# Ollama + Claude Code — Docker Setup

Run **qwen2.5-coder:32b** locally via Ollama, with Claude Code wired up to use it as its backend.

## Prerequisites

- Docker & Docker Compose v2
- NVIDIA Container Toolkit installed on the host
- NVIDIA GPU with 32 GB+ VRAM (or large system RAM for CPU offload)

## Quick Start

```bash
# 1. Clone this repository
git clone https://forge.wilddragon.net/zgaetano/ollama-claude-code.git
cd ollama-claude-code

# 2. Build and start both services
docker compose up -d --build

# 3. Watch Ollama pull the model on first run
docker compose logs -f ollama

# 4. Shell into the Claude Code container
docker compose exec claude-code bash

# 5. Start Claude Code
claude
```

## Environment Variables

### ollama service
| Variable | Default | Description |
|---|---|---|
| OLLAMA_HOST | 0.0.0.0 | Interface Ollama listens on |
| OLLAMA_KEEP_ALIVE | 24h | How long to keep the model loaded |
| OLLAMA_MODEL | qwen2.5-coder:32b | Model to pull on startup |

### claude-code service
| Variable | Default | Description |
|---|---|---|
| ANTHROPIC_BASE_URL | http://ollama:11434/v1 | Points Claude Code at local Ollama |
| ANTHROPIC_API_KEY | ollama | Dummy key (Ollama does not verify it) |
| CLAUDE_CODE_DEFAULT_MODEL | qwen2.5-coder:32b | Model Claude Code will use |

## Volumes
- `ollama_data` - persists downloaded models across restarts
- `workspace` - shared workspace inside the claude-code container

## Stopping
```bash
docker compose down
# To also remove model data:
docker compose down -v
```