Multi-agent architecture
Recon, exploit, post-exploit and reporting subagents coordinate through an orchestrator, each specialised for its phase of the engagement.
← Store · // product
OpenHack.
An AI-powered security-assessment agent for authorized penetration testing. It plans and runs recon, exploitation, post-exploitation and reporting through coordinated specialist agents — every action gated by a hash-signed authorization harness.
from $499support & customisation
⚠ Authorized use only — liability noticeThis is offensive-security software, published for lawful, authorized testing, research and education. You are solely responsible for how you use it, and you must hold explicit written authorization for any system you target. Provided as-is, without warranty; Cyber Sec accepts no liability for misuse.
// what it is
An AI pentest agent that stays inside the scope you signed for
OpenHack is a multi-agent AI assistant that plans and executes authorized security assessments — recon, exploitation, post-exploitation and reporting. It is a hardened fork based on the OpenCode coding agent, rebuilt around a safety and rules-of-engagement layer so the automation can never operate outside a scope you authorized in writing.
Rather than one monolithic model, OpenHack dispatches work to specialist subagents — recon, exploit, post-exploit, C3 and reporting — and routes each task to the one best suited to it. A council review can fan a single finding out to independent reviewer agents for a second opinion before it ever reaches the client report.
Every tool call passes a single runtime enforcement hook that checks the engagement scope and a hash-signed Rules-of-Engagement record before anything is allowed to execute. Findings are written with integrity hashes and uncertainty flags, and secrets are scrubbed automatically on the way out.
// capabilities
What it does
The capabilities below are drawn straight from the project — the safety harness is the point, not an afterthought.
Signed Rules of Engagement
Hash-signed authorization fields with tamper detection; a single runtime hook blocks any tool call that falls outside the signed scope.
MCP tool integration
Bundles opt-in Model Context Protocol servers — recon, findings, reporting and more — all disabled by default until you explicitly enable them.
Integrity-checked findings
Findings carry HMAC integrity and uncertainty flags; secrets are scrubbed automatically before anything is stored, exported or shared.
MoE routing
Dispatches each task to the best-matching specialist and remembers what worked, so routing sharpens as the engagement runs.
Council review
Fan a finding out to independent reviewer agents for an adversarial second opinion before it lands in the report.
// under the hood
Specifications
- Language
- TypeScript on the Bun runtime (Python for MCP servers)
- Version
- 0.2.0
- License
- MIT — a fork based on OpenCode by Anomaly
- Interfaces
- CLI, terminal UI (TUI) and desktop app
- Requirements
- Bun 1.3+, Node.js 22+, Python 3.8+ for the MCP servers
- Intended use
- Authorized security assessments only
// what you pay for
Support, not a paywall
OpenHack’s source is open (MIT). What you pay for is the person who built it.
- Installation & configuration help — stood up in your environment, tuned to your setup and rules of engagement, not left as a README you fight alone.
- Bug fixes & improvements, prioritised for you — issues you hit move to the front of the queue.
- Customisation — new modules, integrations, or behaviour built for your targets and workflow. You keep the source.
- A direct line to the engineer who wrote it — no support tier relaying questions to a team you never meet.
// the fine print
Terms & liability
Authorized, lawful use only. This software is published for legitimate security testing, research and education. You alone are responsible for how you use it, and you must have explicit written authorization for any system, network or application you test.
No liability. To the maximum extent permitted by law, Cyber Sec and Stephan Botes accept no liability for any direct, indirect, incidental or consequential damage, data loss, service disruption or legal consequence arising from the use — or misuse — of this tool. Everything is provided as-is and without warranty of any kind.
What the fee buys. The price shown is a starting point and covers the engineer’s time — setup, fixes, improvements and customisation. Where the source is public, you are paying for the work and the time, not for access.
Questions about scope, licensing, or authorization? Get in touch first.