Summary

Cyber-AutoAgent is an AI-driven penetration testing tool designed for autonomous cyber operations. It enables users to conduct security assessments with natural language reasoning, dynamic tool selection, and automated evidence collection, leveraging large language models. The project targets security professionals and researchers seeking to automate and enhance their penetration testing workflows.

How It Works

The agent operates on the Strands framework, employing a "meta-everything" architecture for dynamic adaptation. It uses metacognitive reasoning to analyze situations, select appropriate security tools (from standard pentesting utilities to dynamically created meta-tools), and execute actions. Evidence is collected and stored persistently via the Mem0 memory system. The system supports multiple LLM providers, including AWS Bedrock, local Ollama, or others via LiteLLM, allowing flexibility in deployment and privacy. An adaptive execution cycle guides strategy based on confidence levels, from direct tool use to swarm intelligence for complex tasks.

Quick Start & Requirements

Installation offers local development via Node.js (v20+) and Python (v3.10+) or containerized deployment using Docker. The recommended interactive mode utilizes a React-based terminal interface.

• Local: Clone repo, npm install (React UI), npm start.
• Docker: docker run ... cyberautoagent/cyber-autoagent or docker compose ... run cyber-autoagent.
• Prerequisites: Node.js v20+, Python v3.10+, Docker. macOS users require Xcode Command Line Tools. Cloud credentials (AWS) or local LLM setup (Ollama) are needed depending on the provider. Optional security tools like nmap, sqlmap may need pre-installation unless running Docker as root.
• Documentation: User Guide, Architecture Guide, Memory System Guide, Observability Guide, Deployment Guide are available within the docs/ directory.
• Community: Discord server available at https://discord.gg/WNHhsnkTc3.

Highlighted Details

• Autonomous Operation: Conducts security assessments with minimal human intervention.
• Intelligent Tooling: Dynamically selects and creates security tools, including meta-tools for novel challenges.
• Metacognitive Reasoning: Adapts strategy based on confidence levels and continuous self-reflection.
• Comprehensive Observability: Integrated Langfuse tracing and Ragas evaluation metrics provide deep insights into agent operations and performance.
• Swarm Intelligence: Supports deploying parallel agents for complex, distributed assessments.
• Flexible Model Providers: Supports AWS Bedrock, local Ollama, and numerous others via LiteLLM.

Maintenance & Community

The project shows active development with GitHub statistics indicating contributors and pull requests. A Discord community server is available for support and discussion.

Licensing & Compatibility

The project is licensed under the permissive MIT License, allowing for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

This is EXPERIMENTAL SOFTWARE and must be used only in authorized, safe, sandboxed environments. Users are solely responsible for legal and ethical compliance. Dynamic tool installation requires running the Docker container as root, which reduces security isolation. Setup complexity can be a barrier, requiring familiarity with Node.js, Python, Docker, and LLM configurations.