localflow — dictation that never leaves this tab
Model:
base.en (~75 MB, faster)
small.en (~250 MB, more accurate)
Copy text
Clear
loading…
Loading transcription engine…
Whisper runs inside this browser tab (WebGPU, WASM fallback). Nothing is uploaded — verify in DevTools → Network. First use downloads the model into browser cache; offline afterwards. Hold the button (touch) or hold
Space
(desktop) to talk.