Skip to main content
Back

ARTEX Global Brings Autonomous AI Hacking to Five Languages

ARTEX Global ports a Chinese autonomous penetration-testing agent framework into English, Korean, Spanish and Japanese with built-in defender tooling.

Repo, Full Access
  • ARTEX Global is an English/Korean/Spanish/Japanese fork of the Chinese Autumn-27/ARTEX autonomous pen-testing framework.
  • LLM multi-agents autonomously plan, run real tools, and chain exploits into a dual-graph of assets and exploration state.
  • Only user-facing output is translated; agent reasoning prompts stay in the original language to preserve benchmarked behavior.
  • Ships as a single Go binary with embedded Next.js UI over PostgreSQL, deployable via Docker Compose on port 8787.
  • Includes Sigma + Suricata detection rules, ATT&CK Navigator layer, MISP indicator feed, and host-triage script.
  • AGPL-3.0 licensed; authors warn of ongoing Korean investigations into misuse of upstream ARTEX in real breaches.

ARTEX Global is an international fork of the original ARTEX, a Chinese open-source framework that coordinates multiple large language model agents to plan and execute penetration tests with minimal human input. The fork localizes human-facing output, supports five languages and adds detection resources for defenders. At the time of writing, its repository had attracted about 1,700 stars and 2,000 forks.

Two graphs give agents memory

ARTEX assigns different roles to LLM agents during a penetration test. A planner breaks the objective into bounded tasks called intents. Workers accept one intent at a time, invoke tools through Kali, Bash or HTTP interfaces, record the results and stop. A reporting agent turns the accumulated findings into a human-readable assessment.

The application is distributed as a Go binary with a Next.js frontend bundled through go:embed. PostgreSQL stores task state, discovered assets, findings and execution history.

Component Purpose
Asset graph Maintains a shared inventory of domains, subdomains, IP addresses, services, applications and endpoints discovered across tasks.
Exploration graph Tracks each task’s goals, intents, facts, findings and hints. Anchors connect those records to entries in the asset graph.
Planner Reads the current graph state and creates an intent when the evidence suggests a new line of investigation.
Workers Run tools for a single intent, write observations back to the graphs and then terminate.