Microphone
Required to capture your voice. Audio is processed locally and never stored or uploaded.
Runs 100% on your device
System requirements
WhisperJot transcribes locally on the Neural Engine — no cloud, no streaming. That means a little hardware does the heavy lifting. Here's what you need.
macOS
The primary, fully-supported platform.
| Component | Minimum | Recommended |
|---|---|---|
| Operating system | macOS 14 Sonoma | macOS 15 Sequoia or later |
| Processor | Apple Silicon (M1) Intel Macs not supported | M2 / M3 / M4 family |
| Memory (unified) | 8 GB | 16 GB+ required for local-LLM cleanup |
| Free disk space | ~3 GB app + one speech model | 10 GB+ room for multiple models |
| Microphone | Any built-in or external mic | |
| Internet | Only for the one-time model download — dictation works fully offline | |
Windows
Native C#/.NET port — in active development.
| Component | Minimum | Recommended |
|---|---|---|
| Operating system | Windows 10 (64-bit, 22H2) | Windows 11 |
| Architecture | x64 ARM64 not yet supported | x64 |
| Processor | Modern quad-core CPU | CPU + DirectML-capable GPU faster ONNX inference |
| Memory | 8 GB RAM | 16 GB+ RAM |
| Free disk space | ~3 GB | 10 GB+ |
| Runtime | Self-contained build — no separate .NET install required | |
Before you start
Permissions & good to know
Accessibility
Lets WhisperJot type the transcribed text into whatever app has focus.
Input Monitoring
Detects your global hotkey (including the fn key) from anywhere on the system.
About the model download
On first dictation, WhisperJot downloads a speech model — Parakeet ≈ 2.3 GB, or a smaller WhisperKit model — to your device. It's a one-time download; afterwards everything runs offline and launches instantly.
Models stay under their publishers' licenses: NVIDIA Parakeet (CC-BY-4.0), OpenAI Whisper (MIT), Silero VAD (MIT).
Optional local LLM
Extra text polish via MLX-LM or Ollama runs on 127.0.0.1.
Plan for 16 GB+ RAM if you enable it.
Specs check out?
Download WhisperJot