RedGrid

LLM-based agentic framework for autonomous black box penetration testing

Agentic
Red Team

A research environment for evaluating autonomous black box penetration testing agents. Run different agents against standardized security benchmarks, watch them reason and execute live, and compare agents and models side by side.

Watch Demo
pentest agents
2+
benchmarks
2+
sandboxed
100%

How RedGrid Works

redgrid · mission consolelive
target 10.20.4.17one-day mode
01/07
Environment Layerobserved facts
  • host: 10.20.4.17
  • ports: 80, 443, 8080
  • stack: nginx + PHP 8.1
  • waf: ModSecurity (detected)
  • creds: admin:••• (harvested)

Recon

Scope & observe

The orchestrator scopes the mission, then runs an autonomous recon pass. Every confirmed fact is written into the Environment Layer — never inferred.

Agent stream
  • ›scope: target + rules of engagement
  • ›auto-recon: nmap · WhatWeb · ObserverWard
  • ›→ 4 facts written to Environment Layer
Vulnerability Graphfrontier
  • sqli_login0.91
    SYNTHESIZING
  • auth_bypass0.78
    SYNTHESIZING
  • xss_search0.54
    SYNTHESIZING
Skill Library
0 skills
awaiting promotion
+
Elapsed0:12
LLM calls6
Est. cost$0.04
Confirmed0
Built withby
KAI Logo
KAI

© 2026 RedGrid, all rights reserved.