OpenRouter

Overview

OpenRouter provides a single API to access models from multiple providers. When integrated with Autohand, you get:

• Access to 200+ models from Anthropic, OpenAI, Google, Meta, and more
• Automatic fallback when a model is unavailable
• Cost optimization with model routing
• Single API key for all providers
• Real-time model availability and pricing

Setup

Get started with OpenRouter in minutes.

Get your API key

1. Go to openrouter.ai and create an account
2. Navigate to Keys in your dashboard
3. Click Create Key and copy your API key

Configure Autohand

Set your API key using environment variable or config file:

# Set environment variable (recommended)
export OPENROUTER_API_KEY="sk-or-v1-xxxxxxxxxxxx"

# Or configure via CLI
autohand config set openrouter.apiKey "sk-or-v1-xxxxxxxxxxxx"

Verify your configuration:

# Check current provider and model
autohand config get provider
autohand config get model

# Test with a simple prompt
autohand --prompt "Hello, which model are you?"

CLI configuration

Configure OpenRouter in your ~/.autohand/config.json:

{
  "provider": "openrouter",
  "openrouter": {
    "apiKey": "${OPENROUTER_API_KEY}",
    "model": "nvidia/nemotron-3-super-120b-a12b:free",
    "fallbackModels": [
      "anthropic/claude-3.5-sonnet",
      "openai/gpt-4o"
    ],
    "maxTokens": 4096,
    "temperature": 0.7,
    "siteUrl": "https://your-app.com",
    "siteName": "Your App"
  }
}

Configuration options

Available models

OpenRouter provides access to models from all major providers. Here are the recommended ones for coding:

Anthropic Claude

OpenAI GPT

Google Gemini

Open-source models

Switch models

# Set default model
autohand config set model "nvidia/nemotron-3-super-120b-a12b:free"

# Use a specific model for one session
autohand --model "openai/gpt-4o"

# Switch during a session
/model openai/gpt-4o

Features

Model fallback

Configure fallback models for automatic failover:

{
  "openrouter": {
    "model": "nvidia/nemotron-3-super-120b-a12b:free",
    "fallbackModels": [
      "anthropic/claude-3.5-sonnet",
      "openai/gpt-4o",
      "google/gemini-pro-1.5"
    ]
  }
}

Cost tracking

Monitor your usage and costs:

# View session costs
/stats

# Check account balance
curl -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  https://openrouter.ai/api/v1/auth/key

Model routing

Let OpenRouter choose the best model for your query:

{
  "openrouter": {
    "model": "openrouter/auto",
    "route": "fallback"
  }
}

Provider transforms

OpenRouter handles provider-specific features automatically:

System prompts

System prompts are converted to the correct format for each provider:

{
  "openrouter": {
    "transforms": ["middle-out"],
    "systemPrompt": "You are an expert software engineer."
  }
}

Context handling

Long conversations are automatically truncated to fit model limits:

{
  "openrouter": {
    "maxContextTokens": 100000,
    "truncateStrategy": "middle-out"
  }
}

Best practices

• Use environment variables: Never commit API keys to version control.
• Set fallback models: Ensure availability with multiple fallback options.
• Match model to task: Use Claude for coding, GPT-4o for multimodal, Gemini for large context.
• Monitor costs: Check usage regularly with /stats command.
• Use site attribution: Set siteUrl and siteName for model leaderboards.

Recommended configurations

Coding tasks

{
  "openrouter": {
    "model": "nvidia/nemotron-3-super-120b-a12b:free",
    "temperature": 0.3,
    "maxTokens": 8192
  }
}

Creative tasks

{
  "openrouter": {
    "model": "anthropic/claude-3-opus",
    "temperature": 0.9,
    "maxTokens": 4096
  }
}

Large codebase analysis

{
  "openrouter": {
    "model": "google/gemini-pro-1.5",
    "maxContextTokens": 500000
  }
}

Troubleshooting

Common issues

Check API status

# List available models
curl https://openrouter.ai/api/v1/models

# Check your credits
curl -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  https://openrouter.ai/api/v1/auth/key

Debug mode

# Enable verbose logging
AUTOHAND_DEBUG=true autohand --prompt "Test"

# View request/response details
autohand --verbose