Robots Atlas>ROBOTS ATLAS
ARTEX
Active

ARTEX

ARTEX is an open-source, autonomous agentic framework for penetration testing (AI autonomous penetration testing). It is model-agnostic — it can use any large language model (LLM) as the agent's "brain" to plan and automate offensive security testing. The project is self-hosted and released under the AGPL-3.0 license.

Its architecture is built on a division of labor: a planner splits a goal into intents, and workers execute them using real tools. The platform includes a recording MITM proxy, approval gates, scope rules and budgets that constrain the agent. The project's tagline is "Autonomous penetration testing that keeps its receipts."

ARTEX originated as a Chinese open-source project. The original upstream repository (Autumn-27/ARTEX) is no longer public. An actively developed English continuation lives at BaaRai/ARTEX (~5.3k stars, AGPL-3.0, Go backend, Next.js UI). A localized fork, jiwoochris/artex-global, exists under the same license.

The tool is dual-use (offensive security). It supports authorized penetration testing performed with the system owner's consent, but according to Microsoft analyses it has also been abused by cybercriminals — including to find and exploit exposed secrets (OAuth tokens) in repositories and in campaigns targeting the financial sector (an actor tracked as Storm-3358).

Self-hosted (lokalnie)CLIFree
Updated 17 Nov 2026

Overview

System type
Input
Text
Output
ReportsRisk alertsCodeText

Access & Pricing

Access channels
Self-hosted (lokalnie)
Access by self-hosting on your own hardware — no vendor cloud, fully local.
CLI
Access via command-line interface.
Subscription model
Free

Open-source software (AGPL-3.0), self-hosted — no charge for the framework itself. The real cost comes only from the chosen LLM (API or your own inference infrastructure) and compute resources.

Capabilities

7 capabilities
Reasoning & Planning
Agency & Action
Task Specialization

Integrations

System tools

Sources

3 sources