A Korean localization of an autonomous AI pentest tool — with a defense-first warning label

What it does

ARTEX is a Go single binary (with an embedded Next.js frontend, PostgreSQL for storage) where multiple LLM-driven agents break down objectives, run real tools, and accumulate discovered assets and vulnerabilities into a graph — an autonomous pentest pipeline with minimal human intervention. This fork localizes it for Korean teams: detection results, summaries, reports, and chat responses are forced into Korean, while commands, payloads, and logs stay in their original form.

The interesting bit

The localization policy is deliberately surgical: internal reasoning prompts are left untranslated to preserve the behavior benchmarked in the original Chinese version — only user-facing output is translated. The repo also leans hard into the defender’s side, shipping detection signatures and a hardening checklist alongside the attack tool, motivated by reports that the original ARTEX was allegedly used in an attack on Korean financial institutions.

Key highlights

• Go backend + embedded Next.js UI in a single binary; PostgreSQL stores findings and activity
• Human-in-the-loop chat lets you nudge agents mid-run, with attack-chain summaries in Korean
• Works with OpenAI-compatible endpoints, so domestic/open models can replace frontier providers
• Includes a defense/detection guide (Sigma-style signatures, hardening checklist) for blue teams
• Original Chinese docs preserved in README.zh.md for easier upstream diffing

Caveats

• The Docker Compose quickstart is currently broken: the upstream autumn27/artex image is unavailable on Docker Hub, so you must compile from source
• Finding titles are model-generated and may mix in English; only the surrounding UI is fully Korean
• AGPL-3.0, with explicit Korean legal warnings (정보통신망법, 개인정보보호법) — authorized targets only

Verdict

Worth a look for Korean-speaking security teams who want to study how autonomous AI attacks operate and how to detect them. If you wanted a turnkey Docker deployment or an English experience, wait — or go upstream.