vxcontrol/pentagi

What it claims

PentAGI describes itself as a "fully autonomous AI Agents system capable of performing complex penetration testing tasks," with a project site at pentagi.com. The repository has about 31.5k stars, 5.0k forks, 169 watchers and 952 commits on main. At capture time (October 2026) it had 21 open issues and 25 open pull requests.

How to use it

The fetched page did not render the README, so no install or run commands are available here. The file list shows what a deployment involves. A base docker-compose.yml sits next to separate Compose files for Langfuse (docker-compose-langfuse.yml), Graphiti (docker-compose-graphiti.yml), observability (docker-compose-observability.yml) and an end-to-end test setup (docker-compose.e2e.yml). The optional services appear to be split out from the core stack. Configuration starts from .env.example. Code is split between backend and frontend directories, and there is an examples directory. A top-level Dockerfile is included.

The repository also contains a CLAUDE.md file, the convention for instructions to Claude Code. This suggests the maintainers use AI coding agents on the codebase.

Limits

The repository is tagged MIT, but it also ships an EULA.md, a NOTICE file and a licenses directory. The page does not show what the EULA covers. Anyone planning to adopt or redistribute PentAGI should read these files before relying on the MIT label alone.

The capture contains only GitHub navigation, the About box, topics and the file tree. It does not include the README's architecture description, supported models, hardware needs, setup steps or any stated safety controls on the autonomous offensive tooling. The topics list offensive-security, so any use must be restricted to systems the operator is authorized to test.