> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mergewatch.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Ollama

> Run models locally with Ollama — experimental, for air-gapped environments.

<Warning>
  **Ollama is experimental.** Review quality is significantly lower than Claude or GPT-4o. Use only if your environment cannot make external API calls.
</Warning>

[Ollama](https://ollama.com) lets you run open-source models locally. This provider is designed for **air-gapped environments** where no external API access is possible.

## When to use Ollama

* Your network has no outbound internet access
* Security policy prohibits sending code to external LLM APIs
* You want to evaluate MergeWatch without any API keys

## Configuration

Set these environment variables in your `.env` file or container environment:

| Variable          | Required | Value                                                                                                                                                                 |
| ----------------- | -------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `LLM_PROVIDER`    | Yes      | `ollama`                                                                                                                                                              |
| `LLM_MODEL`       | Yes      | Ollama model tag (e.g., `qwen2.5-coder:7b`, `llama3`). The Ollama provider has no built-in default — you must pick a model that you've pulled into the Ollama server. |
| `OLLAMA_BASE_URL` | No       | Ollama API endpoint (default: `http://localhost:11434`)                                                                                                               |

## Picking a model

You must set `LLM_MODEL` to a model tag that you've pulled into your Ollama server. There is no built-in default. Suggested starting points:

| Model               | Best for                                       | Approx VRAM |
| ------------------- | ---------------------------------------------- | ----------- |
| `qwen2.5-coder:7b`  | Code-focused reviews on an 8 GB GPU            | \~6–8 GB    |
| `qwen2.5-coder:14b` | Better review quality if you have 16 GB VRAM   | \~12–16 GB  |
| `llama3`            | General-purpose baseline for quick smoke tests | \~4–8 GB    |

## Setup

Pull the model before starting MergeWatch — MergeWatch will not pull it for you:

```bash theme={null}
ollama pull qwen2.5-coder:7b
```

Verify the model is available:

```bash theme={null}
ollama list
```

## Docker Compose setup

Run Ollama as a sidecar alongside MergeWatch:

```yaml docker-compose.yml theme={null}
services:
  mergewatch:
    image: ghcr.io/santthosh/mergewatch:latest
    ports:
      - "3000:3000"
    env_file: .env
    environment:
      LLM_PROVIDER: ollama
      LLM_MODEL: qwen2.5-coder:7b
      OLLAMA_BASE_URL: http://ollama:11434
    depends_on:
      - ollama

  ollama:
    image: ollama/ollama:latest
    ports:
      - "11434:11434"
    volumes:
      - ollama-data:/root/.ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  ollama-data:
```

<Note>
  After starting the stack for the first time, exec into the Ollama container to pull the model:

  ```bash theme={null}
  docker compose exec ollama ollama pull qwen2.5-coder:7b
  ```
</Note>

## Hardware requirements

| Component  | Minimum    | Recommended |
| ---------- | ---------- | ----------- |
| GPU VRAM   | 8 GB       | 16 GB       |
| System RAM | 16 GB      | 32 GB       |
| Disk       | 10 GB free | 20 GB free  |

<Tip>
  Larger models like `qwen2.5-coder:14b` produce better reviews but require 16 GB VRAM. If you have the hardware, set `LLM_MODEL=qwen2.5-coder:14b` for improved quality.
</Tip>

## Next steps

<CardGroup cols={2}>
  <Card title="Air-gapped deployment" icon="lock" href="/self-hosting/platforms/air-gapped">
    Full guide for deploying MergeWatch without internet access.
  </Card>

  <Card title="Anthropic (direct)" icon="bolt" href="/self-hosting/llm-providers/anthropic">
    The recommended provider for the best review quality.
  </Card>

  <Card title="Configure review behavior" icon="sliders" href="/configuration/review-behavior">
    Tune sensitivity, ignored paths, and review focus areas.
  </Card>

  <Card title="Environment variables" icon="gear" href="/reference/env-vars">
    Full list of supported environment variables.
  </Card>
</CardGroup>