vxcontrol/pentagi

What it solves

PentAGI provides an autonomous, self-hosted platform for penetration testing. It allows security professionals to define a target and a goal in plain language, after which a team of specialized AI agents plans the attack, executes professional security tools within an isolated sandbox, and generates a final report.

How it works

The system uses a multi-agent architecture where specialized agents handle research, development, and execution. These agents operate within isolated Docker containers to ensure security. The system maintains long-term memory using a PostgreSQL database with pgvector for embeddings and optionally integrates a Neo4j knowledge graph to track findings and relationships. It supports a wide range of LLM providers, including local models via Ollama or vLLM, and provides a web UI and API for monitoring and automation.

Who it’s for

It is designed for security engineers, penetration testers, and red teams who require an autonomous assistant for authorized security testing on their own infrastructure.

Highlights

• Autonomous multi-agent system for planning and executing penetration tests.
• Includes 20+ professional security tools like nmap, Metasploit, and sqlmap.
• Secure, sandboxed execution environment using Docker.
• Supports 10+ LLM providers, including local models via Ollama and vLLM.
• Advanced memory system using pgvector and optional Neo4j knowledge graph for context retention.
• Comprehensive monitoring and observability via Langfuse, Grafana, and Prometheus.