In short
RedAmon is an open-source, MIT-licensed AI penetration testing framework. An AI agent maps a target's attack surface into a Neo4j graph and runs reconnaissance and exploitation from a Kali Linux sandbox behind human approval gates. It can then open pull requests that fix the flaws it finds. It is dual-use, like most AI offense tools.
Key takeaways
- What it is
- A self-hosted AI pentest framework on GitHub, MIT-licensed, built on LangGraph, Neo4j, MCP servers and Docker.
- How it works
- A six-stage pipeline: recon, exploitation, post-exploitation, AI triage, a CodeFix agent and a GitHub pull request.
- Why it matters
- It puts a full attack chain, with Metasploit, Nmap and Nuclei, behind one agent that anyone can approve step by step.
- What Odin Research saw
- 17 RedAmon endpoints in Monograph 20, each on its own host. Three controllers are confirmed; ten candidate dashboards use port 3000.
- What to do first
- Shrink and patch your external surface, and hunt for unregistered RedAmon containers on your own hosts.
What is RedAmon?
RedAmon is an open-source, self-hosted AI penetration testing framework published on GitHub under the MIT license. It maps a target's attack surface into a graph. An AI agent then runs reconnaissance and exploitation from a Kali Linux sandbox behind human approval gates, and can open pull requests that fix what it finds.
The project is maintained by its creator, Samuele Giampieri, with Ritesh Gohil listed as a co-maintainer. It is one of a new group of AI agent tools that join a large language model to real security tools. The README states that RedAmon is for authorized security testing, education and research only.
Like other AI offense frameworks, RedAmon is dual-use. Most installs are likely to belong to testers and researchers. The same automation helps anyone who wants to probe many web applications and APIs quickly.
How RedAmon works
The README describes a pipeline with six stages:
- Reconnaissance of the target's hosts, services and web content.
- Exploitation from a Kali sandbox. Dangerous tools need manual confirmation before they run.
- Post-exploitation on access that the agent gains.
- AI triage of the findings.
- A CodeFix agent that writes a fix.
- A GitHub pull request with that fix.
Under the hood, RedAmon uses a LangGraph agent, a Neo4j graph database for the attack surface, PostgreSQL, a Next.js web app and Docker containers. Tools are exposed to the agent through Model Context Protocol (MCP) servers. The README counts more than 100 security tools, more than 12,000 Nuclei templates and 22 agent tools over MCP, with GVM/OpenVAS as an optional scanner.
Separate MCP services cover network recon, Nuclei, Metasploit, Nmap and Playwright, each on its own local port. That layout matters for defenders: one RedAmon host can expose several services, not just a web page. For how MCP tool access should be governed, see Securing MCP Servers and Tool Access.
RedAmon requirements, versions and popularity
- Runtime. Docker Engine 28.0.0 or later and Docker Compose v2 or later.
- Hardware. At least 7 GB of RAM without OpenVAS and 10 GB with it, with 16 GB recommended. Disk needs are 80 GB without OpenVAS and 110 GB with it.
- Models. OpenAI, Anthropic, OpenRouter, AWS Bedrock or any OpenAI-compatible endpoint, such as a local Ollama or vLLM server.
- Web app. The README serves the web app on port
3000by default. - Version and reach. The repository showed version 7.9.0, about 3,500 stars and 737 forks on 21 November 2026.
Is it legal? The software is MIT-licensed and legal to download. The project itself warns never to scan, probe or attack a system without ownership or explicit written permission.
Why RedAmon matters to defenders
RedAmon packages a full attack chain, from surface mapping to exploitation, behind one AI agent. That lowers the skill needed to run a structured test. In the wrong hands, it lowers the skill needed to run a structured attack.
Its built-in Metasploit, Nmap and Nuclei services mean its traffic looks like a busy, well-organized scanner. Expect rapid, systematic requests against web applications and APIs, including the public-facing application exploits that agents reach for first. The approval gates and fix-oriented design help honest users, but nothing stops a hostile operator from approving every step.
What Odin Research observed
Monograph 20 studied seven specialized AI vulnerability agents across 285 endpoints. RedAmon sits in the container validation stage, next to Strix and XingRin, which run web and API checks from isolated containers.
- 17 endpoints, one per host. Every RedAmon endpoint sat on its own host.
- Dashboards on 3000. Ten were candidate operator dashboards on port
3000, the project's default. Agent listeners sat on ports80and443. - Three confirmed controllers. Three controllers on port
80were confirmed, on Oracle Cloud in the United States, Contabo in Germany and Leaseweb in Singapore. Monograph 20 lists them for blocking. The other 14 are candidates. - Two families on one server. One VPS in South Korea ran a RedAmon console on
3000next to a Strix console on8080. It was one of only three hosts in the study that ran two agent families. Its address stays withheld because neither service is confirmed.
