hushquill.com / offline dictation
Most voice typing tools are thin clients for a server. Lose the connection and they lose their voice. HushQuill keeps the whole speech engine on your machine, so offline is not a degraded mode. It is just Tuesday.
Why apps go silent
Windows voice typing (Win+H) sends your audio to Azure Speech services, and Microsoft's documentation is explicit that it needs an internet connection to work. Cloud subscription apps like Wispr Flow are built the same way. This is not carelessness, it is architecture: the model that turns sound into text lives in a data centre, so your audio has to travel there and the text has to travel back.
HushQuill inverts that. The Whisper and Parakeet models sit on your disk and run on your CPU. Turn the WiFi off mid-sentence and the sentence still types. Nothing to reach, nothing to wait for, nothing leaving the machine.
Where it matters
Draft the whole report at 11,000 metres. Dictation does not know the plane has no WiFi and does not care.
Client sites, labs, government floors where machines must not talk to the internet. A cloud dictation app is simply not installable policy-wise. A local one is.
Legal drafts, patient notes, unreleased numbers. Offline capable means the audio path physically cannot include a third party.
Hotel WiFi, tethered 4G, rural connections. Local processing has no latency spikes and no dropped takes, because there is no round trip.
Setup
One 8.4 MB installer on Windows, a DMG on Mac. No account, no sign-up, no card.
On first run the app fetches the models you choose, so you decide what lands on disk and when. This is the only step that needs a connection.
From then on, everything runs locally: transcription, punctuation, your personal dictionary, snippets and voice commands. Updates arrive when you are back online, on your schedule.
Questions
Yes. Speech recognition runs on your own CPU using Whisper and Parakeet models stored on your disk. After the one-time model download, you can turn the WiFi off and dictation keeps working: transcription, punctuation, your dictionary, snippets and voice commands all run locally.
Three things: the first-run download of the speech models you choose, app updates, and the optional cloud engines, which are off by default and only activate if you opt in and paste your own API key. Day-to-day dictation needs no connection.
Because their speech engine runs on the vendor's servers. Windows voice typing (Win+H) uses Azure Speech services and Microsoft documents that it requires an internet connection. Cloud subscription apps work the same way. If the engine is remote, no connection means no dictation.
Modern local models closed most of the old gap. HushQuill picks the fastest engine for your language and hardware automatically, and you can choose a bigger model when you want accuracy over speed. An optional cloud engine exists for people who want it, but it is opt-in with your own key, never the default.
Try it
The free build is the whole dictation engine. No account, no trial clock, no word cap.
Windows 10 and 11, macOS 11 and later. Free forever, no telemetry.