Autonomous pentesting gets a multi-agent system: self-hosted agents take on complex attack tasks

Complex penetration testing tasks can be carried out by a fully autonomous multi-agent system rather than by a person driving one tool or one model step by step. Earlier penetration testing automation relied mostly on manual orchestration or a single model, and lacked an open-source scheme that organizes multiple autonomous agents into a complete system. The project describes itself as a fully autonomous AI agent system capable of performing complex penetration testing tasks, and the repository shows about 29.3k stars and 4.4k forks, indicating substantial developer attention.

The system is made of multiple agents that collaborate through a graph structure, with a Go backend, a React frontend and a GraphQL interface layer. The multi-agent label corresponds to graph-structured multi-agent collaboration, not a simple wrapper around one model with tool calls. The repository contains backend, frontend, observability and examples directories, and provides docker-compose.yml along with separate Compose files for Graphiti, Langfuse, observability and end-to-end testing.

The project positions itself as a fully autonomous agent system for performing complex penetration testing tasks, aimed at developers and security teams that need automated security testing, deployed through self-hosting, with Docker Compose and companion configurations for observability, Graphiti, Langfuse and end-to-end testing. The material does not state specific attack task types, target environments or usage limits.

The material comes only from a repository summary and gives no success rate, coverage or comparison with manual testing for penetration testing tasks; star and fork counts reflect attention, not capability evidence.