# Use Ollama on your iPhone or iPad

Connect Vanta to an Ollama server using its OpenAI-compatible API, including LAN binding, model identifiers and connection checks.

Canonical: https://vantaclient.app/docs/ollama/
Updated: 2026-10-02

Vanta connects to Ollama through its OpenAI-compatible API. Ollama runs the model on your computer or server; Vanta sends messages from your iPhone or iPad to that server. Use a locally hosted model if you want processing to stay on that machine.

## Prepare Ollama

Install Ollama and download a model that runs on your host. The [Ollama OpenAI compatibility guide](https://docs.ollama.com/api/openai-compatibility) documents the `/v1` API. It implements a subset of that API, so support for optional features varies.

Ollama normally binds to loopback, which another device cannot reach. For LAN access, configure `OLLAMA_HOST` and restart Ollama. On macOS, when using the Ollama application:

```sh
launchctl setenv OLLAMA_HOST "0.0.0.0:11434"
```

Quit and reopen Ollama after setting it. On Linux or Windows, follow the operating-system-specific steps in the [Ollama FAQ](https://docs.ollama.com/faq). Binding to `0.0.0.0` listens on all IPv4 interfaces; restrict access with your network and firewall.

Keep the phone and host on a trusted reachable network. Check the model installed on your host with `ollama list`.

## Add Ollama in Vanta

1. Open **Settings → Providers → Add provider**.
2. Name the provider **Ollama**.
3. Enter `http://192.168.1.10:11434/v1`, using your host's real address.
4. Leave the API key empty for an ordinary local Ollama server. If you use an authenticated proxy, enter its key.
5. Check and save the provider, then create a model preset using the exact installed model identifier, including its tag.
6. Select the preset in **Settings → Models** and send a short test message.

Do not enter `/api/chat` as the base URL. Vanta uses the OpenAI-compatible chat endpoint under `/v1`. The normalizer adds `/v1` to an address with no path, but preserves a path you explicitly enter.

<img src="https://vantaclient.app/media/providers.webp" width="720" height="1564" loading="lazy" alt="Vanta's provider settings showing an Ollama base URL ending in port 11434 and /v1." />

The address in this example belongs to the example server. Enter the address of your own host.

## Verify from the network

```sh
curl http://192.168.1.10:11434/v1/models
```

A JSON model list confirms the API is reachable from the device running the check. It does not prove your phone can reach it; check the saved provider from Vanta too.

## Common problems

| Symptom | Check |
| --- | --- |
| Connection refused | Ollama is running, the LAN binding took effect, and the firewall permits port `11434`. |
| “Model not found” | The preset identifier matches an installed model and tag. |
| A 404 response | The saved base path is `/v1`, rather than Ollama's native `/api` path. |
| Long wait before a reply | The host may be loading the model. Try a smaller model and a short prompt. |
| Tool or image request fails | Both the chosen model and Ollama's compatible API must support the feature. |

## Privacy and remote access

For a locally hosted model, inference takes place on your Ollama host. Ollama also offers cloud models; selecting one changes where processing happens. A blank local API key does not secure a server exposed to the internet. Use a private VPN or an authenticated HTTPS proxy for remote access.

This guide documents setup, not measured model performance. See [on-device models](https://vantaclient.app/docs/on-device/) for running directly on the phone, or [troubleshooting](https://vantaclient.app/docs/troubleshooting/) if the server connection fails.