A model that does not fit in your phone can still be useful from your phone. OGAM (Off Grid AI Mobile) can connect to a larger model running in OGAD (Off Grid AI Desktop) on your own computer. The computer holds the model in memory. Your phone sends the question and displays the answer.
| Download OGAM for Android or iPhone | 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 a small phone model struggles with a writing task, a longer question, or a piece of code. You can try a model your computer can run without putting that model’s full download and memory load on your phone.
The route below uses your home Wi-Fi. After app and model downloads, the connection can work without internet as long as the local network stays available. Your phone still needs a connection to the computer.
Can a phone with limited RAM use a larger local AI model?
Yes, by running the model on another computer you own. OGAM acts as the chat client and OGAD supplies the model. This changes where the work happens. It does not increase the phone’s RAM or make an oversized model run directly on the phone.
Choose the route that fits where you will use it:
| What you need | Where the model runs | Connection needed |
|---|---|---|
| AI while the phone has no network connection | On the phone, using a model that fits | None after setup |
| A larger model while at home | On your own computer | Local Wi-Fi or another reachable local network |
| A home model while away | On your home computer | A private remote connection; normally internet at both ends |
This guide covers the middle option. The basic remote-server connection is a core feature. Pro device sync is separate and is not required to send a chat request to your own model server.
What do you need?
Install OGAM on Android or iPhone and OGAD on your computer. Download a text model that the computer can run. Connect both devices to the same trusted home network, keep the computer awake, and leave OGAD running.
The relevant releases are OGAM 0.0.111 and OGAD 0.0.51. Use current compatible versions on both devices.
Start with a model that runs reliably on the computer. A larger download does not guarantee a better answer, and its file size is not its full working memory requirement. The host also needs memory for the model’s context and other running apps.
How do you prepare the computer?
Open Models in OGAD, download your chosen text model, and select it. Ask a short question on the computer first. Once it answers, use the computer’s local network address to reach its gateway from the phone.
- Prepare the model in OGAD and confirm it answers a normal chat message.
- Open Gateway to check the API service.
- Find the computer’s local IP address in its network settings.
- Keep the computer awake and connected to the same network as the phone.
OGAD’s gateway uses port 7878. A computer with the example address 192.168.1.50 would have this server address:
http://192.168.1.50:7878
Replace that address with your own computer’s address. Do not use 127.0.0.1 on the phone: that would point back to the phone.
The gateway described here does not require an API key. Restrict access to your trusted devices with the computer’s firewall and network settings. Do not forward port 7878 from the router to the public internet.
How do you select the computer’s model on your phone?
Open Settings → Remote Servers in OGAM. Add the computer’s address, test the connection, then select a text model supplied by that server. Models from the server appear separately from models stored on the phone.
- Add a server and give it a clear name, such as Home computer.
- Enter the base address, such as
http://192.168.1.50:7878. - Leave the API key empty for this OGAD gateway.
- Tap Test connection.
- Choose the discovered Text model, then tap Add server.
- Open the chat model selector and choose the model under your computer’s name.
You can also use Scan network to look for a reachable OGAD server. Manual entry is useful when network discovery cannot find the computer but its address is reachable.
Wait for model selection to finish before sending a question. The model may need to load on the computer. You do not need to download the same model onto the phone.
What should you try first?
Start a new chat with a short task whose result you can judge. For example, paste a rough paragraph and ask:
Rewrite this as a clear email. Keep every date and price unchanged. Put the requested action in the first sentence.
Check the selected server and model, then review whether the answer kept the facts. After that works, try the longer task that was difficult with the phone model.
The answer should appear in OGAM while the computer runs the model. Keeping the model on the computer does not remove all phone memory use: the app still holds the conversation, receives text, and displays the result.
If you later switch to a downloaded phone model, new requests use that local choice. Check the model selector before a private or important request, especially if you have saved more than one server.
Why is the home model unavailable?
Most connection problems come from the address, network, sleeping computer, or model state. Check the simplest part first: can the model answer inside OGAD on the computer?
| Symptom | What to check |
|---|---|
| Test connection fails | Confirm the IP address, port 7878, firewall access, and that OGAD is open. |
| Discovery finds no computer | Try the explicit local address. Guest Wi-Fi may prevent devices from reaching each other. |
| Server connects but the desired model is missing | Finish its download on the computer, then test or refresh the server again. |
| Model fails to load | Choose a smaller host model or free memory on the computer. |
| Requests stop when you leave home | The local IP is only reachable on that network. Use a separate private remote setup for access away from home. |
Do not raise the phone’s memory budget to fix a model running on the computer. The host’s available memory is the relevant limit for that model.
Does the prompt leave the phone?
Yes. It travels from OGAM to the computer you selected. With an OGAD host running a downloaded local model, inference takes place on your own hardware. If the selected server forwards work to a cloud provider, that is a different route.
This setup is useful precisely because the phone can use another device’s resources. Keep the network trusted and choose a server you control. Local Wi-Fi does not by itself mean that an unauthenticated HTTP service is suitable for a shared public network.
Try your existing computer before replacing the phone
Install OGAD, run a model that fits your computer, and connect to it from OGAM. Start with one short chat on your home network. Once that works, you can keep using the phone you have while the larger model runs on hardware you already own.