Athena

Athena, ancient Greek goddess associated with wisdom, warfare, and handicraft.

A proxy server that maps Anthropic's API format to OpenAI API format, allowing you to use Claude Code with OpenRouter's vast selection of AI models.

Features

• 🔄 API Translation: Maps Anthropic API calls to OpenRouter format
• 🌊 Streaming Support: Full SSE streaming for real-time responses
• 🛠️ Tool Calling: Complete function/tool calling support
• 🎯 Model Mapping: Configurable mappings for Opus, Sonnet, and Haiku models
• 🔀 Provider Routing: Automatic Groq provider routing for Kimi K2 models
• ⚙️ Flexible Configuration: CLI flags, config files, environment variables, and .env files
• 🚀 Minimal Dependencies: Lightweight with only essential external packages (Cobra CLI, YAML parser)

Quick Start

One-line install:

curl -fsSL https://raw.githubusercontent.com/martinffx/athena/main/install.sh | bash

Manual installation:

1. Download the latest release from GitHub Releases
2. Extract and move to your PATH
3. Set up configuration (see below)

Configuration

The proxy looks for configuration in this priority order (highest to lowest):

1. Command line flags - CLI arguments override everything
2. Environment variables - ATHENA_* prefixed env vars
3. Local config file - ./athena.yml in current directory
4. Global config file - ~/.config/athena/athena.yml
5. Built-in defaults - Hardcoded fallback values

Config File Example (YAML):

# ~/.config/athena/athena.yml
port: "12377"
api_key: "your-openrouter-api-key-here"
base_url: "https://openrouter.ai/api"
model: "moonshotai/kimi-k2-0905"
opus_model: "anthropic/claude-3-opus"
sonnet_model: "anthropic/claude-3.5-sonnet"
haiku_model: "anthropic/claude-3.5-haiku"

Note: The default model moonshotai/kimi-k2-0905 automatically uses Groq provider routing for optimal performance.

Advanced: Provider Routing

For fine-grained control over provider routing, add provider configurations to your YAML config:

port: "12377"
api_key: "your-openrouter-api-key-here"
base_url: "https://openrouter.ai/api"
model: "moonshotai/kimi-k2-0905"
default_provider:
  order:
    - Groq
  allow_fallbacks: false
opus_model: "anthropic/claude-3-opus"
opus_provider:
  order:
    - Anthropic
  allow_fallbacks: true
sonnet_model: "anthropic/claude-3.5-sonnet"
haiku_model: "anthropic/claude-3.5-haiku"

Provider routing allows you to:

• Force requests through specific providers (e.g., Groq, Anthropic)
• Control fallback behavior when the primary provider is unavailable
• Configure different providers for different model tiers

Environment Variables:

export OPENROUTER_API_KEY="your-key"
export OPUS_MODEL="anthropic/claude-3-opus"
export SONNET_MODEL="anthropic/claude-3.5-sonnet"
export HAIKU_MODEL="anthropic/claude-3.5-haiku"
export DEFAULT_MODEL="moonshotai/kimi-k2-0905"
export PORT="12377"

.env File:

# ./.env
OPENROUTER_API_KEY=your-openrouter-api-key-here
OPUS_MODEL=anthropic/claude-3-opus
SONNET_MODEL=anthropic/claude-3.5-sonnet
HAIKU_MODEL=anthropic/claude-3.5-haiku

Usage

Daemon Management

# Run in foreground (default)
athena

# Run as background daemon
athena start

# Stop daemon
athena stop

# Check daemon status
athena status

# View logs (daemon mode)
tail -f ~/.athena/athena.log

Custom Configuration

# Use specific models and port (foreground)
athena -port 9000 -api-key YOUR_KEY

# Or run as daemon with custom port
athena start -port 9000 -api-key YOUR_KEY

# Enable debug logging to see full request/response bodies
athena --log-level debug

Using with Claude Code

# Start Athena daemon
athena start

# Configure Claude Code to use the proxy
export ANTHROPIC_BASE_URL=http://localhost:12377
export ANTHROPIC_API_KEY=your-openrouter-key

# Run Claude Code
claude

How It Works

The proxy server:

1. Receives Anthropic API calls from Claude Code on /v1/messages
2. Transforms the request format to OpenAI-compatible format
3. Forwards to OpenRouter's /v1/chat/completions endpoint
4. Converts the response back to Anthropic format
5. Streams the response back to Claude Code

Model Mapping

When Claude Code requests a model:

• claude-3-opus* → Your configured opus_model
• claude-3.5-sonnet* → Your configured sonnet_model
• claude-3.5-haiku* → Your configured haiku_model
• Models with / (e.g., openai/gpt-4) → Passed through as-is
• Other models → Your configured default_model

Building from Source

git clone https://github.com/martinffx/athena.git
cd athena
go build -o athena ./cmd/athena

API Compatibility

The proxy provides a fully compatible Anthropic Messages API that supports:

• ✅ Text generation
• ✅ Streaming responses
• ✅ System messages
• ✅ Tool/function calling
• ✅ Multi-turn conversations
• ✅ Content blocks (text, tool_use, tool_result)
• ✅ Usage tracking
• ✅ Stop reasons

Endpoints

• POST /v1/messages - Anthropic Messages API (proxied to OpenRouter)
• GET /health - Health check endpoint

Supported Platforms

• Linux (AMD64, ARM64)
• macOS (Intel, Apple Silicon)
• Windows (AMD64, ARM64)

Configuration Examples

Use Claude Code with GPT-4:

opus_model: "openai/gpt-4"
sonnet_model: "openai/gpt-4-turbo"
haiku_model: "openai/gpt-3.5-turbo"

Use Claude Code with Gemini:

opus_model: "google/gemini-pro"
sonnet_model: "google/gemini-pro"
haiku_model: "google/gemini-pro"

Use Claude Code with Local Ollama:

base_url: "http://localhost:12377/v1"
opus_model: "llama3:70b"
sonnet_model: "llama3:8b" 
haiku_model: "llama3:8b"

Troubleshooting

Port already in use:

# Use a different port
athena -port 9000

API key issues:

# Check if key is set
echo $OPENROUTER_API_KEY

# Test the proxy directly
curl -X POST http://localhost:12377/v1/messages \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: your-key" \
  -d '{"model":"claude-3-sonnet","messages":[{"role":"user","content":"Hi"}]}'

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.