plano/demos/llm_routing/claude_code_router/README.md

# Claude Code Router - Multi-Model Access with Intelligent Routing

Plano extends Claude Code to access multiple LLM providers through a single interface. Offering two key benefits:

1. **Access to Models**: Connect to Grok, Mistral, Gemini, DeepSeek, GPT models, Claude, and local models via Ollama
2. **Intelligent Routing via Preferences for Coding Tasks**: Configure which models handle specific development tasks:
   - Code generation and implementation
   - Code reviews and analysis
   - Architecture and system design
   - Debugging and optimization
   - Documentation and explanations

Uses a [1.5B preference-aligned router LLM](https://arxiv.org/abs/2506.16655) to automatically select the best model based on your request type.

## Benefits

- **Single Interface**: Access multiple LLM providers through the same Claude Code CLI
- **Task-Aware Routing**: Requests are analyzed and routed to models based on task type (code generation, debugging, architecture, documentation)
- **Provider Flexibility**: Add or remove LLM providers without changing your workflow
- **Routing Transparency**: See which model handles each request and why

## How It Works

Plano sits between Claude Code and multiple LLM providers, analyzing each request to route it to the most suitable model:

```
Your Request → Plano → Suitable Model → Response
             ↓
    [Task Analysis & Model Selection]
```

**Supported Providers**: OpenAI-compatible, Anthropic, DeepSeek, Grok, Gemini, Llama, Mistral, local models via Ollama. See [full list of supported providers](https://docs.planoai.dev/concepts/llm_providers/supported_providers.html).


## Quick Start (5 minutes)

### Prerequisites
```bash
# Install Claude Code if you haven't already
npm install -g @anthropic-ai/claude-code

# Ensure Docker is running
docker --version
```

### Step 1: Get Configuration
```bash
# Clone and navigate to demo
git clone https://github.com/katanemo/arch.git
cd arch/demos/llm_routing/claude_code_router
```

### Step 2: Set API Keys
```bash
# Copy the sample environment file
cp .env .env.local

# Edit with your actual API keys
export OPENAI_API_KEY="your-openai-key-here"
export ANTHROPIC_API_KEY="your-anthropic-key-here"
# Add other providers as needed
```

### Step 3: Start Plano
```bash
# Install using uv (recommended)
uv tool install planoai
planoai up
# Or if already installed with uv: uvx planoai up

# Or install using pip (traditional)
pip install planoai
planoai up
```

### Step 4: Launch Enhanced Claude Code
```bash
# This will launch Claude Code with multi-model routing
planoai cli-agent claude
# Or if installed with uv: uvx planoai cli-agent claude
```
![claude code](claude_code.png)

### Monitor Model Selection in Real-Time

While using Claude Code, open a **second terminal** and run this helper script to watch routing decisions. This script shows you:
- **Which model** was selected for each request
- **Real-time routing decisions** as you work

```bash
# In a new terminal window (from the same directory)
sh pretty_model_resolution.sh
```
![model_selection](model_selection.png)

## Understanding the Configuration

The `config.yaml` file defines your multi-model setup:

```yaml
llm_providers:
  - model: openai/gpt-4.1-2025-04-14
    access_key: $OPENAI_API_KEY
    routing_preferences:
      - name: code generation
        description: generating new code snippets and functions

  - model: anthropic/claude-3-5-sonnet-20241022
    access_key: $ANTHROPIC_API_KEY
    routing_preferences:
      - name: code understanding
        description: explaining and analyzing existing code
```

## Advanced Usage

### Override Model Selection
```bash
# Force a specific model for this session
planoai cli-agent claude --settings='{"ANTHROPIC_SMALL_FAST_MODEL": "deepseek-coder-v2"}'

### Environment Variables
The system automatically configures these variables for Claude Code:
```bash
ANTHROPIC_BASE_URL=http://127.0.0.1:12000  # Routes through Plano
ANTHROPIC_SMALL_FAST_MODEL=arch.claude.code.small.fast    # Uses intelligent alias
```

### Custom Routing Configuration
Edit `config.yaml` to define custom task→model mappings:

```yaml
llm_providers:
  # OpenAI Models
  - model: openai/gpt-5-2025-08-07
    access_key: $OPENAI_API_KEY
    routing_preferences:
      - name: code generation
        description: generating new code snippets, functions, or boilerplate based on user prompts or requirements

  - model: openai/gpt-4.1-2025-04-14
    access_key: $OPENAI_API_KEY
    routing_preferences:
      - name: code understanding
        description: understand and explain existing code snippets, functions, or libraries
```

## Technical Details

**How routing works:** Plano intercepts Claude Code requests, analyzes the content using preference-aligned routing, and forwards to the configured model.
**Research foundation:** Built on our research in [Preference-Aligned LLM Routing](https://arxiv.org/abs/2506.16655)
**Documentation:** [docs.planoai.dev](https://docs.planoai.dev) for advanced configuration and API details.