Skip to content
This repository was archived by the owner on Aug 5, 2025. It is now read-only.

OPENROUTER_MIGRATION.md

Latest commit

 

History

History
264 lines (204 loc) · 7.31 KB

OPENROUTER_MIGRATION.md

File metadata and controls

264 lines (204 loc) · 7.31 KB

OpenRouter Migration Guide

This guide helps you migrate your workflows from Claude Code CLI to OpenRouter, giving you access to 400+ AI models from multiple providers.

Overview

Benefits of Migration:

  • 🌐 Access to 400+ models from providers like OpenAI, Anthropic, Google, Meta, Mistral
  • 💰 Cost optimization with automatic model routing and fallbacks
  • 🔑 Single API key instead of managing multiple provider keys
  • 🚀 Better performance with direct API calls vs CLI subprocess overhead
  • 🎯 Model flexibility - switch models without changing your code

Quick Start

1. Install Dependencies

pip install openai>=1.0.0  # OpenRouter uses OpenAI SDK

2. Get OpenRouter API Key

  1. Sign up at openrouter.ai
  2. Get your API key from the dashboard
  3. Set environment variable:
export OPENROUTER_API_KEY="your-api-key-here"

3. Choose Your Migration Path

Option A: Use New OpenRouter Workflows (Recommended)

# Research workflow
python workflows/research/research_openrouter.py "your research topic"

# Code review workflow  
python workflows/code_review/review_openrouter.py 123 --repository owner/repo

# JIRA task workflow
python workflows/jira_task/enhanced_workflow_openrouter.py PROJ-123 "implement feature"

Option B: Replace Claude Code Client in Existing Workflows

Use the OpenRouterClient class to replace subprocess.run(["claude", ...]) calls:

from openrouter_client import OpenRouterClient, WorkflowType

# Replace Claude Code CLI calls
client = OpenRouterClient(workflow_type=WorkflowType.RESEARCH)
result = client.execute_prompt("Your prompt here")

Recommended Models by Workflow

Research Workflow

  • Primary: perplexity/llama-3.1-sonar-large-128k-online ($0.001/1K tokens)
    • Excellent for research with web access
  • Alternative: anthropic/claude-3-haiku ($0.00025/1K tokens)
    • Fast and cost-effective

Code Review Workflow

  • Primary: anthropic/claude-3.5-sonnet ($0.003/1K tokens)
    • Best for code analysis and complex reasoning
  • Alternative: deepseek/deepseek-coder ($0.00014/1K tokens)
    • Specialized for code review tasks

JIRA Task Development

  • Primary: anthropic/claude-3.5-sonnet ($0.003/1K tokens)
    • Excellent for development tasks
  • Alternative: openai/gpt-4-turbo ($0.01/1K tokens)
    • Great for comprehensive analysis

Migration Examples

Before: Claude Code CLI

import subprocess

command = ["claude", "-p", "--verbose", "--model", "sonnet", "--allowedTools", "bash,webSearch"]
result = subprocess.run(command, input=prompt, capture_output=True, text=True, timeout=1800)
response = result.stdout.strip()

After: OpenRouter API

from openrouter_client import OpenRouterClient, WorkflowType

client = OpenRouterClient(workflow_type=WorkflowType.RESEARCH)
result = client.execute_prompt(prompt, timeout=1800)
response = result["content"]

Workflow-Specific Migration

Research Workflow Migration

Old Command:

python workflows/research/research.py "AI trends 2024"

New Command:

python workflows/research/research_openrouter.py "AI trends 2024"

Key Differences:

  • ✅ Direct API calls (faster)
  • ✅ Better error handling
  • ✅ Cost estimation
  • ✅ Multiple model options
  • ✅ Usage tracking

Code Review Migration

Old Command:

python workflows/code_review/review.py 123 --repository owner/repo

New Command:

python workflows/code_review/review_openrouter.py 123 --repository owner/repo

Enhanced Features:

  • --list-models: See recommended models
  • --estimate-cost: Estimate review cost before running
  • --model: Specify exact model to use

JIRA Task Migration

Old Command:

python workflows/jira_task/enhanced_workflow.py

New Command:

python workflows/jira_task/enhanced_workflow_openrouter.py PROJ-123 "implement user authentication"

Enhanced Features:

  • Context preservation across interactions
  • Dynamic model switching
  • Better error recovery
  • Cost tracking

Configuration

Environment Variables

# Required
export OPENROUTER_API_KEY="your-api-key"

# Optional (for OpenRouter leaderboards)
export OPENROUTER_SITE_URL="https://github.com/yourusername/yourproject"
export OPENROUTER_SITE_NAME="Your Project Name"

Model Selection

# List available models for a workflow type
python -c "
from openrouter_client import OpenRouterModels, WorkflowType
models = OpenRouterModels.get_recommended_models(WorkflowType.RESEARCH)
for model in models:
    print(f'{model.name}: {model.description}')
"

Cost Comparison

Provider Model Cost per 1K tokens Best for
Claude Code claude-3.5-sonnet $0.003 All workflows
OpenRouter anthropic/claude-3.5-sonnet $0.003 Same model, more flexibility
OpenRouter anthropic/claude-3-haiku $0.00025 Cost-effective tasks
OpenRouter deepseek/deepseek-coder $0.00014 Code-focused tasks
OpenRouter perplexity/sonar-large $0.001 Research with web access

Troubleshooting

Common Issues

"OpenRouter API key required"

export OPENROUTER_API_KEY="your-key-here"
# Or pass directly:
python script.py --api-key "your-key-here"

"Model not found"

# List available models
python script.py --list-models

Import errors

pip install openai>=1.0.0

Getting Help

  1. Check model availability: Use --list-models flag
  2. Estimate costs: Use --estimate-cost flag
  3. Enable verbose logging: Use --verbose flag
  4. Test with simple prompts: Start with basic requests

Advanced Usage

Custom Model Configuration

from openrouter_client import OpenRouterClient

client = OpenRouterClient()
# Use specific model
result = client.execute_prompt(
    prompt="your prompt", 
    model="anthropic/claude-3.5-sonnet",
    max_tokens=4096,
    temperature=0.7
)

Cost Estimation

cost_info = client.estimate_cost("your prompt", model="anthropic/claude-3.5-sonnet")
print(f"Estimated cost: ${cost_info['estimated_cost']:.4f}")

Multiple Models in One Workflow

# Use different models for different tasks
research_client = OpenRouterClient(workflow_type=WorkflowType.RESEARCH)
code_client = OpenRouterClient(workflow_type=WorkflowType.CODE_REVIEW)

# Each optimized for their respective tasks
research_result = research_client.execute_prompt("research prompt")
code_result = code_client.execute_prompt("code review prompt")

Next Steps

  1. Start with one workflow: Migrate your most-used workflow first
  2. Test with small tasks: Verify functionality before large projects
  3. Monitor costs: Use the cost estimation features
  4. Experiment with models: Try different models for different use cases
  5. Optimize for your needs: Adjust models based on quality vs cost preferences

Support

Ready to migrate? Start with the research workflow using:

python workflows/research/research_openrouter.py "test migration" --list-models