vxcontrol/pentagi ? reverse-engineered prompt

I want to build a self-hosted platform for doing automated penetration testing using AI. It should have a team of autonomous AI agents that can work together to test a target system. Some agents should be specialized for research, using search engines to gather information, while others are specialists at using security tools like nmap or Metasploit to find and exploit vulnerabilities.

The whole thing needs to run in a secure, sandboxed Docker environment so it's isolated. I need a modern web UI to manage the system, start tests, and monitor what the agents are doing. It should also have good observability with dashboards and generate a detailed report when the test is done. Make it flexible so I can plug in different LLM providers like OpenAI, Anthropic, or even a local one using Ollama. Package it all with Docker Compose for easy setup.