You can speak an email or a paragraph and insert the text into the Mac app where you are already writing. OGAD (Off Grid AI Desktop) Pro provides local dictation through its Voice feature. Download a speech model, enable Paste at cursor, and use the dictation shortcut. The transcription can run without internet after setup.
| Get OGAD for Mac | Desktop releases |
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.
The workflow is useful when you can say a sentence faster than you can type it: a project update, a rough paragraph, or a reply that needs more than one line. The text goes into the target app so you can edit it there.
“Any Mac app” means an app with a text field that accepts a normal paste. A locked document, read-only view, or field that blocks pasting can still prevent insertion. Start in a blank TextEdit document to check the setup before trying your usual editor.
What do you need for local Mac dictation?
You need OGAD Pro on a supported Mac, a downloaded local transcription model, microphone access, and Accessibility permission for automatic insertion. The published desktop requirements include Apple Silicon and macOS 13 or later. Complete installation, Pro setup, and model downloads while connected.
This guide uses the Voice feature in the 0.0.51 desktop release. It is separate from the microphone inside a chat composer: that microphone enters text into the chat, while Voice dictation can paste into another app.
| Requirement | What it does |
|---|---|
| Local transcription model | Converts the recording into text on the Mac |
| Microphone access | Allows OGAD to record your speech |
| Accessibility permission | Supports the global hold shortcut and automatic paste |
| Editable field in the target app | Receives the completed transcript |
A large chat model is not required just to turn speech into text. The basic dictation path uses the speech model. Optional memory processing is a separate setting.
How do you prepare the speech model?
Open Voice, then Voice settings. Under Transcription, choose Whisper as the engine for this first setup. Download a model from the model list. If a suitable model is already installed, choose Use and check that it shows Active.
You can also open Models to manage speech-to-text downloads. Wait for the model to finish downloading before starting a recording.
Choose Language to match what you plan to say. Auto-detect is available, but an explicit language removes one source of confusion for a short sentence. For languages other than English, use a multilingual model rather than an English-only variant.
The settings also offer another transcription engine. Use it only after its model and runtime are ready. The instructions here use one engine so the first check has fewer moving parts.
How do you make the words appear in another app?
In Voice settings, enable Paste at cursor, leave Auto-send off, and choose an activation mode. Then click the destination text field, start dictation, speak, and stop. OGAD transcribes the recording and pastes the final text into the app you started from.
1. Set the output behavior
Use these settings for a first attempt:
- Paste at cursor: on — inserts the completed text.
- Append a space: on — leaves room for the next dictated phrase.
- Auto-send: off — lets you review the text before submitting anything.
- Feed to memory: off — keeps this first workflow focused on recording and inserting text.
Auto-send presses Enter after pasting. In a messaging app, Enter may send the message. In a document, it may add a new paragraph. Leave it off until you know how the destination handles that key.
2. Choose when the microphone runs
Under Activation > Mode, choose one of these controls:
| Mode | What you do |
|---|---|
| Hold | Hold the shortcut while speaking; release it to stop |
| Toggle | Press the shortcut to start; press it again to stop |
| Both | Tap to toggle or hold for push-to-talk |
The default shortcut is Option+Space, and the default mode is Hold. Toggle can be easier for a longer paragraph because you do not need to keep the keys down.
3. Allow the Mac permissions
Allow microphone access when asked. For automatic paste and the hold shortcut, allow OGAD in System Settings > Privacy & Security > Accessibility. If macOS asks you to quit and reopen the app after a permission change, do so.
You may also see a macOS request related to controlling another app when the paste action first runs. Follow the system prompt for the feature you are using.
4. Start in a blank document
Open TextEdit and create a blank document. Click inside it so the insertion cursor is visible.
Start dictation with your selected shortcut. Wait for the visible recording indicator before speaking. Try:
The first draft is ready. I will check the figures before I send it.
Stop with the control for your chosen mode. Wait for the final transcript to appear before clicking another field or starting a new recording.
5. Review the result
Check that both sentences arrived in the intended place. Correct any missing word or punctuation in TextEdit. Recognition depends on the recording and model; the example is a sentence to test, not a promise of exact output.
