Local Whisper DictationState-of-the-Art ASR on Your GPU.
Harness open-weights OpenAI Whisper models on your own machine. Sub-200ms latency, 99 supported languages, and 0 bytes transmitted over the network.
The Problem with Cloud Dictation
Raw Whisper scripts require manual Python environments, lack global desktop hotkey hooks, struggle with real-time streaming latency, and don't provide automatic app formatting.
Physical Hardware Sovereignty
HushWrite wraps whisper.cpp in a high-performance native Rust desktop application, providing an instant floating pill UI, custom phonetic biasing, and background warm VRAM states.
Local Architecture vs. Cloud SaaS
| Buyer Concern | HushWrite (Local-First) | Competitor / Cloud Architecture |
|---|---|---|
Inference Stack Zero cloud API costs and zero network latency | Native whisper.cpp with Metal/DirectML | Cloud API wrapper (OpenAI Whisper API) |
Custom Vocabulary Biasing Flawless accuracy on proprietary jargon | Local phonetic dictionary steers beam search | Generic prompt prefixes or none |
Background Warm State Dictation starts the instant you press the hotkey | VRAM model persistence (<5ms wakeup) | Cold start delays or server queueing |
License & Ownership Inspect, audit, and modify the code freely | 100% Open Source (MIT) | Proprietary closed-source SaaS |
Open source beats closed cloud SaaS on speed and sovereignty.
Why pay OpenAI or cloud wrappers per-minute API fees when you can run optimized Whisper models locally with zero marginal cost?
“Whisper.cpp compiled into a native Rust desktop app with universal hotkey injection. This is how local AI should be built.”
How to Verify This Claim Yourself
Clone the repo and run `cargo bench`. Benchmark Whisper Base Q5_0 decode latency on your own hardware.
Frequently Asked Questions
Which Whisper model sizes are supported?
HushWrite supports Tiny, Base, Small, Medium, and Large-v3-Turbo in quantized Q4_0, Q5_0, and F16 formats.
Can I bring my own fine-tuned Whisper model?
Yes. HushWrite supports loading custom GGUF and whisper.cpp compatible model weights.
Ready to dictate without cloud exposure?
Download HushWrite for macOS and Windows. Free, open source, and 100% on-device.