Use Ollama from your phone over home Wi-Fi or a VPN
Let a phone app use an Ollama model on your computer: open Ollama to your home network, find the base URL, test it and keep it off the internet.
To use Ollama from your phone, let Ollama on your computer accept connections from your home network, then point a phone app at http://<computer-ip>:11434/v1 with the exact model name. Away from home, use a private VPN, not an open router port.
This guide covers the computer, the network and a first test with any app that accepts an OpenAI-compatible base URL. It doesn’t cover running a model on the phone itself. Mono by Reunified accepts an endpoint like this as its text model, but it isn’t out yet.
1. Know what runs on each device
The model runs on the computer. The phone only sends requests and shows replies, so:
localhoston your phone means the phone itself. The app needs the computer’s address.- The computer has to be on, awake and reachable while you chat.
The example in this guide is made up. A Windows desktop called den-pc is on the home network at 192.168.1.20 with one model, llama3.2, already pulled. The phone is on the same home Wi-Fi.
2. Let Ollama listen beyond localhost
By default, Ollama binds to 127.0.0.1 on port 11434, so only the computer itself can reach it. To change that:
- Ollama app on macOS or Windows: open Settings and turn on “Expose Ollama to the network”. That’s the change on den-pc.
- Linux service: run
systemctl edit ollama.service, addEnvironment="OLLAMA_HOST=0.0.0.0:11434"under[Service], then runsystemctl daemon-reloadandsystemctl restart ollama. - No such switch: set the
OLLAMA_HOSTenvironment variable to0.0.0.0:11434(all networks) and restart Ollama. Ollama’s FAQ gives the steps for each system.
If your firewall asks about Ollama, allow it. On Windows, allow private networks only and set your home network’s profile type to Private (Windows 11 starts new networks as Public).
3. Find the address your phone should use
You need three values:
- Base URL:
http://<computer-ip>:11434/v1, with the computer’s local IP from its network settings (oripconfigon Windows). Ollama’s OpenAI-compatible API lives under/v1. Apps made for Ollama’s own API want the address without/v1. - Model name: run
ollama liston the computer and copy the name with its tag, such asllama3.2:latest. - API key: Ollama ignores it. Leave it empty, or type any placeholder if the app insists.
For den-pc, open http://192.168.1.20:11434/v1/models in the phone’s browser. A short JSON list with your model means the network path works.
Reserve the computer’s IP in your router’s settings so it doesn’t change.
4. Keep it off the public internet
Ollama’s API has no sign-in, so anyone who can reach port 11434 can run or delete your models. To keep it to your own devices:
- Don’t forward the port on your router.
- Use it on Wi-Fi you control, not a guest or shared network. Plain
http://isn’t encrypted. - On a laptop, Ollama now listens on café or office Wi-Fi too. Undo the change from section 2 before you leave.
- Away from home, use a private VPN that only your devices join, with the computer’s VPN address in the base URL.
- The FAQ’s proxy and tunnel examples have no sign-in, and a tunnel address is public. Add sign-in before you use one.
To chat from a café, the example reader installs the same private VPN on den-pc and the phone, then changes only the IP in the base URL. The AI privacy checklist helps you trace what leaves your devices.
5. Connect an app and run a first test
Enter the example’s values. Field names vary by app (see connecting a custom model).
| Field | Value |
|---|---|
| Base URL or Endpoint URL | http://192.168.1.20:11434/v1 |
| API key | Empty, or a placeholder |
| Model name or Model ID | llama3.2:latest, copied from ollama list |
Send: “Return the three words red, green, blue on separate lines.” Then ollama ps on the computer should list your model as loaded.
Then quit Ollama and send another message. You should get an error. A reply means the app used another model.
In Mono, turn on Text model under Settings > Integrations > Custom endpoints and enter these values. The “Use default model if unreachable” switch is on by default, so run the stopped-server test there too. With your own text model, chat doesn’t use skills or any tools, and its replies are free in Mono.
6. Fix common connection problems
| Symptom | What to check |
|---|---|
| Works on the computer, times out on the phone | Network switch on or OLLAMA_HOST set, Ollama restarted, firewall allows it on this network |
| Worked before, now times out | Computer asleep, its IP changed, phone on mobile data or guest Wi-Fi |
| Not found error for the path | /v1 missing from the base URL, or added twice |
| Error says the model isn’t found | Name and tag exactly as ollama list shows them |
| iPhone browser test works, app can’t connect | The app’s access under Settings > Privacy & Security > Local Network |
| First reply is slow | The model is loading. By default, Ollama unloads it after 5 minutes unused |
If the computer keeps sleeping, see how to keep your computer awake. Mono isn’t available yet, but you can join the waitlist.
