On this page
Your Android phone can run a 2B to 4B model on its own. For anything bigger, like GPT-OSS 20B, Gemma 4 31B or Llama 3.1 70B, run the model on your own computer and use it from your phone. Off Grid AI Mobile (OGAM), the free Android app, finds Off Grid AI Desktop, Ollama or LM Studio on your Wi-Fi. Pick a model from the computer and chat. The computer does the work, your phone shows the answer, and nothing goes to a cloud.
Setup takes a few minutes if you already run Ollama or LM Studio. The steps use OGAM 0.0.124.
| Get OGAM for Android | Get Off Grid AI Desktop |
Why your phone can’t hold the big ones
A running model needs its file plus room for the conversation, and Android gives one app well under the phone’s total RAM. So OGAM sizes every file against your phone:
| Phone RAM | Runs well | Tight | Too big |
|---|---|---|---|
| 8 GB | under 4.2 GB | 4.2 to 5 GB | over 5 GB |
| 12 GB | under 6.2 GB | 6.2 to 7.4 GB | over 7.4 GB |
| 16 GB | under 8.3 GB | 8.3 to 9.9 GB | over 9.9 GB |
That’s models up to about 7B to 9B, such as Gemma 4 E4B, Qwen 3.5 4B or 9B, or Llama 3.2 3B. Good for quick questions, rewrites and offline use. Now look at what people actually pull on Ollama:
| Model | Ollama download | Computer memory to aim for |
|---|---|---|
| GPT-OSS 20B | 14 GB | 24 GB |
| Gemma 4 26B | 16 to 19 GB | 32 GB |
| Qwen 3.5 27B | 17 to 20 GB | 32 GB |
| Gemma 4 31B | 19 to 20 GB | 32 GB |
| Llama 3.1 70B | 43 GB | 64 GB |
| GPT-OSS 120B | 65 GB | 96 GB or more |
Download sizes are from the Ollama library on 16 Nov 2026. The memory column adds room for the conversation and the system, so treat it as a rough guide. None of these fit any phone. A 32 GB Mac or PC covers most of them.
Step 1: Load the model on your computer
Pick one of these. You only need one.
Off Grid AI Desktop. Install it, open Models, download a model and select it. Ask it one question on the computer to check it answers. Its gateway listens on port 7878.
Ollama. Pull the model (ollama pull gpt-oss:20b, for example). Ollama only listens to the computer itself until you open it to your network. On Mac or Linux:
OLLAMA_HOST=0.0.0.0 ollama serve
On Windows, add a system environment variable OLLAMA_HOST with the value 0.0.0.0 and restart Ollama.
LM Studio. Download the model, open the Developer tab, load it, start the server and turn on Serve on Local Network.
Keep the computer awake and on the same Wi-Fi as your phone.
Step 2: Find the computer from your phone
- Open OGAM and go to Settings → Remote servers.
- Under Scan for, leave on the servers you use: Off Grid AI Desktop (7878), Ollama (11434) and LM Studio (1234).
- Tap Scan network. OGAM checks your Wi-Fi and lists what answered.
- Turn on Auto-discover on Wi-Fi if you want OGAM to look each time you open this screen.
If the scan finds nothing, tap Add manually. Enter a name and the computer’s address with its port, for example http://192.168.1.50:11434 for Ollama. Tap Test connection, then Add server. Don’t use 127.0.0.1. On the phone, that points back at the phone.
If Off Grid AI Desktop asks for a key, OGAM tells you. Copy one from Desktop’s Gateway → API keys and paste it into the server’s API key field. The key stays on your phone.
Step 3: Pick the big model and chat
Open a chat and tap the model name to open the model selector. Your phone’s own models sit under Local Models. The computer’s models sit under the server’s name. Tap GPT-OSS 20B, or whichever you loaded, and ask.
The model shows a Remote badge under Currently Loaded, so you always know where it runs. You can switch back to an on-device model in the same chat when you leave the house.
Symptom, check, fix
| Symptom | Check | Fix |
|---|---|---|
| Scan finds nothing | Same Wi-Fi on both? | Join the same network, or Add manually |
| Ollama not found | OLLAMA_HOST set? |
Set it to 0.0.0.0 and restart Ollama |
| LM Studio not found | Serve on Local Network on? | Turn it on and start the server |
| Server shows Not answering | Computer asleep? | Wake it and keep it awake |
| Desktop found but no models | API key | Paste a key from Gateway → API keys |
| Answers are slow | Model size vs computer memory | Pick a smaller model from the table |
Away from home
This works on your home network. To use it from outside, put both devices on a private network such as Tailscale. OGAM accepts Tailscale addresses the same way it accepts home addresses. See how to reach your home AI from anywhere with Tailscale. Don’t forward ports 7878, 11434 or 1234 to the internet.
Need the full-size model?
Some models are too big for your computer too. GPT-OSS 120B wants about 96 GB of memory, and most laptops have 16 to 32 GB. Off Grid AI Boost runs open models on our GPUs at one flat price from $4.99 a month for the first cohort (regular price $9.99, plans up to $199.99). There’s no per-token billing, and we don’t store your prompts or train on them. The full privacy policy comes with launch.
Boost is a pre-order. It starts in January; pre-order now to lock in half price, and you’ll get an API key. Usage limits for each plan aren’t published yet. Keep a small model on your phone for offline use, your computer for the big ones, and Boost for the ones your computer can’t hold.
FAQ
Can an Android phone run a 20B or 70B model? Not on the phone itself. GPT-OSS 20B is a 14 GB download, and OGAM’s limit on a 16 GB phone is about 8 GB. Run it on your computer and chat from the phone.
Does my computer need a GPU? No. An Apple silicon Mac or a PC with enough RAM runs these on the CPU or built-in graphics. A GPU makes answers faster.
Does this need the internet? No. Your phone talks to your computer over your home Wi-Fi. After the model is downloaded, the internet can be off.
Do I need Pro? No. Remote servers are part of the free app.
Can I still use AI when I leave home? Yes. Download a small model to the phone, like Gemma 4 E4B or Qwen 3.5 4B, and switch to it in the same chat.
Open Settings → Remote servers in OGAM and tap Scan network. If your computer is already running a model, you’ll be chatting with it in under a minute.