Content
# 🏭 Multi-Agent Industrial AI System with A2A Protocol
Build a distributed multi-agent AI system for industrial automation using Google's Agent-to-Agent (A2A) protocol, enabling seamless communication between specialized AI agents.
## 🎯 Overview
Transform your monolithic industrial AI agent into a **distributed multi-agent system** where specialized agents collaborate through standardized communication protocols.
### The Evolution
**Previous Approach (Parts 1-4):**
- Single monolithic agent handling everything
- Complex, hard to maintain
- Limited scalability
**New Multi-Agent Approach:**
- **Specialized agents** with focused responsibilities
- **Standardized communication** via A2A protocol
- **Scalable architecture** - add agents as needed
- **Vendor agnostic** - agents can use different frameworks
### The Three Agents
1. **🔧 Equipment Monitoring Agent**
- Specializes in real-time data
- Connects to OPC UA server
- Monitors tank levels and machine states
2. **📊 Material Calculating Agent**
- Specializes in product knowledge
- Queries database for recipes
- Calculates material requirements
3. **🎯 Orchestrator Agent**
- Coordinates other agents
- Makes production decisions
- Provides unified interface
## ✨ Key Features
### A2A Protocol Implementation
- **Standardized messaging** using JSON-RPC 2.0
- **Dynamic agent discovery** through agent cards
- **Task orchestration** with state management
- **Bi-directional communication** - agents can be both clients and servers
### Core Capabilities
- ✅ Agent self-description via agent cards
- ✅ Automatic capability discovery
- ✅ Asynchronous task handling
- ✅ Multi-modal data exchange
- ✅ Built-in error handling
- ✅ Scalable architecture
## 🏗️ System Architecture
```mermaid
graph TB
subgraph "User Layer"
U[User CLI Application]
end
subgraph "Orchestration Layer"
O[Orchestrator Agent<br/>Port: 40004]
R[Agent Registry]
end
subgraph "Specialist Agents"
E[Equipment Monitoring Agent<br/>Port: 40002]
M[Material Calculating Agent<br/>Port: 40003]
end
subgraph "Data Sources"
OPC[OPC UA Server]
DB[TimescaleDB]
end
U -->|A2A Request| O
O -->|Discover| R
R -->|Agent Cards| O
O -->|Delegate Task| E
O -->|Delegate Task| M
E -->|Read Data| OPC
M -->|Query| DB
E -->|Response| O
M -->|Response| O
O -->|Final Decision| U
style O fill:#f9f,stroke:#333,stroke-width:4px
style E fill:#bbf,stroke:#333,stroke-width:2px
style M fill:#bfb,stroke:#333,stroke-width:2px
```
### A2A Protocol Components
**1. Agent Cards** - Self-description
```json
{
"name": "EquipmentMonitoringAgent",
"description": "Monitors material availability and machine states",
"url": "http://localhost:40002/",
"skills": [{
"id": "material-availability",
"name": "Get Material Levels",
"examples": ["What materials are available?"]
