A voice-dictation app that never touches the cloud

Dictation on macOS almost always means handing your voice to a third party. For anyone writing something private — notes, messages, code comments — that's a real cost for a small convenience. The brief we set ourselves: build the fast, cloud-grade dictation experience people already expect, but keep every byte of audio on the device it was recorded on.

Hush lives in the menu bar and does its real work through a single global hotkey — hold to talk, or toggle to dictate hands-free. There's no window to manage and no workflow to learn: record, speak, release, and the text is already where the cursor was. The interface is the absence of one.

WhisperKit runs entirely on the Neural Engine, so transcription happens locally at speed — no upload, no latency waiting on a server, no data ever leaving the Mac. Six model sizes are bundled, from fast-and-light to large-and-precise, covering twelve-plus languages, so users pick their own trade-off between speed and accuracy.

Microphone and Accessibility access are non-negotiable for a tool like this, so we built a 4-step onboarding that explains exactly why each permission is needed before macOS asks for it — trust before the prompt, not after. If Accessibility is ever denied, auto-paste degrades gracefully to the clipboard instead of failing silently.