Voicy
Open-source macOS dictation utility: hold a global shortcut, speak, and text appears in the app you were using. Built with Tauri 2, React, TypeScript, Rust, and whisper.cpp - all inference stays on-device. No cloud, no audio uploads. MIT licensed at github.com/vishuRizz/voicy.
Client / Type
Personal / Open Source
Stack / Services
Tauri 2, React, TypeScript, Rust, whisper.cpp
Hold to talk, release to type
Hold a configurable global shortcut (default Option+Space), see a live provisional transcript in an always-on-top overlay, then release to finalize and insert text into whatever app had focus. Menu-bar app with settings for shortcut, Whisper model, and language. Cancel with Esc anytime.
Local Whisper · Rust core
Microphone capture via cpal (16 kHz mono PCM), rolling-window preview every ~1s, and a full-utterance pass on key-up. Rust owns the state machine (IDLE → LISTENING → FINALIZING → INSERTING). Session UUIDs drop stale inference. Audio stays in RAM only - never written to disk or sent to a server.
Privacy-first by design
All transcription runs on your Mac with whisper.cpp. No cloud API keys, no telemetry of transcript content. Clipboard-paste fallback for insertion today; Accessibility-based direct insert is on the roadmap. Windows support planned as a later phase.