Speak a paragraph and place the text in the app where you are writing. OGAD (Off Grid AI Desktop) Pro beta 114 includes local dictation on Linux. Prepare a speech model, set Paste at cursor, and check one short sentence before using the shortcut for longer work.
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
What do you need?
Use the x64 beta package for your platform with Pro active. Prepare a working microphone and a downloaded local transcription model. Dictation uses the speech model; a large chat model is not required just to recognize speech.
Initial app and model downloads need internet. With the local speech engine ready, the transcription can run on your computer without cloud AI processing. The app where you paste may have its own network behavior.
On Linux, global shortcut and paste support depend on the desktop session and available native helpers. Wayland restrictions can affect app focus and simulated input. Check the full route in the app where you plan to write.
Prepare one local transcription route
- Open Voice, then Voice settings.
- Under Transcription, choose Whisper for this first check.
- Download a suitable model, select Use, and confirm it is active.
- Set Language to the language you plan to speak, or use automatic detection with a multilingual model.
- Enable Paste at cursor and leave Auto-send off.
- Under Activation, choose Toggle for a simple first attempt and inspect the configured Shortcut.
Use a multilingual model for languages other than English. An English-only model does not become multilingual because you changed the Language control.
The activation choices include Hold, Toggle, and Both. Hold records while you keep the shortcut down. Toggle starts and stops on successive presses. Begin with the mode that is easiest for you to control, then check the visible recording state.
Dictate into a harmless document
Open a text editor and click a blank editable field. Start dictation with the configured shortcut and wait for the recording indicator. Say:
Please review the draft on Thursday. I will add the final figures before lunch.
Stop with the control for your selected mode. Wait for transcription and insertion to finish before selecting another field. Read the words and correct the result.
You should see the paragraph in the intended field. If it appears elsewhere, check focus before trying again. A recognized sentence and a successful automatic paste are separate results.
For a longer note, dictate one complete thought at a time. Review names, dates, and numbers before adding the next passage. A quiet microphone recording is more useful than repeating a long paragraph over background noise.
If the text is recognized but not pasted
Check Paste at cursor, the destination field, and the native helper or session support. If automatic insertion fails, use the saved transcript in Voice and Copy transcript, then paste it manually into the intended app.
You do not need to record the sentence again merely because insertion failed. Retain the recognized text and resolve the paste route separately.
If the shortcut opens another feature, choose a different chord in Voice settings > Shortcut. Escape cancels shortcut selection. Do not assume it is a universal stop key for a recording. Use the visible stop control when needed.
Improve words that matter to your work
Add unusual names and terms under Custom words. They provide vocabulary to the recognizer; they do not guarantee a correct spelling. Check an uncommon client name before you send a message.
Try the same short recording with a different downloaded speech model when recognition remains poor. Compare the actual result on your speech. A larger download does not guarantee a better result for every microphone or language.
Choose what is retained
Voice keeps a recording history. Use Keep and Auto-delete in Voice settings to select its limits. Feed to memory is a separate choice for adding spoken context to the memory workflow.
Leave Auto-send off while checking the text. When enabled, it presses Enter after insertion; the destination app decides what Enter does. Correct a draft before you choose to send it.
This guide covers OGAD beta 0.0.55-beta.114. The release and build record identify the desktop version and its included Pro code. Setup details follow that code; they are not a device test.
Download OGAD and check one sentence in your normal writing app. Once recording, recognition, and insertion work, use the same route for a paragraph.
Top comments (0)