beta, out the same day as Windows. It has the same trial and subscription. Transcription is fast on Apple Silicon. Intel Macs run version 1.3.2, and it is slow there. Updating from 1.3 asks for the macOS permissions one last time; later updates keep them. Get in touch if you hit something broken — you'll be first to know when Mac reaches a signed 1.0.
What's the cancellation and refund policy?
Nothing is charged yet, so there is nothing to cancel. Subscriptions open in a few days. The 14-day trial needs no credit card. Full details: Refund & Cancellation Policy.