jiwoochris/artex-global

ARTEX – Global Edition

What it is – An open‑source, autonomous penetration‑testing platform that lets large language‑model (LLM) agents run a full security assessment from reconnaissance to data‑exfiltration. The system is built as a single Go binary that embeds a Next.js web UI and stores all findings in PostgreSQL.

Key capabilities

• LLM‑driven multi‑agent workflow – a planner breaks a testing goal into sub‑tasks, multiple worker agents execute real tools (nmap, curl, etc.), and a reporter aggregates results into a graph of assets and vulnerabilities.
• Human‑in‑the‑loop chat – operators can intervene, give hints, or ask for summaries while the run is in progress.
• Multilingual output – user‑facing findings, reports and chat replies can be displayed in English, Korean, Chinese or Spanish. The internal reasoning prompts stay in the original language to preserve model performance.
• Configurable LLM back‑end – works with Anthropic, OpenAI, or any OpenAI‑compatible endpoint; model and token limits are adjustable via the UI or environment variables.
• Built‑in tooling – the Docker image ships with common security utilities (nmap, ripgrep, curl, etc.) and a traffic‑recording MITM proxy that captures all command output for later enrichment.
• Defensive resources – the repo includes a “Defense & Detection Guide” with signatures, hardening checklists, and recommendations for detecting autonomous AI attacks.

Architecture at a glance

Frontend (Next.js) ←embedded→ Go server ←→ PostgreSQL
   │                         │                │
   └─ REST API + SSE        └─ Engine (planner + N workers)
                              │
                              └─ norma SDK agents (goal decomposition, execution)

• The UI visualises tasks, assets, coverage graphs and activity streams.
• The engine schedules planner loops and worker goroutines, persisting all data in two graphs: an asset graph (targets, discovered services) and an exploration graph (steps taken, tool output).
• Supporting subsystems handle traffic proxying, tool‑approval guarding, DNS/HTTP enrichment, and extensible “skills” modules.

Getting started

1. Clone the repo and copy .env.example → .env.
2. Provide a PostgreSQL password and optionally an Anthropic or OpenAI API key.
3. Run docker compose up -d – this starts the artex container and a PostgreSQL instance.
4. Open http://localhost:8787, complete the one‑time admin setup using the printed setup token, and begin creating tasks.

Alternative installs – an install script, pre‑compiled binaries, or a manual go build -tags embedui also work.

Safety notice – The tool is deliberately powerful; it can perform real exploits. The repository includes a prominent misuse warning and is released under AGPL‑3.0 with additional usage‑restriction clauses that require authorized testing only. The authors stress that the software is intended for defenders to understand and detect autonomous AI attacks.

Technology stack

• Backend: Go (net/http, go:embed) with the norma SDK for LLM agents.
• Frontend: Next.js (static build) bundled into the binary.
• Database: PostgreSQL.
• Containerisation: Docker + Docker‑Compose.
• LLM integration: Anthropic or OpenAI‑compatible APIs (configurable via env vars or UI).
• Tools bundled: nmap, curl, ripgrep, vim, npm, plus custom “skills” scripts.

License – AGPL‑3.0 with an additional disclaimer that restricts use to authorized environments and disallows malicious exploitation.

---

This summary is based solely on the repository’s README; no external information has been added.