Hold a key. Say the thing. Cleaned-up text lands at your cursor in any app — transcribed, polished, and typed entirely on your machine. No cloud. No subscription. No excuses.
Browser demo — hold, then actually speak. Uses this page's mic + your browser's speech recognition to show the loop live. The Speakr app itself never sends audio anywhere.
Unzip, then: run.bat on Windows · bash package_mac.sh on Mac. First run downloads the model — that's the only network call it ever makes.
Hold. Speak. Done.
Right Ctrl on Windows, fn on Mac. Release the key and clean text appears at your cursor — fillers stripped, self-corrections applied, lists formatted.
Zero cloud. Period.
Whisper transcription and LLM polish run entirely on-device. The only network call, ever, is the one-time model download. Audit the code — it's all there.
It learns you.
Your names, your jargon, your CamelCase — Speakr picks up recurring vocabulary and spells it right, getting sharper the more you talk. Locally, obviously.