PentestGPT

An LLM-powered penetration testing framework built around a 5-agent orchestration loop with multi-provider LLM routing.

Paper: AutoPentester: arXiv:2510.05605

Quick Start

git clone <repo-url>
cd AutoPentester

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip install -e .

cp .env.example .env
# Add at least one API key to .env (OPENAI_API_KEY, ANTHROPIC_API_KEY, or XAI_API_KEY)

pentestgpt --target 10.10.10.100 --preset high-quality

Run pentestgpt --help for all command-line options, or see the getting-started guide.

For Docker:

docker-compose build
docker-compose up -d
docker attach pentestgpt-app

How It Works

PentestGPT uses five cooperating agents to automate penetration testing:

User → Orchestrator
         ├── Strategy Agent        — MITRE-aligned task planning
         ├── Repetition Detector   — loop detection
         ├── Command Generator     — RAG-enhanced command synthesis
         ├── Results Verifier      — output parsing and error detection
         └── Tool Executor         — sandboxed command execution

Tasks are routed to the best-suited LLM provider (OpenAI, Anthropic, xAI, or local Ollama) based on task type and cost constraints.

Documentation

Full documentation lives in docs/:

• Getting Started — Installation, configuration, first run
• Architecture — System design, agent communication, diagrams
• API Reference — REST API endpoints
• Configuration — Config options and environment variables
• Contributing — How to contribute
• Deployment — Production deployment guide

License

MIT — see LICENSE.