Your voice should not need an upload
Most modern dictation tools send your audio to cloud servers for transcription. That is a fine engineering choice - and someone else's computer listening to everything you say.
How cloud dictation works
The popular dictation apps of the last few years share an architecture: your microphone audio streams to the vendor's servers, a large model transcribes it there, and the text comes back. It works well. It also means your voice - meetings, drafts, half-formed thoughts, names of clients - transits and often persists on infrastructure you do not control, under terms that can change.
How MyQuietShadow works instead
- Local speech recognition. Whisper-class models run on your own machine. The audio never leaves it.
- No training on you. Your voice cannot become training data, because we never receive it.
- Push-to-talk by default. The mic engages when you double-tap Ctrl - not before. No always-hot microphone.
- Works offline. On a plane, on a locked-down network, in a client's office - dictation does not care.
- Any app. The words land in whatever has focus. No per-app integrations to wait for.
The trade-offs, honestly
Cloud models can be bigger than what fits on a laptop, and for some accents and edge cases the biggest models still win. Local recognition has closed most of that gap and keeps closing it - and we think "very good and private" beats "slightly better and listened to" for the words you speak at your own desk all day. MyQuietShadow is in early access; dictation is the first thing we ship because it is the thing we use ourselves, every day.
Read our privacy stance - it is the whole reason this product exists.