AI API Proxy Switch Skill

English | 简体中文

Overview

AI API Proxy Switch Skill is a Hermes Agent skill that allows users to dynamically switch between various advanced AI models provided by AI API Proxy, without modifying Hermes Agent's core configuration files.

Key Features

• Dynamic Model Switching: Switch between different AI models without modifying configuration files
• Multi-Provider Support: Support for major AI providers like OpenAI, Anthropic, Google
• Unified Interface: Consistent calling interface for different models
• Cost Monitoring: Real-time token consumption monitoring via Clawsocket dashboard
• Easy Configuration: Simple setup through environment variables

Quick Start

Installation

# Clone repository
git clone https://github.com/G-Joker-G/ai-api-proxy-switch.git

# Copy to skills directory
cp -r ai-api-proxy-switch ~/.hermes/skills/software-development/

Configuration

1. Register at api.clawsocket.com and get your API Key
2. Set environment variables:

# Temporary setup
export OPENROUTER_API_KEY='your_cla..._key'
export OPENROUTER_BASE_URL='https://api.clawsocket.com/v1'

# Or permanent setup (add to ~/.bashrc or ~/.zshrc)
echo 'export OPENROUTER_API_KEY="your_cla..._key"' >> ~/.bashrc
echo 'export OPENROUTER_BASE_URL="https://api.clawsocket.com/v1"' >> ~/.bashrc
source ~/.bashrc

Basic Usage

# Use GPT-5.2 for complex reasoning
hermes chat --skills ai-api-proxy-switch --model gpt-5.2 --query "Your question"

# Use Claude-Sonnet for creative writing
hermes chat --skills ai-api-proxy-switch --model claude-sonnet-4-6 --query "Write a poem"

# Test connection
hermes chat --skills ai-api-proxy-switch --query "/test-connection"

Supported Models

OpenAI Series

• gpt-5.2 - Latest GPT-5 model, strongest reasoning capability
• gpt-5-mini - Lightweight GPT-5 model, fast response
• gpt-5.3-codex - Code-optimized model

Anthropic Series

• claude-opus-4-6 - Latest Claude Opus version
• claude-sonnet-4-6 - Latest Claude Sonnet version
• claude-haiku-4-6 - Latest Claude Haiku version

Google Series

• gemini-3.1-pro-preview-thinking - Gemini deep thinking mode
• gemini-3.1-pro-preview - Gemini standard version

Troubleshooting

Common Issues

1. Error: Provider resolver returned an empty API key

   • Solution: Ensure OPENROUTER_API_KEY environment variable is set

2. Error: Model not available

   • Solution: Check model name, run /list-models to see available models

3. API call failed

   • Solution: Check if API Key is valid, network connection is working

Diagnostic Commands

# Check environment variables
echo $OPENROUTER_API_KEY

# Test API connection
curl -X GET "https://api.clawsocket.com/v1/models" \
  -H "Authorization: Bearer $OPENROUTER_API_KEY"

Best Practices

Model Selection Guide

Task Type	Recommended Model	Description
Complex Reasoning	gpt-5.2	Strongest reasoning and analysis
Creative Writing	claude-sonnet-4-6	Excellent creative expression
Code Generation	gpt-5.3-codex	Code-optimized
Fast Response	gpt-5-mini	Fast response, low cost
Deep Analysis	gemini-3.1-pro-preview-thinking	Deep thinking mode

Cost Optimization

1. Use lightweight models for simple tasks
2. Regularly monitor token consumption
3. Batch process related tasks

Advanced Features

Built-in Commands

• /list-models - View available model list
• /test-connection - Test API connection
• /show-config - Show current configuration

Batch Processing

for model in "gpt-5.2" "claude-sonnet-4-6"; do
    echo "=== Using $model ==="
    hermes chat --skills ai-api-proxy-switch --model $model --query "Question"
done

Contributing

Issues and Pull Requests are welcome!

License

MIT License