For context, the same study counted 40 Strix endpoints and 11 XingRin endpoints. The three co-hosted servers show some operators combine tools. They do not show a single operator running the whole chain.
How to detect RedAmon
- Unapproved containers. Watch container hosts for RedAmon services, a Neo4j graph and a Kali sandbox that nobody registered.
- Port layout. A web app on
3000plus several MCP services on adjacent local ports is a stronger sign than any single page title. - Agent-style scanning. Bursts of Nmap, Nuclei and Playwright activity from one source, with steady pacing and systematic parameter tests, suggest an automated agent.
- Not port 3000 alone. Port 3000 is the default for many web frameworks, so treat it as a lead, not a verdict.
Monograph 20 publishes a Suricata signature for RedAmon agent beacons and a YARA rule for Strix, RedAmon and XingRin orchestration files. It adds a Sigma rule for unapproved recon or swarm workers and a Microsoft Sentinel KQL hunt. All are in the Defense Kit.
How to defend against AI-driven testing tools
- Shrink the surface. Agents start with reconnaissance. Use external attack surface management to find exposed apps and APIs first.
- Patch fast. Prioritize the KEV catalog and internet-facing apps.
- Rate-limit and watch APIs. Web application firewalls and API gateways can slow and flag rapid, systematic probing.
- Govern your own use. If your team runs RedAmon, keep it on registered test hosts with written scope and logged approvals. See Red Teaming AI Agents and the AI Agent Governance Policy Template.
RedAmon compared with other AI pentest agents
| Tool | Role in Monograph 20 | Endpoints seen | Notes |
|---|---|---|---|
| RedAmon | Container validation of web apps and APIs | 17 (3 confirmed) | Dashboards on 3000; full recon-to-fix pipeline in the README |
| Strix | Container validation with one Docker sandbox per task | 40 | Highest port-to-host density in the study |
| XingRin | Distributed web scanner nodes | 11 | Nodes in China and Hong Kong share a self-signed certificate |
| HexStrike AI | MCP tool bridge for AI clients | Covered in other Odin studies | See our HexStrike AI page |
Frequently asked questions about RedAmon
What is RedAmon?
RedAmon is an open-source, MIT-licensed AI penetration testing framework. An AI agent maps the attack surface into a graph, runs reconnaissance and exploitation from a Kali sandbox behind approval gates, and can open pull requests with fixes.
Is RedAmon malware?
No. RedAmon is a dual-use security tool built for authorized testing. Attackers could run it against systems they have no permission to test, which is why defenders track its servers.
Is RedAmon legal to use?
Yes, on systems you own or are authorized to test. The project's README says never to scan, probe or attack systems without ownership or explicit written permission.
Who made RedAmon?
The README names Samuele Giampieri as creator and maintainer, with Ritesh Gohil as a maintainer. The code is on GitHub under the MIT license.
Which AI models does RedAmon support?
The README lists OpenAI, Anthropic, OpenRouter, AWS Bedrock and any OpenAI-compatible endpoint, which includes local models served through Ollama or vLLM.
What does RedAmon need to run?
Docker Engine 28.0.0 or later and Docker Compose v2 or later, at least 7 GB of RAM without OpenVAS and 10 GB with it, and 80 to 110 GB of disk.
How do I detect RedAmon?
Look for unregistered RedAmon containers, a web app on port 3000 with several MCP services beside it, and agent-style scanning bursts. Odin Research publishes Suricata, YARA, Sigma and KQL rules in its Defense Kit.
RedAmon vs Strix: what is the difference?
Both validate web apps and APIs from containers. RedAmon adds a graph-based recon-to-fix pipeline. Strix gives each task its own Docker sandbox and port, which produced the densest port layouts in Odin's data.
Sources
- samugit83, RedAmon repository and README, GitHub. github.com/samugit83/redamon (checked 21 November 2026)
Related tools and terms
What is CyberStrikeAI?
CyberStrikeAI is a public AI-driven offensive security platform. It spreads its operator interface, API and agent listener across several ports on one host and exposes tools through MCP.
AI offensive toolWhat is PentAGI?
PentAGI is an open-source autonomous penetration testing system that coordinates AI agents and security tools in containers.
AI offensive toolWhat is HexStrike AI?
HexStrikeAI is a public project that exposes many security tools to AI agents through the Model Context Protocol (MCP).
AI security toolWhat is DeepAudit?
DeepAudit is an open-source AI code auditing tool. Application security teams use it to find flaws in source code.
AI security toolWhat is AutoCVE?
AutoCVE is an AI-assisted tool that turns new vulnerability advisories into HTTP checks.
AI security toolWhat is Cairn?
Cairn is a reconnaissance tool that keeps an asset graph built from DNS zones, TLS certificates and HTTP banners.
Odin Research
Get the rules, not just the definition
Every Odin Research report ships with verified indicators and detection rules you can load today.