DictationDesk docs
1. Install and sign in
- The Windows app is coming soon: it is in private testing. Get notified. The steps below describe how it works.
- DictationDesk sits in the system tray. Click the tray icon to open settings.
- In the web dashboard open Devices → New token, copy the
sk_dev_…token, paste it in the app under Account, then press Sync. - No account? The app still works fully offline in English.
2. Hotkeys
| Ctrl+Win (hold) | Dictate. Release to type. Pressing another key while holding cancels. |
| Ctrl+Win+Alt (hold) | Command mode: select text first, then say “make this formal”, “shorter”, “translate to Hindi”… |
| Settings → Hold to talk off | Press once to start, again to stop (long dictation). |
| Settings → Dictation hotkey | Any shortcut such as Ctrl+Shift+Space, or a modifier combo such as Ctrl+Win. (macOS is planned, no date yet.) |
3. Speaking tips
- Say
comma,full stop,question mark,new line,new paragraphfor explicit punctuation. - Correct yourself naturally: “at five, no wait, at six” → “at six”. “Scratch that” deletes the last sentence.
- In code editors:
camel case user name→userName,snake case,open paren,equals,dot. - Emails get a greeting line and a sign-off block automatically; chat messages drop the trailing full stop.
4. Languages and scripts
English today. Indian languages, Hinglish and speak-Telugu-get-English are built and coming soon, billed at Sarvam's cost.
5. Privacy
Surge Software Solutions Pvt Ltd's company-wide policy: surgesoftware.co.in/privacy. This section explains how DictationDesk applies it.
- No screenshots, ever. The app only reads the name of the focused app (e.g.
slack.exe) to choose formatting. - Audio is processed in memory and never stored, on our servers or on your PC.
- Text is not stored on our servers unless you turn on Cloud history. Local history (last 500) can be turned off.
- Offline only mode (tray menu) keeps everything on your PC using whisper.cpp.
- Cloud speech-to-text and AI polish run on our own servers in Bengaluru. If they are down, a fallback provider is used, which may be outside India (Anthropic for AI polish). Indian-language speech (when switched on) is processed by Sarvam AI. Emails are sent through Resend.
6. Offline mode
The Windows app (coming soon) includes whisper.cpp with the English base model (148 MB on disk, used only while transcribing). More models (multilingual small, more accurate small.en, tiny tiny) can be downloaded from Settings → Offline model. Offline dictation is free and unlimited.
7. API (for the desktop app)
POST /api/v1/transcribe: multipartaudio(WAV) +lang,mode,script,context,polishPOST /api/v1/command:{instruction, text}GET /api/v1/settings: dictionary, snippets, language, cloud hours left, creditPOST /api/v1/usage: offline word count (no text)- Auth:
Authorization: Bearer sk_dev_…