free, open-source, 100% local ai voice dictation. hold a key, talk, let go: clean text lands in whatever app you are in.
an open alternative to wispr flow. mit licensed. no account, no subscription, no cloud.
demo
what it does
- hold to talk
- hold the hotkey, speak, release. the text is pasted into any app with focus.
- ai cleanup
- drops fillers, fixes punctuation, formats lists, applies self-corrections ("no wait, tuesday").
- command mode
- a second hotkey edits the selected text: "make this more concise".
- per-app styles
- a different tone per app or site.
- dictionary
- learns the words you correct after a paste.
- multilingual
- whisper fallback for languages the default speech model does not cover.
- offline
- speech, cleanup and voice detection run on your machine. no telemetry.
- platforms
- macos, windows, linux. one codebase.
how it was built
by ai agents: pi on deepseek v4.1 flash, plus claude code.
status
the pipeline is verified end to end on a headless linux box with real models. the mac hotkey and paste path and the gpu backends are not verified on real hardware yet. no signed installer yet; build from source.
openflow is an independent project, not affiliated with or endorsed by wispr. "wispr flow" is a trademark of its owner.