sotto voce · in an undertone
Dictation, kept to yourself.
Press a shortcut, speak, and Sotto types it at your cursor — in any app. Transcription runs on your computer: no account, no cloud, and audio is never saved.
Free · no account, no subscription · all releases & checksums
To: Elena
CtrlShiftSpace starts and stops — from any app
legato · one smooth motion
Press. Speak. It's typed.
No window to find, no button to click. Sotto waits in the menu bar or system tray and works wherever your cursor already is.
Press the shortcut
One global shortcut starts listening from any app. Change it any time in Settings; Escape cancels.
Say it how you'd say it
Speak naturally, then press the shortcut again. The speech model runs locally — it works on a plane.
It's typed at your cursor
Sotto copies the transcript, then pastes it where you were typing. If an app blocks pasting, the text is already in your clipboard.
pianissimo · as quiet as software gets
What leaves this computer
Nothing, by default. Sotto has no analytics, no crash upload, no update telemetry, and no account. Here is the complete list — including the two honest exceptions, both opt-in.
| Your data | Leaves? | Details |
|---|---|---|
| Audio | Never | Held in memory only and discarded the moment transcription finishes. Never written to disk. |
| Transcripts | Never* | *Unless you turn on optional AI cleanup with your own OpenRouter key — off by default. Text only, never audio; offline, you get the raw local transcript. |
| History | Never | Optional, stored locally, searchable, bounded, and clearable in one click. |
| Usage & telemetry | Never | None exists to send. No analytics, no crash reports, no phoning home. |
| Model downloads | Opt-in | The Standard model ships in the installer — nothing downloads on first run. The optional Multi-lingual model downloads from Hugging Face after you review its size and license; that request carries ordinary metadata like your IP, nothing more. |
tutti · everything included
Small app, complete instrument
About 500 MB installed, with a ready-to-use speech model in the box. No Python, no setup, no model download on first run.
Two modes, both local
Standard ships in the installer — instant English dictation. Add the optional Multi-lingual mode to dictate in other languages; it downloads only with your consent.
Streaming transcription
Sotto transcribes while you talk, so even long dictations finish almost immediately after you stop.
Personal dictionary
Teach Sotto the names, products, and jargon it should always get right.
Clipboard first, always
Every transcript is copied before pasting. A blocked paste never loses a word.
Local history
Optional and searchable, with retention you control and a clear-all that means it.
Quiet by design
Lives in the menu bar or tray, follows your light or dark theme, respects reduced motion, and can launch at login.
coda · take it home
Download Sotto
Free, no account. Formerly TalkType — your settings, history, and models migrate automatically on first launch.
Windows
Windows 10 or 11, 64-bit · about 500 MB installed · 1.5 GB free during install
Sotto-Setup-3.3.4.exe
Download installerUnsigned build: Windows may show a SmartScreen prompt. Choose More info → Run anyway. Installs per-user; no admin rights needed.
macOS
macOS 11 (Big Sur) or newer · Apple silicon only · about 500 MB installed
Sotto-3.3.2-arm64.dmg
Download disk imageUnsigned build: if the first launch is blocked, open System Settings → Privacy & Security → Open Anyway. Intel Macs are not supported.
Verify your download against SHA256SUMS.txt on the release page · all versions