Strix setup: autonomous testing with GPT-5.4 or Llama-4

This tutorial walks through installing Strix, an open source AI penetration testing framework, and launching an assessment against an authorized application. The demonstrated setup requires a Unix-like terminal, Python 3.12 or higher, a running Docker daemon and an OpenAI API key. The presenter checks Docker, runs the quick-start installer and validates the CLI with its help command.

Strix runs its offensive utilities inside a prebuilt Kali Linux container. The speaker describes reconnaissance agents passing endpoint information to exploitation agents, followed by a validation agent that checks the exploit. The container provides isolation, though the video's claim that this makes autonomous exploitation entirely safe is not an established guarantee.

The model configuration uses OpenAI/GPT-5.4. The presenter also describes Llama-4 as a local LLM option for offline use, without walking through that alternative's setup or hardware requirements.

The scan example selects a target URL and quick mode, then sets an API spending limit. According to the tutorial, Strix stops the sandbox when aggregate spending reaches that limit. Local repositories exceeding 1,024 MB hit the default file-streaming limit; the mount option instead exposes the directory as read-only inside Docker. The final section covers reports with validated proof-of-concept findings and headless mode for CI/CD pull-request scans.