PyTorch-Jadex Integration: Intelligent BDI Agent with Python ML Model

This project demonstrates how to integrate a Jadex BDI agent with a Python-based PyTorch model to enable intelligent decision-making in agent-based systems.

Project Overview

The agent extracts feature values, sends them to a Python ML model via REST API, and uses the returned decision to act within its environment. This enables hybrid AI combining symbolic (BDI) and sub-symbolic (ML) methods.

---

Structure

├── jadex-bdi-agent/                # Java-based BDI agent (Jadex)
│   └── DecisionAgent.java          # Main agent with PyTorch integration
├── python-ml-server/              # Python REST API with PyTorch model
│   └── app.py                     # Flask server that serves model predictions
└── README.md                      # Project documentation

---

🚀 Setup Instructions

✅ Prerequisites

• Java 11+
• Jadex Platform (v3.0+)
• Python 3.8+
• Flask (pip install flask)
• PyTorch (pip install torch)

###⚙️ Run Instructions

1. Start the Python Flask Server

Navigate to python-ml-server/ and run:

python app.py

This will start a server on http://localhost:5000/predict that accepts JSON input like:

{
  "features": [0.1, 0.4, 0.6, 0.8]
}

And responds with a single integer decision, e.g. 1.

2. Run the Jadex Agent

Use Jadex or your Java IDE to run the agent defined in DecisionAgent.java. The agent will:

• Initialize with default feature values
• Trigger a goal
• Call the Python model
• Print the result

---

Example Communication

Request sent from agent:

{
  "features": [0.1, 0.4, 0.6, 0.8]
}

Response from Python server:

1

---

🛠️ Future Extensions

• Add feature extraction from environment
• Multiple agents with distributed decision making
• Secure communication between agent and server (e.g., token auth)
• Use FastAPI instead of Flask for async support

---

📚 References

• Jadex BDI Framework: https://www.activecomponents.org
• PyTorch: https://pytorch.org
• Flask: https://flask.palletsprojects.com