garagehq/nightcrawler

What it solves

Nightcrawler automates the process of penetration testing—simulating cyberattacks to find security vulnerabilities—directly from a smartphone. It eliminates the need for cloud connectivity or constant human intervention, allowing a device to be deployed as a "drop box" on a network to autonomously discover and test targets.

How it works

The system runs a small AI model (LFM2.5-1.2B) locally on a phone's GPU using OpenCL. The agent follows a continuous loop: it selects a target, generates a command based on current network observations, validates that command through a safety "scope proxy" to ensure it stays within authorized boundaries, and executes the tool via a Kali MCP server. It progresses through phases of reconnaissance, enumeration, and exploitation, using a database of over 24,000 CVEs and 27 pre-defined attack playbooks to identify weaknesses.

Who it’s for

It is designed for professional penetration testers and security researchers who need an autonomous, stealthy, and portable tool for authorized network security assessments.

Highlights

• 100% Local Inference: Runs entirely on the device GPU, requiring no internet or cloud APIs.
• Stealth-First Approach: Uses slow scan rates, host rotation, and cover traffic to avoid detection by network monitoring systems.
• Scope Enforcement: A two-layer defense system prevents the agent from attacking hosts or ports outside the authorized scope.
• Autonomous WiFi Breach: Can autonomously crack WPA2 networks using an external USB WiFi adapter.
• Integrated C2 Dashboard: A web-based interface for real-time monitoring, network mapping, and manual steering.