---
title: Other Local Servers
description: Connect local OpenAI compatible model servers
---

# Connect Other Local Servers

Connect to llama.cpp, vLLM, LocalAI, LiteLLM Proxy, and other servers
that expose OpenAI compatible routes.

SurfSense discovers models from:

```text
/v1/models
```

Chat requests use the same `/v1` base URL.

## Pick Your Setup

Use one of these URL patterns.

### SurfSense Runs in Docker

Use this when SurfSense is running from Docker and the model server is running on your computer.

```text
http://host.docker.internal:<port>/v1
```

Common ports:

| Server | Port |
|---|---|
| llama.cpp | `10000` |
| vLLM | `8000` |
| LocalAI | `8080` |
| LiteLLM Proxy | `4000` |
| text-generation-webui | `5000` |

### SurfSense Runs Without Docker

Use this when SurfSense and the model server both run directly on the same computer.

```text
http://localhost:<port>/v1
```

### Model Server Runs on Another Computer

Use this when the model server is running on another machine in your network.

```text
http://<host>:<port>/v1
```

## Add the Connection

1. Open Search Space Settings.
2. Go to Models.
3. Select OpenAI Compatible.
4. Set API Base URL.
5. Add an API Key only if your server requires one.
6. Select the models you want to enable.
7. Save the connection.

If you enter the URL without `/v1`, SurfSense adds `/v1` for requests.

## Verify

From the host:

```bash
curl http://localhost:<port>/v1/models
```

From the SurfSense backend container:

```bash
docker compose exec backend curl http://host.docker.internal:<port>/v1/models
```

A working server returns JSON with a `data` array.

## When Not to Use This

Use the Ollama provider for Ollama. It uses native routes such as `/api/tags`.

Use the LM Studio provider for LM Studio. Its default URL is already set.

## Troubleshooting

### Endpoint returned 404

The server does not expose `/v1/models`.

Enable the server's OpenAI compatible mode.

### Connection refused

The backend cannot reach the server.

Check that the server is running and that the port is open.

### No models found

The server returned an empty model list.

Load or serve a model, then refresh model discovery in SurfSense.