Watch on YouTube

Watch on YouTube

• Autonomous Agent - Agentic pipeline for intelligent, autonomous penetration testing
• Session Persistence - Save and resume penetration testing sessions
• Docker-First - Isolated, reproducible environment with security tools pre-installed

In Progress: Multi-model support for OpenAI, Gemini, and other LLM providers

• AI-Powered Challenge Solver - Leverages LLM advanced reasoning to perform penetration testing and CTFs
• Live Walkthrough - Tracks steps in real-time as the agent works through challenges
• Multi-Category Support - Web, Crypto, Reversing, Forensics, PWN, Privilege Escalation
• Real-Time Feedback - Watch the AI work with live activity updates
• Extensible Architecture - Clean, modular design ready for future enhancements

• Docker (required) - Install Docker
• LLM Provider (choose one):
  • Anthropic API Key from console.anthropic.com
  • Claude OAuth Login (requires Claude subscription)
  • OpenRouter for alternative models at openrouter.ai
  • Tutorial: Using Local Models with Claude Code

# Clone and build
git clone --recurse-submodules https://github.com/GreyDGL/PentestGPT.git
cd PentestGPT
make install

# Configure authentication (first time only)
make config

# Connect to container
make connect

Note: The --recurse-submodules flag downloads the benchmark suite. If you already cloned without it, run: git submodule update --init --recursive

cd benchmark/standalone-xbow-benchmark-runner
python3 run_benchmarks.py --range 1-1 --pattern-flag

See Benchmark Documentation for detailed usage.

# Interactive TUI mode (default)
pentestgpt --target 10.10.11.234

# Non-interactive mode
pentestgpt --target 10.10.11.100 --non-interactive

# With challenge context
pentestgpt --target 10.10.11.50 --instruction "WordPress site, focus on plugin vulnerabilities"

Keyboard Shortcuts: F1 Help | Ctrl+P Pause/Resume | Ctrl+Q Quit

PentestGPT supports routing requests to local LLM servers (LM Studio, Ollama, text-generation-webui, etc.) running on your host machine.

• Local LLM server with an OpenAI-compatible API endpoint
  • LM Studio: Enable server mode (default port 1234)
  • Ollama: Run ollama serve (default port 11434)

# Configure PentestGPT for local LLM
make config
# Select option 4: Local LLM

# Start your local LLM server on the host machine
# Then connect to the container
make connect

Edit scripts/ccr-config-template.json to customize:

• localLLM.api_base_url: Your LLM server URL (default: host.docker.internal:1234)
• localLLM.models: Available model names on your server
• Router section: Which models handle which operations

• Connection refused: Ensure your LLM server is running and listening on the configured port
• Docker networking: Use host.docker.internal (not localhost) to access host services from Docker
• Check CCR logs: Inside the container, run cat /tmp/ccr.log

PentestGPT collects anonymous usage data to help improve the tool. This data is sent to our Langfuse project and includes:

• Session metadata (target type, duration, completion status)
• Tool execution patterns (which tools are used, not the actual commands)
• Flag detection events (that a flag was found, not the flag content)

No sensitive data is collected - command outputs, credentials, or actual flag values are never transmitted.

# Via command line flag
pentestgpt --target 10.10.11.234 --no-telemetry

# Via environment variable
export LANGFUSE_ENABLED=false

PentestGPT includes 104 XBOW validation benchmarks for comprehensive testing and evaluation.

cd benchmark/standalone-xbow-benchmark-runner

python3 run_benchmarks.py --range 1-10 --pattern-flag   # Run benchmarks 1-10
python3 run_benchmarks.py --all --pattern-flag          # Run all 104 benchmarks
python3 run_benchmarks.py --retry-failed                # Retry failed benchmarks
python3 run_benchmarks.py --dry-run --range 1-5         # Preview without executing

PentestGPT achieved an 86.5% success rate (90/104 benchmarks) on the XBOW validation suite:

• Cost: Average $1.11, Median $0.42 per successful benchmark
• Time: Average 6.1 minutes, Median 3.3 minutes per successful benchmark
• Success rates by difficulty:
  • Level 1: 91.1%
  • Level 2: 74.5%
  • Level 3: 62.5%

For detailed benchmark results, analysis, and automated testing instructions, see the Benchmark Documentation.

The previous multi-LLM version (v0.15) supporting OpenAI, Gemini, Deepseek, and Ollama is archived in legacy/:

cd legacy && pip install -e . && pentestgpt --reasoning gpt-4o

If you use PentestGPT in your research, please cite our paper:

@inproceedings{299699,
  author = {Gelei Deng and Yi Liu and Víctor Mayoral-Vilches and Peng Liu and Yuekang Li and Yuan Xu and Tianwei Zhang and Yang Liu and Martin Pinzger and Stefan Rass},
  title = {{PentestGPT}: Evaluating and Harnessing Large Language Models for Automated Penetration Testing},
  booktitle = {33rd USENIX Security Symposium (USENIX Security 24)},
  year = {2024},
  isbn = {978-1-939133-44-1},
  address = {Philadelphia, PA},
  pages = {847--864},
  url = {https://www.usenix.org/conference/usenixsecurity24/presentation/deng},
  publisher = {USENIX Association},
  month = aug
}

Distributed under the MIT License. See LICENSE.md for more information.

Disclaimer: This tool is for educational purposes and authorized security testing only. The authors do not condone any illegal use. Use at your own risk.

• Research supported by Quantstamp and NTU Singapore

(back to top)