You can have a spoken AI conversation on your Android phone without sending the audio to a cloud service. OGAM (Off Grid AI Mobile) Pro can listen when you speak, finish a turn after a pause, and read the answer aloud. Download the local speech, chat, and voice models first, then test with internet access off.
| Get OGAM on Google Play | Mobile features |
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.
Hands-free mode is useful when the phone is beside you and you want to continue a conversation without pressing the microphone for every question. This guide covers an active conversation inside the app. It does not set up a system-wide wake word or promise listening with the phone locked.
What do you need for an offline voice conversation?
You need OGAM Pro, microphone access, and local models for speech recognition, answers, and spoken replies. The Android app’s published minimum is Android 10 or later with at least 4 GB of RAM. Larger models need more memory than that minimum provides.
Complete the app installation, Pro setup, and all model downloads while connected.
| Part | What it does | What to prepare |
|---|---|---|
| Speech recognition | Turns your question into text | A downloaded local transcription model |
| Chat | Generates the answer | A downloaded on-device text model |
| Speech output | Reads the answer aloud | A downloaded local voice model and selected voice assets |
Use small models for the first test. Running a chat model alongside speech models adds memory pressure. A model fitting by itself does not establish that every combination will fit at once.
Spoken output and the Audio interface used here are Mobile Pro features. Free speech-to-text alone does not provide the full conversation described in this article.
How do you prepare the models?
In Model Settings > Transcription (Speech to Text) > Transcription model, select and download an on-device speech model. Choose a multilingual model for languages other than English. Then select a local text model for answers and download the voice model under Models > Voice.
For a first multilingual test, Base, 99 languages is a small transcription option. Do not choose an English-only model if you plan to speak another language.
Set the spoken language in Chat Settings > SPEECH TO TEXT > Language. Set the reply voice in the text-to-speech section. These settings do different jobs: one helps recognize your speech; the other determines how the answer sounds.
Wait for the voice-ready message after selecting a language and voice. Switching languages can download additional assets even when another voice already works.
If you used a remote server before, check all active selections. Local transcription with a remote chat or voice model is not a fully offline conversation.
How do you turn on Hands-free mode?
Open Chat Settings and select Audio under TEXT TO SPEECH > Interface Mode. In SPEECH TO TEXT, set Voice turns to Hands-free. This makes a turn start when you speak and finish after the selected pause. It applies to the Audio interface, rather than ordinary text-chat dictation.
The three turn choices behave differently:
| Voice turns | Behavior |
|---|---|
| Manual | You start and stop the recording |
| Auto | You start it; a pause ends the turn |
| Hands-free | Listening starts the turn when speech is detected; a pause ends it |
For your first test, leave time to think:
- Set End of turn to 5s.
- Set Listen again to 2s.
These are the current defaults. You can shorten them later if the conversation feels too slow.
End of turn is the pause that ends your question. Listen again is the wait after the reply before the microphone reopens. A short setting can feel quicker but gives less room for a pause or lingering speaker sound.
What if you want silence to stop input, but prefer to start each question yourself?
Choose Auto under Voice turns. You use the microphone control to begin, then a pause ends the recording and submits the recognized question. This gives you control over when listening starts without needing to reach for Stop after each sentence.
End of turn offers 1s, 2s, 3s, and 5s. Start at 5 seconds. Shorter delays suit short, prepared questions, but a thinking pause can submit an unfinished thought. The setting detects audio silence; it cannot know whether you have finished your idea.
Try one complete sentence, then a question with a natural pause. Check that the transcript contains the final words before shortening the timer. Background speech or music can keep input open after you stop talking. Use Manual when you need longer pauses or a noisy room makes automatic timing unreliable.
Auto and Hands-free apply to the Audio interface. Use ordinary Chat dictation if you want to review and edit the recognized words before sending them.
How do you test a complete conversation offline?
Finish the model setup, keep the app open, and turn on airplane mode. Confirm Wi-Fi is off too. In Audio mode, wait for the listening state, speak a short question, then pause. The app should transcribe it, generate a local answer, and play the reply.
1. Start with a quiet room
Put the phone where its microphone can hear you clearly. Keep its speaker volume at a comfortable level. Allow Android microphone access when asked.
When the empty conversation shows Waiting for your voice, speak. If it shows Tap to speak, use the microphone control to start.
2. Ask one short question
For example:
Give me two questions to ask when planning a weekend trip.
Pause for the selected end-of-turn period. In Audio mode, the recognized question is sent automatically. You do not get the same before-send editing step as normal text-chat dictation.
3. Continue with a follow-up
After the answer, wait for listening to resume. Ask:
Make the first question more specific.
This checks the repeat conversation cycle, rather than only a single recording. Check the displayed transcript if the answer seems unrelated.
4. Leave voice mode when finished
Return Interface Mode to Chat. Changing out of Audio mode cancels an active recording and resets the voice session.
Use the visible listening and recording indicators to understand the current state. Hands-free means fewer controls during a conversation; it does not mean permanent background listening.
How do you stop a reply and speak again?
Use the square Stop control while the assistant is generating or playing a reply. Then use the microphone control when it is available to begin your next turn. A deliberate stop can also stop automatic listening, so wait for a visible listening or recording state before speaking.
Do not rely on speaking over the reply. In release 0.0.111, recording waits while the assistant speaks to prevent its own voice from becoming the next input. Earlier release notes described talk-over interruption, but that is not the behavior to expect from this release.
What should you change if the conversation cuts you off?
| Problem | Check | Action |
|---|---|---|
| The question sends before you finish | End of turn | Choose a longer pause, such as 5s, or use Manual turns. |
| The phone hears its own answer | Speaker volume and Listen again | Reduce the volume and increase the wait before listening resumes. |
| You have to tap for every question | Interface mode and Voice turns | Use Audio plus Hands-free. Auto still requires you to start each turn. |
| You hear no answer | Local voice readiness and Android audio output | Finish voice downloads and check volume, headphones, or Bluetooth routing. |
| It works only while connected | Active transcription, chat, and voice models | Confirm that all three are local and already downloaded. |
| The answer is unrelated | The recognized words | Check the transcript. Set the correct speech language and try a clearer recording. |
| Models fail to load | Available memory | Use a smaller chat or transcription model and close memory-heavy apps. |
The silence detector uses the audio level to decide when a turn ends. Background speech or music can delay that decision. Use Manual mode when you need direct control over a difficult recording.
Does your voice leave the phone?
With local transcription, chat, and speech output selected, the conversation is processed on the Android phone. The initial downloads and Pro setup are separate network steps.
Remote model selections or connected tools can change the network behavior. For the offline test, use a simple conversation that does not ask for web information or external actions.
Android hands-free FAQ
Does Hands-free work in normal Chat mode?
Hands-free turn detection is intended for the Audio interface. Normal text-chat dictation keeps its own recording behavior and editable message box.
Does this replace Android’s system assistant?
No. This guide configures a conversation in OGAM. It does not register a system-wide wake word or promise lock-screen access.
Will every language work equally well?
No. Speech recognition and available voices have separate language support. Choose matching supported settings and test your own voice and recordings.
Is a network needed between each question?
Local processing does not need a connection after the required setup. Keep all three model choices local and prepare each voice you intend to use.
Try a short question and one follow-up with airplane mode enabled. Adjust the two pause settings only after that basic conversation works.