openhackai/OpenHack

OpenHack – AI‑powered, open‑source security scanner

What it is – OpenHack is a command‑line tool that uses a pipeline of LLM‑driven agents to automatically look for security bugs in a codebase. It mimics the “Claude Code Security / Codex Security” services but is fully open source and can run with any open‑source or commercial LLM you connect.

How it works – The scan proceeds through four stages:

1. Recon – the agents read every file, build a structural model of the project and ingest any extra context you provide.
2. Hunters – a set of specialized agents (e.g., input‑validation, SQL‑injection, access‑control hunters) scan the model for likely vulnerable patterns. A second “feature‑hunter” pass dives deeper into risky code areas.
3. Validation – another agent reviews each candidate, checks its impact and decides whether it is a true finding.
4. Verification – (beta) the tool can spin up the application in Docker and run a headless browser to actually exploit the finding. Successful exploits are marked with a ✓.

Key UI – OpenHack ships with an interactive terminal UI (TUI). While a scan runs you see a Trace pane that shows the recon‑→‑hunter‑→‑validator flow in real time, and a Findings pane that lists confirmed issues with severity, CVSS score, file location, code snippet and a suggested fix. All of this can be navigated with keyboard shortcuts or mouse.

CLI / CI mode – For automation you can run the same pipeline head‑less:

openhack --scan /path/to/repo   # prints progress, writes JSON report

Other flags let you resume a paused scan, list past sessions, classify frameworks, or run a single “hack” task.

Model flexibility – After the initial login you can connect any LLM provider (OpenAI, Azure OpenAI, Anthropic, DeepSeek, Kimi, etc.) via the /connect command. The UI shows a searchable picker that pulls model metadata from Models.dev and also ships an offline catalog. You can switch models on the fly with /models.

Sandbox verification – If Docker is available, /verify sandbox launches the app inside a container and sends real HTTP requests to test each finding. /verify browser adds a headless browser layer to catch client‑side bugs like XSS or CSRF.

Installation – The package is on PyPI and can be installed with any Python installer:

pipx install openhack      # isolated, recommended
# or
uv tool install openhack   # with uv
# or
pip install openhack

Privacy – Source code is processed locally; only the prompts that are sent to the selected LLM provider leave your machine. If you use the hosted OpenHack inference service, prompts go through their API; otherwise they go directly to the provider you configured.

License – MIT, fully open source. Contributions are welcomed via GitHub.

---

Bottom line – OpenHack is a genuine, AI‑driven security‑analysis tool that lets developers run multi‑agent vulnerability scans on their own code, optionally verify exploits in a Docker sandbox, and stay in control of which LLM powers the analysis.