OGAD (Off Grid AI Desktop) and OGAM (Off Grid AI Mobile) can read an assistant reply aloud using a voice model on your device. Download the voice first, enable text-to-speech, and select Speak on a completed reply. To create new answers without internet too, select a downloaded local text model. Text generation and spoken playback are separate steps.
| Get OGAD | Get OGAM |
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.
These steps cover Android, iPhone, and Apple Silicon Mac. Local speech output on Windows is not covered by this guide.
You can keep the written answer on screen and listen when it helps. This is useful for a short explanation or a checklist you want to review without reading every line. You do not need to record your own voice to use it.
What must be downloaded for offline spoken replies?
Download a local text model for new answers and a local voice model for playback. Complete any resources required by the selected language and speaker. Once those files are ready, you can type a question, receive a local answer, and hear that answer without a cloud speech API.
| Stage | What it needs |
|---|---|
| Type a question | The chat interface; microphone access is not required |
| Generate an answer offline | A downloaded local text model |
| Read the answer aloud offline | A downloaded local voice and its required resources |
| Use another language | Text in that language and a suitable voice, downloaded before disconnecting |
On mobile, spoken replies are a Pro feature. The Mac desktop app includes local voice output in its free core features. The Pro recording library on desktop is a separate feature; it is not required to hear a chat reply.
How do you enable spoken replies on Android or iPhone?
Download the local voice under Models > Voice, then open Chat Settings > TEXT TO SPEECH. Choose Chat mode and switch on Enable TTS. Select the language and speaker. Wait for any extra voice files to download before going offline.
Use mobile version 0.0.111 or later for these controls. The published requirements are Android 10 or later with at least 4 GB of RAM, or iPhone 12 or newer on iOS 17 or later.
- Open Models > Voice and download the on-device voice model.
- Select that local voice model if a remote voice service was active.
- Open Chat Settings > TEXT TO SPEECH.
- Select Chat, then turn on Enable TTS.
- Choose the Language and a Voice.
- Wait for the selected voice to finish downloading and loading.
- Open a chat and ask your local text model for a short answer.
- When the answer finishes, long-press the assistant message and select Speak.
The speech action applies to assistant replies. Long-pressing your own question does not provide the same action. The message may also show a playback control once speech is enabled.
How do you enable spoken replies on your Mac?
On your Mac, download and select a local voice model in Models. Open Settings > Voice. Set Interface mode to Chat, switch Text-to-speech to On, and select the language and speaker. Then use Speak in the assistant message’s actions menu.
These steps use OGAD for Mac. Use the current stable Mac build from the desktop downloads.
- Finish the local voice model and runtime setup in Models.
- Open Settings > Voice.
- Select Chat under Interface mode.
- Set Text-to-speech to On.
- Choose the voice language and speaker, and wait for any first-use downloads.
- Generate a short answer with a downloaded local text model.
- Open the reply’s actions menu and select Speak.
Use a voice that matches the answer’s language. A Spanish voice is a voice for Spanish text; it does not translate an English answer into Spanish.
How can you check that both stages work offline?
Finish setup, disconnect the device, and ask a new question. Check that the written answer appears, then play it. A successful replay of an old audio clip only shows that the clip is available. A new answer followed by new speech checks the local text and voice models together.
On a phone, use airplane mode and confirm that Wi-Fi is off. On a computer, disconnect its network connection for the check.
Try a request with no need for live information:
Explain the difference between a file and a folder in four short sentences. Use plain language.
Wait for the completed answer, then select Speak. You should see the text and hear spoken playback. A question about current weather or a live website adds a separate network task, so use a simple explanation for this check.
How should you write prompts for answers you want to hear?
Ask for short sentences, explicit names, and a clear order. Dense tables and long code blocks are harder to follow through speech. Keep the screen available for material whose layout or exact spelling matters.
For a spoken checklist:
Give me a short checklist for preparing a presentation. Use one action per sentence. State the step number before each action. Avoid a table.
For a technical explanation:
Explain this concept for listening. Define an abbreviation before using it. Keep the answer to five short paragraphs and put code in a separate section.
Check important facts in the written reply before relying on the audio. Speech playback changes the way you receive the answer; it does not check whether that answer is correct.
Can it speak in another language?
Yes, when the selected voice supports that language and the required files are downloaded. Ask the text model to answer in the desired language, then select the matching voice. Check the text model’s language ability separately from the voice model’s support.
The mobile voice list includes English US and UK options, French, Spanish, Italian, Portuguese, Hindi, Polish, and German. Other runtimes can offer a different set. Use the language choices shown by your installed voice model.
Changing languages can trigger a download even when your first voice already works offline. Prepare each voice you plan to use before leaving the connection.
How do you choose a voice you can understand clearly?
Select the language first, then compare available speakers with the same short assistant reply. Start at 1.0x speed. Include a name, a number, and a term you use at work. Check the written text before judging pronunciation; the voice cannot correct an incorrect answer.
On mobile, use Chat Settings > TEXT TO SPEECH > Voice. On Mac, use Settings > Voice. Wait for each selected voice to finish its setup before playing the reply. Some languages offer only one speaker. Prepare each voice you need before disconnecting.
Change playback speed after choosing the speaker. Test a mixed-language passage separately: a voice suited to one language can mispronounce words from another.
Why is there no Speak option, or no sound?
First check that text-to-speech is enabled and the local voice is ready. Then check the type of message and its state. On mobile, the action is available for completed assistant messages, rather than user messages or answers still being generated.
| Problem | What to check |
|---|---|
| Speech controls are missing on mobile | Pro access, Enable TTS, and a ready voice model |
| Speech controls are missing on desktop | Voice > Text-to-speech > On |
| Playback works only online | Whether the active voice is remote, or its local files are incomplete |
| There is text but no sound | Media volume and the selected output device or headphones |
| There is no new answer offline | Whether the active text model is local and downloaded |
| Words sound wrong | Voice language, text language, names, and abbreviations |
Start at 1.0x speed and use a short answer. Once the local text model and voice both work while disconnected, you can use longer replies and adjust the pace.