KodaVoice is push-to-talk voice typing for Windows. Hold Right Ctrl in any app — your editor, browser, chat, or a Koda agent pane — say what you're thinking, release, and your words are typed at the cursor. Transcription runs entirely on your machine with Whisper: private, offline, no cloud round-trip.
Windows 10/11 · local Whisper transcription · works offline
One hotkey, every app. KodaVoice stays out of the way in your system tray and turns speech into text wherever your cursor is.
KodaVoice runs faster-whisper locally. Audio is recorded, transcribed, and discarded on your own hardware — there's no KodaVoice server, no account, and nothing to subscribe your microphone to.
Same principle as everything in the Koda ecosystem: local-first, and honest about what runs where.
Koda gives your coding agents a command deck. KodaVoice gives you a faster way to drive it — hold the key and speak your prompts, task cards, and replies straight into any agent pane instead of typing them. Two apps, one design language, one local-first philosophy.
No. Transcription runs entirely on your machine using faster-whisper. The speech model is downloaded once (~1.6 GB) on first launch, and after that KodaVoice works fully offline. Your audio is never uploaded anywhere.
Hold the hotkey (default Right Ctrl), speak, and release. A beep confirms recording started and stopped, a live meter shows your level, and a moment later your words are typed into whatever window has focus. Every transcript is also kept in the app's history panel. You can rebind the hotkey to any key in Settings.
Any reasonably modern Windows 10/11 PC works on CPU. For much faster transcription, flip on Use GPU in Settings — it supports NVIDIA (CUDA) and AMD / Intel (via Vulkan, no CUDA or ROCm install needed). The toggle only enables when a compatible GPU is detected.
Yes — KodaVoice types wherever your cursor is, and Koda's agent panes are ordinary terminals. Hold the key, describe the task, release, and your prompt lands in the pane ready to send. It's the fastest way to brief an agent.
Yes. The Vocabulary setting takes whole-word find→replace rules applied to every transcript — perfect for names, jargon, and acronyms the recognizer trips on (e.g. "clod" → "Claude"). You can also say "new line" or "new paragraph" while dictating to insert line breaks.
KodaVoice isn't code-signed with a paid certificate yet, so Microsoft Defender SmartScreen shows a one-time caution — the same one Koda shows. Click More info → Run anyway, once. Signed builds are on the roadmap.
Windows 10/11 today. KodaVoice installs per-user (no admin rights), lives in your system tray, can start with Windows, and updates itself — when a new release is out, an Update button appears and the app swaps itself over in one click.
Get KodaVoice and put a push-to-talk key under everything you write.
First launch: if SmartScreen appears, click More info → Run anyway, once.