#1313th of 40 in Security

redamon

Self-hosted AI pentest agent with Neo4j attack-surface graph and auto-fix PRs

Releases and problems feed (RSS)

Stars
3.4k
License
MIT
Last commit
Nov 2026
Last release
Nov 2026
Language
Python
Stars, last 29 weeks+1,654
29 Apr 2026: 1,791 stars6 May 2026: 1,817 stars13 May 2026: 1,844 stars20 May 2026: 1,874 stars27 May 2026: 1,899 stars3 Jun 2026: 1,923 stars10 Jun 2026: 1,947 stars17 Jun 2026: 1,981 stars24 Jun 2026: 2,007 stars1 Jul 2026: 2,103 stars8 Jul 2026: 2,143 stars15 Jul 2026: 2,169 stars22 Jul 2026: 2,193 stars29 Jul 2026: 2,223 stars5 Aug 2026: 2,251 stars12 Aug 2026: 2,284 stars19 Aug 2026: 2,311 stars26 Aug 2026: 2,331 stars2 Sept 2026: 2,355 stars9 Sept 2026: 2,401 stars16 Sept 2026: 2,434 stars23 Sept 2026: 2,552 stars30 Sept 2026: 2,851 stars7 Oct 2026: 2,946 stars14 Oct 2026: 3,053 stars21 Oct 2026: 3,170 stars28 Oct 2026: 3,226 stars4 Nov 2026: 3,292 stars11 Nov 2026: 3,374 stars18 Nov 2026: 3,445 stars

1,791 on 29 Apr 20263,445 on 18 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

RedAmon runs a parallel recon pipeline, stores the attack surface in a Neo4j graph, and hands it to an LLM agent that attempts exploitation behind human approval gates. A triage agent and a CodeFix agent then rank findings and open GitHub pull requests with fixes. It has multi-user RBAC and audit trails, and supports OpenAI, Anthropic, Bedrock, OpenRouter and OpenAI-compatible endpoints such as Ollama.

Who it is for: Security teams running authorized pentests on infrastructure they own

Strengths

  • Claims 101/104 on the XBOW web benchmark black-box, with raw agent sessions published
  • Recon results merge into a Neo4j graph the agent can query in natural language
  • Closes the loop: triage agent plus CodeFix agent that opens GitHub PRs
  • Multi-user RBAC, Rules of Engagement enforcement, approval gates and audit trails

Weaknesses

  • Heavy footprint: 9 containers, 7 GB RAM and 80 GB disk minimum without OpenVAS
  • OpenVAS option needs about 10 GB RAM and a ~30 minute first feed sync
  • Benchmark and coverage figures are self-reported; the audit lists 210 classes still in backlog
  • Autonomous exploitation tooling carries legal risk if pointed at targets you lack permission to test

What it needs

  • RAM ≥ 7 GB
  • no GPU
  • Compose
  • Needs Docker Engine 28+, Docker Compose v2, Neo4j, PostgreSQL, Kali sandbox, Metasploit, OpenVAS (optional)
  • Models: OpenAI, Anthropic, OpenRouter, AWS Bedrock, OpenAI-compatible (Ollama, vLLM, Groq)
  • port 3000

What changed

  1. v7.8.0 ↗ (opens in a new tab)Security fix

    GHSA-J3R5-X2GG-2HGV, GHSA-3CJG-8CP8-J269, GHSA-WGXP-VQ94-73H8, GHSA-W454-6HWP-FM58, GHSA-VHJW-GFMX-J3FX, GHSA-8WHG-RVCR-R88R. “Six advisories are fixed:”

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Security

See all 40
Also in Security
RankProjectScore
1shannonAutonomous pentester that exploits web apps and APIs using source analysis52.1k stars+3,054 (+6.2%) in 30 daysAGPL-3.092 out of 100
2ifixaiPython CLI that audits AI agents for policy, access and reliability failures46.7k stars+17,519 (+60%) in 30 daysApache-2.087 out of 100
3StrixAutonomous multi-agent pentesting CLI that validates vulnerabilities with working PoCs76.6k stars+6,499 (+9.3%) in 30 daysApache-2.085 out of 100
#4ai-infra-guardRed-teaming platform for AI infra CVEs, MCP servers, agent skills and jailbreaks7.6k starsApache-2.084 out of 100
#5decepticonAutonomous red-team agents that run multi-stage attack chains from an approved plan Live demo ↗ (opens in a new tab)6.6k starsApache-2.084 out of 100