Nightcrawler: A Local AI Pentesting Agent Running on a Smartphone

Autonomous Mobile Pentesting

Nightcrawler is a new autonomous penetration testing agent designed to run entirely on a smartphone. It allows users to drop a phone onto a network, and the agent then independently discovers hosts, maps services, identifies vulnerabilities, and produces a pentest report. This process occurs without requiring any cloud connectivity.

Local AI-Powered Operations

The system utilizes a small AI model, LFM2.5-1.2B-Instruct-Heretic (1.2 billion parameters), which operates locally on the phone's GPU. This AI model guides the agent's actions, determining which hosts to probe, which tools to employ, and what information to seek. The local execution eliminates the need for an internet connection or cloud API access during the assessment.

Pentesting Capabilities

Nightcrawler includes several core functionalities: optional WiFi breaching for WPA2 networks using an external USB adapter, stealthy reconnaissance to discover network devices, enumeration of services like web servers and SSH, exploitation testing for known vulnerabilities and default credentials, and structured report generation with findings and remediation advice. The agent's operational method involves rotating across hosts and performing small actions incrementally, which makes it less detectable than conventional vulnerability scanners that conduct simultaneous, broad scans.

System Architecture

The agent's architecture on a OnePlus 8 phone running NetHunter involves the LFM2.5 1.2B model on the GPU, an Agent Loop (main.py) for decision-making, and a Scope Proxy that validates commands to prevent out-of-scope actions. A Kali MCP Server executes actual commands like nmap and curl, while a Web Dashboard allows monitoring and steering, and an SQLite database stores discovered hosts, vulnerabilities, credentials, and commands.