Speak into your phone. Let your computer handle the transcription. OGAM (Off Grid AI Mobile) can send a voice recording to a transcription model in OGAD (Off Grid AI Desktop), then put the returned words into your chat input. You can review and edit them before sending a message.

Download OGAM

Download OGAD

---

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.

This is useful when you want to dictate a longer thought without keeping a speech model on the phone. It also lets you use a transcription model that fits your computer better than your phone.

The guide uses the phone’s microphone and ordinary chat dictation. Once the apps and model are downloaded, your own local network can carry the recording and transcript without an internet connection.

How does computer-powered phone transcription work?

OGAM records your speech, sends the audio to the transcription server you selected, and receives text. With OGAD running a downloaded local model, your computer does the speech recognition. In ordinary Chat mode, the words go into the editable message input rather than being sent as a question automatically.

There are two separate model choices:

Model	What it does
Transcription model	Turns the recording into text
Chat model	Answers the text if you choose to send it

Choosing a remote chat model alone does not select remote transcription. Select the computer’s Transcription model as well.

The core remote-server and chat-dictation route does not require Pro device sync or Pro voice conversation mode. This guide is based on OGAM 0.0.111 and the gateway in OGAD 0.0.51.

What do you need?

Use OGAM on Android or iPhone and OGAD on a computer on the same trusted local network. Download a transcription model on the computer and allow microphone access on the phone. Keep the computer awake with OGAD open while you dictate.

Prepare these before the first recording:

• A desktop transcription model that fits the computer’s memory.
• A working local connection between phone and computer.
• A place quiet enough to record clear speech.
• A short sentence that you can check against the returned text.

Use a multilingual model if you plan to speak a language other than English. The model choice matters even though the recording starts on the phone. A remote connection does not add language support to an English-only model.

How do you connect the phone to your transcription model?

In OGAM, add your computer under Settings → Remote Servers, test the connection, and choose its transcription model. Then select that remote model in the transcription picker so new recordings use it.

1. In OGAD, download a local transcription model from Models.
2. Open Gateway and check that the app’s API service is available.
3. Find the computer’s local IP address in its network settings.
4. On the phone, open Settings → Remote Servers and add a server.
5. Give it a name such as Home transcription and enter http://YOUR-COMPUTER-IP:7878.
6. Tap Test connection, choose a Transcription model, and tap Add server.
7. Open Settings → Model Settings → Transcription (Speech to Text) → Transcription model.
8. Choose the model under Remote models, identified by your computer’s name.

For the OGAD gateway used here, leave the API key empty. Its port is 7878. Keep access limited to your trusted devices through the computer’s firewall; do not open the port to the public internet.

If you can connect but see no transcription model, complete the model download on the computer and test the server again. A text model in the list is not a substitute for a speech-to-text model.

How do you dictate your first note?

Open an ordinary chat in OGAM and use the microphone control to record a short passage. End the recording and wait for the transcript. Read the words in the message input before you press Send.

Try this practical note:

The desk needs to move closer to the window. Keep the lamp on the left. Measure the space before ordering a new shelf.

Check that all three instructions appear and that the transcript has not changed a key word. Correct any errors in the input. You can now keep working with the text or send it to your chosen chat model.

For example, after checking the transcript, add:

Turn these notes into a short checklist. Do not add measurements I did not give you.

The transcription step captures what you said. The chat step organizes it. Keeping those two steps separate makes it easier to spot whether an error came from speech recognition or from the answer that followed.

For a longer thought, record one useful section at a time. Shorter sections are easier to review, and you can correct names and figures while you still remember what you said.

Can you use a larger transcription model this way?

Yes, if the computer can run it. The phone does not need to hold the selected desktop model in memory. You can compare a larger computer model with a smaller one using the same short recording or sentence.

Choose based on the result you need. For a rough idea, a small model may be sufficient. For names or speech that a small model repeatedly misses, another model may be worth trying. Review the actual words rather than assuming that a larger model always fixes the problem.

The computer still needs memory for transcription. Long recordings, other running models, and busy applications can affect whether the task finishes. Start with a short recording before attempting longer dictation.

Why is the transcript missing or wrong?

First check whether the phone recorded audible speech. Then check the active transcription model and connection. A working chat response from the computer does not prove the transcription model is ready.

Problem	Check and action
Microphone will not start	Allow microphone permission and check whether another app is using it.
Remote model does not appear	Finish the host model download and test the server again.
Recording ends but no text arrives	Keep OGAD awake, confirm the host address, and check the phone remains connected.
Names or numbers are wrong	Correct them before sending; record more clearly or compare a different model.
Words are sent immediately	Return to ordinary Chat dictation instead of a voice conversation mode that sends turns automatically.

Speak close enough to the microphone for a clear recording. More computer memory cannot remove every problem caused by background sound or several people speaking at once.

Is this cloud transcription?

With the setup above, the recording travels from your phone to your own computer and is transcribed there. It leaves the phone, but it does not require uploading to a cloud transcription provider.

Keep both devices on the local network for the internet-free route. If you later connect to a server run by someone else, that operator receives your audio. If you choose a separate cloud chat model after transcription, the text you send also follows that model’s route.

This guide does not require keeping the raw recording as a voice note. Ordinary chat dictation gives you editable text. Use a separate recording workflow when retaining the original audio is part of your task.

Try one spoken note

Install OGAM and connect it to a transcription model in OGAD. Record one short note, check the text, and send it only when it says what you meant. Your phone stays the convenient place to speak while your own computer supplies the speech model.