Open-source speech-to-text that runs entirely on your Mac. Hold a key, speak, release — text appears at your cursor. 100% local. No cloud. No subscription.
Get OpenWhisper on GitHubOpenWhisper uses WhisperKit to run OpenAI's Whisper model directly on your Mac's Neural Engine. Fast, accurate, and completely private.
All speech recognition runs on your Mac. No audio is ever uploaded to any server. Your voice data stays on your machine.
After a one-time model download, OpenWhisper works without any internet connection. Transcribe speech to text anywhere.
Hold the Right Option key, speak, and release. Your transcribed text appears instantly at the cursor. Simple and fast.
VS Code, Terminal, Chrome, Slack, Notes, Pages, Word — OpenWhisper pastes text anywhere you can type on macOS.
Optional local LLM via Ollama removes filler words like "um" and "uh", fixes grammar and punctuation — all on-device.
English, Spanish, French, German, Hindi, Chinese, Japanese, and 22 more languages with automatic language detection.
See how OpenWhisper compares to cloud-based speech-to-text services like Wispr Flow, Otter.ai, and others.
| OpenWhisper | Cloud Services | |
|---|---|---|
| Privacy | ✓ 100% local — nothing leaves your Mac | ✗ Voice uploaded to remote servers |
| Internet | ✓ Works offline | ✗ Requires internet |
| Cost | ✓ Free and open-source | ✗ $10-20/month subscriptions |
| Latency | ✓ Instant on-device processing | ✗ Network round-trip delay |
| Resource usage | ✓ Under 100 MB RAM when idle | ✗ 400-800 MB RAM |
| Data collection | ✓ None | ✗ Voice data stored on third-party servers |
| Accuracy | ✓ OpenAI Whisper (state-of-the-art) | Varies |
Build from source in under two minutes. Requires macOS 14.0+ and Apple Silicon.
Download the OpenWhisper source code from GitHub to your Mac.
Run the build script. The first build downloads dependencies and takes about two minutes.
Open the app, grant microphone and accessibility permissions, and start dictating.
Common questions about OpenWhisper, the free voice-to-text app for macOS.
Yes. OpenWhisper is completely free and open-source under the MIT license. There are no subscriptions, no trials, no in-app purchases, and no hidden costs. You can use it, modify it, and distribute it freely.
No. All speech recognition happens locally on your Mac using WhisperKit, which runs OpenAI's Whisper model on Apple's Neural Engine. No audio data ever leaves your machine. The only network calls are a one-time model download and optional Ollama API calls on localhost.
Yes. After the initial one-time model download (around 140 MB for the base model), OpenWhisper works completely offline. No internet connection is needed for speech-to-text transcription.
OpenWhisper uses OpenAI's Whisper model via WhisperKit, which is one of the most accurate speech recognition models available today. The base model handles everyday dictation well, and the small model provides even higher accuracy for longer passages and multilingual content.
No. Ollama is entirely optional. It provides AI-powered grammar cleanup that removes filler words and fixes punctuation. Without Ollama, OpenWhisper still transcribes perfectly — you simply get the raw Whisper output without post-processing.
macOS built-in dictation sends your audio to Apple's servers for processing. OpenWhisper processes everything locally on your Mac, offering stronger privacy, full offline support, multiple model choices, and optional LLM-powered grammar cleanup. It also works consistently without depending on Apple's server availability.