SPEACH

Talk. Don't type. Hold a global hotkey anywhere on your machine, speak, and release — your words are transcribed right on your device and dropped into whatever field has focus: a chat box, a code editor, an email, an LLM prompt. Free, open source, and private by default.

What it does

• Hold-to-talk, anywhere — press and hold your hotkey in any app, speak, release. The text lands in the focused field via clipboard-paste, with a keystroke-typing fallback for apps that refuse synthetic paste.
• Transcribes files too — drop in audio or video and export a clean transcript as .txt, .srt, or .vtt; translate non-English audio to English in one step.
• Ready in five minutes — a tiny Whisper model ships bundled, so the first transcription works offline immediately. No terminal, no Python, no model-wrangling — a guided first-run grants the two permissions it needs.
• As accurate as you want — optional beam-search decode, a custom-vocabulary field, a live caption while you speak, and an on-device (or bring-your-own-key cloud) AI proofread — every dial off by default.

Private by architecture

The default engine runs whisper.cpp in-process inside a native Rust core — no separate process, no network surface at all. Your audio never leaves the machine; the cloud proofread is opt-in, BYO-key, and clearly disclosed. It's a lightweight Tauri app, not a heavy Electron shell, so it idles quietly in your tray all day.

Availability

SPEACH runs today on macOS (Apple Silicon); Windows is built in code and planned. It's free and open source. The public download isn't ready yet — check back shortly.