Summary Decepticon provides an autonomous, multi-agent red teaming service to counter evolving AI-driven cyber threats. It enables proactive defense through AI-powered offensive security, allowing human experts to focus on strategic oversight rather than manual execution.

How It Works Built on LangChain/LangGraph, Decepticon uses a multi-agent system architecture (Swarm, planned Supervisor/Hybrid). It features specialized Red Team agents (e.g., Reconnaissance, Initial Access) and Utility agents (e.g., Planner, Summary) that collaborate autonomously. The system integrates cloud/local AI models and uses the LangGraph MCP Adapter for flexible tool loading, ensuring rapid adaptation and scalability.

Quick Start & Requirements Installation requires cloning the repo and setting up dependencies via uv (uv venv, uv sync, uv pip install -e .). Configure .env.example with API keys for cloud models (OpenAI, Anthropic, OpenRouter) and LangSmith. Docker is supported (docker-compose up -d --build, ~10-20 min build). MCP servers can be run via scripts or manually. Launch the CLI (python frontend/cli/cli.py) or Streamlit web interface (streamlit run frontend/streamlit_app.py).

Highlighted Details

• Agent Specialization: Red Team agents (recon, initial access) and Utility agents (planner, summary); advanced agents planned.
• Flexible Architectures: Swarm (peer-to-peer), with Supervisor/Hybrid architectures planned.
• Replay Functionality: Enables community collaboration via exportable/replayable execution logs.
• MCP Support: Integrates tools via LangGraph MCP Adapter (mcp_config.json), supporting stdio and streamable_http.

Maintenance & Community Active community contributions are encouraged for migrating security tools to MCP, developing ReAct agents, and architecting multi-agent flows. A Discord server facilitates collaboration and support.

Licensing & Compatibility Licensed under the Apache-2.0 License, generally permissive for commercial use and closed-source integration.

Limitations & Caveats Decepticon is experimental, not yet stable, and may contain bugs or undergo breaking changes. Several advanced agent types and architectural patterns are still in planning stages.