Then repeat the same short check in your email editor, notes app, or document tool. A successful TextEdit paste establishes the basic workflow. A failure in one other app points you toward that app’s field or focus behavior.
How do you use push-to-talk for short writing bursts?
Choose Hold in Voice settings. Click the destination field, hold Option+Space, wait for the recording indicator, and speak one complete sentence. Release the keys after the last word. Wait for final transcription and paste, then read the text before starting the next turn.
This lets you pause to think between recordings without a silence timer cutting a sentence short. Draft an email in three turns: the current status, the open question, and the next action. Review names and figures after each insertion.
If holding the keys gets tiring, choose Toggle for a longer paragraph. If releasing the keys does not stop recording, first check that Hold is selected. A missing native shortcut helper or Accessibility permission can make the shortcut fall back to toggle behavior. Check the visible recording state and use the recording control to stop.
To change the shortcut, click Shortcut and press another chord. Escape cancels shortcut selection; it is not a documented universal cancel key for a live recording.
How do you improve names and rough spoken text?
Add unusual names and terms under Custom words in Voice settings. They give the recognizer useful vocabulary for a transcript. They do not guarantee spelling, so still review names, numbers, dates, and addresses before you send the result.
For a first check, include one term you use often: a project name, a colleague’s name, or a technical abbreviation. Compare the transcript with the intended word. Add the correct spelling if needed, then try another short recording.
Remove filler words can make a cleaned copy by removing supported fillers and stutters. Paste cleaned text selects that copy for insertion. The raw transcript remains in the Voice library.
This is a deletion-based cleanup feature. Do not expect it to rewrite an argument, add facts, or turn an unfinished thought into a polished email. Edit the text in the destination app when you need those changes.
What happens if automatic paste fails?
If the automatic paste cannot run, the transcript is retained on the clipboard for a manual Command+V. You can also open Voice, find the recording, and use Copy transcript. You do not need to record the same sentence again just because insertion failed.
| Symptom | Check | Action |
|---|---|---|
| No recording starts | Microphone access and model | Allow access and activate a downloaded speech model |
| Shortcut opens another feature | Shortcut conflict | Choose another chord in Voice settings |
| Hold behaves like a toggle | Accessibility or shortcut helper | Check permission; use Toggle while resolving it |
| Text is saved but not inserted | Paste at cursor and Accessibility | Enable the setting, allow permission, or paste manually |
| Text goes to the wrong field | Cursor position | Click the intended field before starting and keep it selected |
| A name is wrong | Language and vocabulary | Set the language, add the name to Custom words, and review |
| The second recording will not start yet | Final transcription is still running | Wait for processing to finish |
The paste operation temporarily uses the clipboard. It attempts to restore prior text after a successful paste. That does not preserve every clipboard format: an image or file you copied should not be assumed to survive the operation.
Does the audio leave your Mac?
The local dictation engine processes the recording on the Mac. Prepare the model first, then turn off Wi-Fi and disconnect any other network connection to check the same short TextEdit workflow offline.
The destination app has its own behavior. Dictating into a cloud document or an email app does not make that app local. It may sync the inserted text when connected.
Voice also keeps a recording history. Use Keep and Auto-delete in Voice settings to choose its limits. Turn off Feed to memory if you do not want dictation to feed the separate memory workflow. Local processing and saved history are different choices.
Mac dictation questions
Is this the same as the free chat microphone?
No. This guide uses the Pro Voice feature to insert text into other apps. Chat dictation fills the chat composer.
Can I save speech without pasting it anywhere?
Yes. Turn off Paste at cursor and use the Voice library to review the transcript later.
Do I have to keep the model loaded?
No. Keep model in memory is optional. It keeps the speech model available between uses, but it also keeps memory in use. Choose based on how often you dictate and the other work on your Mac.
Will it send messages for me?
Only if you enable Auto-send, and the target app treats Enter as Send. Leave that setting off when you want to review each message.
Start with one paragraph in TextEdit. Once the words arrive where you expect, use the same shortcut in the app where you normally write.