Semantic Kernel Agent 2 Agent Protocol Demo

This console application demonstrates the implementation of Agent 2 Agent protocol using Microsoft Semantic Kernel with Azure OpenAI. The application supports both built-in AI agents and custom deployed agents across AWS, Azure, and GCP platforms.

Features

• Agent 2 Agent Protocol: Implements Google's Agent 2 Agent communication protocol
• Custom Agent Integration: Connect to your own deployed AI agents across cloud platforms
• Multi-Cloud Agents: Built-in agents representing AWS, Azure, and GCP platforms
• Configurable Workflow: Mix and match custom and built-in agents
• Sequential Workflow: Tasks are delegated across agents in sequence
• Unique Agent Identities: Each agent has a unique human name and favorite rainbow color
• Response Aggregation: Collects and summarizes responses from all agents
• Error Handling: Graceful handling of agent failures with fallback responses

Setup

1. Install Dependencies:

   pip install -r requirements.txt

2. Configure Azure OpenAI:

   • Update the .env file with your Azure OpenAI credentials:
     • AZURE_OPENAI_API_KEY: Your Azure OpenAI API key
     • AZURE_OPENAI_ENDPOINT: Your Azure OpenAI endpoint
     • AZURE_OPENAI_DEPLOYMENT_NAME: Your GPT model deployment name
     • AZURE_OPENAI_API_VERSION: API version (default: 2024-11-20)

3. Configure Custom Agents (Optional):

   python config_manager.py

   Or manually edit custom_agents.env to add your deployed agents.

4. Run the Application:

   python main.py

Custom Agent Configuration

Adding Custom Agents

You can integrate your own deployed AI agents using the configuration manager:

python config_manager.py

This interactive tool allows you to:

• Add custom agent endpoints
• Enable/disable custom agents
• Test agent connectivity
• View configuration status

Manual Configuration

Edit custom_agents.env to add your agents:

# Enable custom agents
USE_CUSTOM_AGENTS=true
USE_BUILTIN_AGENTS=true

# AWS Custom Agent Example
AWS_AGENT_1_NAME=MyCustomAWS-DataProcessor
AWS_AGENT_1_ENDPOINT=https://my-aws-agent.amazonaws.com/api/chat
AWS_AGENT_1_API_KEY=your-api-key
AWS_AGENT_1_COLOR=Red

Custom Agent API Requirements

Your deployed agents should accept POST requests with this JSON structure:

{
  "message": "task description",
  "context": "additional context",
  "agent_info": {
    "name": "agent name",
    "platform": "cloud platform", 
    "favorite_color": "color"
  },
  "metadata": {
    "timestamp": 1234567890,
    "session_id": "session_id"
  }
}

And return responses in one of these formats:

{"message": "response text"}
{"response": "response text"}  
{"content": "response text"}
{"data": {"message": "response text"}}

Or OpenAI-compatible format with choices array.

Agents

Built-in Agents (Semantic Kernel + Azure OpenAI)

1. Alexandra (AWS) - Favorite Color: Red
2. Benjamin (Azure) - Favorite Color: Orange
3. Catherine (GCP) - Favorite Color: Yellow
4. David (AWS) - Favorite Color: Green
5. Elena (Azure) - Favorite Color: Blue
6. Francis (GCP) - Favorite Color: Indigo
7. Gabriel (Multi-Cloud) - Favorite Color: Violet

Custom Agent Support

• AWS: Amazon EC2, ECS, Lambda deployments
• Azure: App Service, Container Instances, Functions
• GCP: Cloud Run, Compute Engine, Cloud Functions
• Multi-Cloud: Cross-platform orchestrators

Workflow

1. Agent Discovery: Load configured custom and built-in agents
2. Task Delegation: Main task is delegated to each agent sequentially
3. Agent Processing: Each agent processes the task while maintaining their identity
4. Response Collection: Responses are collected from both custom and built-in agents
5. Aggregation: Final report aggregates all agent responses with platform and type information

Configuration Options

Agent Workflow Settings

• USE_CUSTOM_AGENTS: Enable/disable custom deployed agents
• USE_BUILTIN_AGENTS: Enable/disable built-in Semantic Kernel agents
• CUSTOM_AGENT_TIMEOUT: Timeout for custom agent requests (seconds)
• CUSTOM_AGENT_RETRIES: Number of retry attempts for failed requests

Testing and Management

• Test Setup: python test_setup.py
• Demo Mode: python demo.py (no API keys required)
• Config Manager: python config_manager.py
• Connectivity Test: Built into config manager

Output

The application produces:

• Real-time progress updates during workflow execution
• Agent type identification (Custom vs Built-in)
• Individual agent responses with cloud platform and favorite color
• Final aggregated report summarizing all agent contributions
• Cloud platform grouping and summary statistics
• Error handling and fallback responses for failed agents