Hinln/ARTEX

What it solves

ARTEX is an autonomous AI-driven penetration testing system designed to automate the complex process of security vulnerability discovery and exploitation. It replaces manual, repetitive scanning and exploration with a multi-agent system that can independently plan and execute attack chains, track assets, and verify vulnerabilities.

How it works

The system employs a dual-graph architecture to separate the "what" from the "how":

• Asset Graph: A global, shared library of truth containing the hierarchy of domains, IPs, services, and endpoints.
• Exploration Graph: A task-specific map of the "thinking process," linking goals, intents, facts, and findings via a lineage chain (e.g., a goal spawns an intent, which yields a fact, which proves a vulnerability).

It operates via an event-driven loop involving a Planner (which generates intents based on the current state of the exploration graph) and multiple Workers (which execute those intents using real-world tools like Kali, Bash, or HTTP). To maintain stability in multi-step attack chains, the Planner uses a persistent, shared todolist to track dependencies across multiple sessions.

Who it’s for

Security researchers, penetration testers, and developers interested in autonomous AI agents for cybersecurity, specifically for those needing a system that can automate the discovery of complex, multi-step vulnerability chains.

Highlights

• Autonomous Attack Chains: Uses a shared todolist to execute sequential dependencies (e.g., Injection $\rightarrow$ Credentials $\rightarrow$ Lateral Movement $\rightarrow$ Privilege Escalation).
• Cross-Worker Information Exchange: Workers can search through the execution traces of other workers to reuse observations without duplicating effort.
• Human-in-the-Loop: Includes a chat interface for human interaction and an intercept/approval gate for tool execution.
• Visualized Exploration: Provides a dashboard with asset coverage maps and exploration link graphs.
• Integration: Supports asset synchronization from ScopeSentry.