All work
Case study
LocalWhisper
Voice-to-text for the Mac that runs entirely on your device. It works offline and keeps every word private.
- Type
- macOS app
- Our role
- Product, design, and engineering
macOSOn-device AIWhisperApple SiliconOffline
The problem
Most transcription tools upload your audio to a server. For a journalist protecting a source, a clinician taking notes, or a lawyer on a privileged call, that is a risk they shouldn’t have to take. Once audio leaves the device, you’ve lost control of it.
We wanted transcription that is accurate and fast, with no audio ever leaving the Mac.
What we built
LocalWhisper is a native macOS app that runs the Whisper speech model on your own machine. It uses the Neural Engine in Apple Silicon Macs, so it’s fast without a network connection.
100% local
All processing happens on the Mac. No network requests and no data collection.Fast
No upload and no network delay. Speed depends only on the Mac itself.Works offline
Airplane mode, bad Wi-Fi, or a deliberately disconnected machine. It keeps working.Pay once
No subscription and no usage limits. Running locally means no per-minute server costs to pass on.What it shows about how we work
- Private by design. Running the model on-device removed a whole category of security and compliance questions.
- Right-size the infrastructure. No servers means nothing to scale, patch, or pay for per user.
- Use modern AI where it fits. Open models on consumer hardware can now match cloud accuracy for many tasks. We help teams spot those chances.
Want the product story? Read the launch post.