}]
}
```
**2. JSON-RPC Messages** - Standardized communication
```json
{
"jsonrpc": "2.0",
"method": "tasks/send",
"params": {
"id": "task-123",
"message": {
"role": "user",
"parts": [{"type": "text", "text": "Can I produce 4 batches?"}]
}
}
}
```
**3. Task Management** - Orchestrated workflow
- Task submission
- Progress tracking
- Result aggregation
## 📋 Prerequisites
- Python 3.8+
- PostgreSQL/TimescaleDB
- OPC UA Server (real or simulated)
- Anthropic API key (for Claude LLM)
## 🚀 Installation
### 1. Clone Repository
```bash
git clone https://github.com/yourusername/a2a-industrial-multiagent.git
cd a2a-industrial-multiagent
```
### 2. Create Virtual Environment
```bash
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
```
### 3. Install Dependencies
```bash
pip install -r requirements.txt
```
### 4. Configure Environment
Create `.env` file:
```env
# LLM Configuration
ANTHROPIC_API_KEY=your_anthropic_api_key
# Database Configuration
DB_USER=your_db_user
DB_PASSWORD=your_db_password
DB_HOST=localhost
DB_PORT=5432
# OPC UA Configuration
OPC_SERVER_URL=opc.tcp://desktop-fjjsr46:26543/BatchPlantServer
```
## 📖 Usage
### Starting the Multi-Agent System
**1. Start Equipment Monitoring Agent**
```bash
python -m agents.equipment_monitoring_agent --port 40002
```
Output:
```
INFO: EquipmentMonitoringAgent running at http://localhost:40002
INFO: A2A server started - listening for tasks
```
**2. Start Material Calculating Agent**
```bash
python -m agents.material_calculating_agent --port 40003
```
Output:
```
INFO: MaterialCalculatingAgent running at http://localhost:40003
INFO: A2A server started - listening for tasks
```
**3. Start Orchestrator Agent**
```bash
python -m agents.orchestrator_agent --port 40004
```
Output:
```
INFO: Discovering agents from registry...
INFO: ✓ EquipmentMonitoringAgent discovered
INFO: ✓ MaterialCalculatingAgent discovered
INFO: OrchestratorAgent running at http://localhost:40004
```
**4. Start User CLI Application**
```bash
python -m app.cmd --target http://localhost:40004
```
### Example Interactions
**Direct Agent Query:**
```bash
# Connect to Equipment Monitoring Agent
python -m app.cmd --target http://localhost:40002
> What is the material availability and machine states?
Agent says:
{
"machine_states": {
"mixer_state": "running",
"reactor_state": "idle",
"filler_state": "idle"
},
"material_availability": {
"tank1_material_level": 8543.21,
"tank2_material_level": 12456.78,
"tank3_material_level": 9823.45
}
}
```
**Orchestrated Query:**
```bash
# Connect to Orchestrator
python -m app.cmd --target http://localhost:40004
> Can I produce 4 batches of Product A?
Agent says:
{
"decision": "Yes",
"reasoning": "Sufficient materials and operational equipment available",
"sufficient_materials": true,
"machine_states": {...},
"material_availability": {...},
"tools_used": ["MaterialCalculatingAgent", "EquipmentMonitoringAgent"]
}
```
## 📁 Project Structure
```
a2a-industrial-multiagent/
├── agents/ # Agent implementations
│ ├── equipment_monitoring_agent/
│ │ ├── __main__.py # Entry point & A2A server
│ │ ├── agent.py # LangChain agent logic
│ │ ├── opcua_data_access.py # OPC UA client
│ │ ├── task_manager.py # A2A task handler
│ │ └── tools.py # LangChain tools
│ │
│ ├── material_calculating_agent/
│ │ ├── __main__.py # Entry point & A2A server
│ │ ├── agent.py # LangChain agent logic
│ │ ├── database_data_access.py # DB queries
│ │ ├── task_manager.py # A2A task handler
│ │ └── tools.py # LangChain tools
│ │
│ └── orchestrator_agent/
│ ├── __main__.py # Entry point & A2A server
│ ├── agent.py # Orchestration logic
│ ├── agent_connect.py # Agent discovery/connection
│ └── task_manager.py # A2A task handler
│
├── app/ # User applications
│ ├── __init__.py
│ └── cmd.py # CLI interface
│
├── shared/ # Shared A2A implementation
│ ├── a2a/
