Skip to content

Welcome to the CybMASDE Documentation

CybMASDE (Cyber Multi-Agent System Development Environment) is a modular platform designed to model, train, analyze, and deploy intelligent multi-agent systems (MAS) based on the MAMAD methodology (MOISE+MARL Assisted MAS Design).
It combines performance from multi-agent reinforcement learning (MARL) and organizational modeling (MOISE+) for controlling and explaining emerging agents' behaviour with practical tools for experimentation, transfer, and explainability in complex cyber-physical environments.

CybMASDE can be used:

  • 🧠 As a research framework, to study agent autonomy, coordination, adaptation, and explainability.
  • ⚙️ As an engineering tool, to design and deploy agent-based behaviors in real or simulated infrastructures.

  • 💻 Through three interfaces:

  • a CLI (Command Line Interface) for automation and batch workflows,

  • a Python library for direct integration into experiments,

  • and a Graphical User Interface (GUI) built with Angular for visual project editing and monitoring.

Table of Contents

Overview

CybMASDE was designed to bridge the gap between research prototypes and operational systems by providing:

  • A structured workflow for the lifecycle of multi-agent systems (from simulation to deployment).
  • An organizationally-aware reinforcement learning engine (MOISE+MARL).
  • Integrated support for world models, multi-agent policy training, and automatic explainability (Auto-TEMM).
  • A transfer component to synchronize simulated and real environments (via REST APIs).

Each phase of the MAMAD method is implemented as a separate module:

  1. Modeling: world model training or handcrafted environment definition.
  2. Training: organizationally guided MARL policy optimization.
  3. Analyzing: behavioral explainability and stability assessment.
  4. Transferring: policy deployment and iterative refinement in the target environment.

Example Workflow

A typical workflow using the CLI may look like this:

# Create and validate a new project

cybmasde init -n overcooked_project --template worldmodel
cybmasde validate

# Run the full MAMAD pipeline automatically

cybmasde run --full-auto --reward-threshold 3.5 --max-refine 5

# Deploy the learned joint policy remotely

cybmasde deploy --remote --api http://localhost:8080/api