│ │ ├── agent_registry.json # Agent registry
│ │ ├── client.py # A2A client
│ │ ├── discovery.py # Agent discovery
│ │ ├── server.py # A2A server
│ │ └── task_manager.py # Task management
│ │
│ └── models/ # A2A data models
│ ├── agent.py # AgentCard, Skills
│ ├── json_rpc.py # JSON-RPC models
│ ├── request.py # Request/Response
│ └── task.py # Task models
│
├── .env # Environment variables
└── requirements.txt # Dependencies
```
## 🔄 How It Works
### 1. Agent Registration & Discovery
```python
# Agent publishes its card at /.well-known/agent.json
agent_card = AgentCard(
name="EquipmentMonitoringAgent",
skills=[AgentSkill(id="monitor", name="Monitor Equipment")]
)
# Orchestrator discovers agents
discovery = DiscoveryClient()
agents = await discovery.list_agent_cards()
```
### 2. Task Delegation
```python
# Orchestrator delegates to specialist
connector = AgentConnector("EquipmentMonitoringAgent", "http://localhost:40002")
result = await connector.send_task("Check material levels", session_id)
```
### 3. Response Aggregation
```python
# Combine results from multiple agents
materials = await material_agent.check_requirements()
equipment = await equipment_agent.check_availability()
decision = analyze(materials, equipment)
```
## 🧪 Testing
### Unit Tests
```bash
pytest tests/
```
### Integration Test
```python
# Test agent communication
async def test_agent_communication():
# Start agents
equipment_agent = EquipmentMonitoringAgent()
orchestrator = OrchestratorAgent()
# Send task
response = await orchestrator.invoke("Can we produce 5 batches?")
assert response["decision"] in ["Yes", "No"]
```
### Health Check
```bash
# Check if all agents are running
curl http://localhost:40002/.well-known/agent.json
curl http://localhost:40003/.well-known/agent.json
curl http://localhost:40004/.well-known/agent.json
```
## 🚀 Extending the System
### Adding New Agents
1. **Create Agent Structure**
```python
class NewSpecialistAgent:
SUPPORTED_CONTENT_TYPES = ["text"]
async def invoke(self, query: str) -> str:
# Agent logic here
return result
```
2. **Define Agent Card**
```python
agent_card = AgentCard(
name="NewSpecialistAgent",
skills=[AgentSkill(id="new-skill")]
)
```
3. **Register in Registry**
```json
{
"agents": [
{"url": "http://localhost:40005"}
]
}
```
## 🚧 Troubleshooting
### Common Issues
**1. Agent Discovery Fails**
- Verify all agents are running
- Check agent_registry.json paths
- Ensure network connectivity
**2. Task Timeout**
```python
# Increase timeout in orchestrator
self.child_agent_timeout = 60.0 # seconds
```
**3. JSON-RPC Errors**
- Validate message format
- Check method names match exactly
- Ensure params structure is correct
## 🛣️ Roadmap
- [ ] Add WebSocket support for real-time updates
- [ ] Implement agent health monitoring dashboard
- [ ] Add authentication/authorization
- [ ] Create agent deployment with Docker
- [ ] Build web-based orchestration UI
- [ ] Add more specialized agents (maintenance, quality, etc.)
## 🤝 Contributing
Contributions welcome! Please:
1. Fork the repository
2. Create a feature branch
3. Implement with tests
4. Submit a Pull Request
## 📝 License
MIT License - see [LICENSE](LICENSE) file
## 🙏 Acknowledgments
- **Kudzai Manditereza** - Tutorial creator and Industrial AI Expert
- **Google** - A2A Protocol specification
- **Anthropic** - Claude LLM
- **LangChain** - Agent framework
## 🔗 Resources
- [Video Tutorial](https://youtu.be/9gmpHMjwbJs)
- [A2A Protocol Spec](https://github.com/google/agent-to-agent)
- [LangChain Docs](https://python.langchain.com/)
- [OPC UA Foundation](https://opcfoundation.org/)
## 📧 Contact
- YouTube: [Kudzai Manditereza](https://www.youtube.com/@industry40tvonline)
- LnkedIn: [Kudzai Manditereza](https://www.linkedin.com/in/kudzaimanditereza/)
---
**⭐ Star this repo to support multi-agent industrial AI development!**
