j3ssie/osmedeus
Osmedeus is a powerful and modern orchestration engine designed for reconnaissance and penetration testing workflows, written in Go.
Awesome AI Agents › Red-Teaming Platforms
qualifire-dev/rogue
Rogue is an AI agent evaluator and red team platform from Qualifire that stress-tests a deployed agent by driving adversarial and policy-checking conversations against it. It works in two modes. Automatic evaluation tests an agent against declared business policies and expected behaviours: the operator defines scenarios and expected outcomes, watches the probing conversations live, and receives pass or fail reports with the judge reasoning attached, which suits regression testing and policy compliance. Red teaming simulates attacks to surface security weaknesses, covering more than seventy-five vulnerabilities across twelve security categories and twenty attack techniques including encoding tricks, social engineering and injection, with CVSS-based risk scoring. Findings are mapped onto eight compliance frameworks, among them OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF, GDPR and the EU AI Act. The architecture is client-server: a server holds the evaluation and red team logic, a terminal interface written in Go with Bubble Tea provides the interactive client, and a CLI runs non-interactively for CI/CD pipelines. The system under test can be reached through three transports: Google A2A over HTTP, the Model Context Protocol over SSE or streamable HTTP using a send_message tool, or a plain Python entrypoint file exposing a call_agent function that takes a message list and returns a string, which avoids any network protocol. Installation is through uvx as rogue-ai, with subcommands selecting server, TUI or CLI mode, and a bundled example T-shirt store agent for a first run. CLI options cover the evaluated agent URL, the judge LLM, a business context string or file, an input scenarios file and an output report path. Scan presets range from a basic scan of five curated vulnerabilities taking a few minutes to a full scan of seventy-five-plus vulnerabilities and forty-plus attacks running roughly thirty to forty-five minutes, plus custom selections. Requires Python 3.10+ and an OpenAI, Anthropic or Google API key.
https://github.com/qualifire-dev/rogue
Also in Red-Teaming Platforms
Tencent/AI-Infra-Guard
AI-Infra-Guard by Tencent is a comprehensive AI Red Teaming platform offering intelligent vulnerability scanning, MCP server risk detection, and jailbreak evaluation to enhance AI security.
beclab/Olares
Olares is an open-source sovereign cloud operating system that enables users to run local AI models, manage personal data, and self-host various applications securely and privately on their own hardware.
GH05TCREW/pentestagent
Python terminal agent for AI-assisted penetration testing that drives nmap, sqlmap, a browser and MCP servers, with assist, autonomous agent, multi-agent crew and guided interactive modes.
larlarua/AutoCVE
Agent-driven platform that automates CVE discovery end to end, using multiple coordinated agents to audit source code, verify vulnerabilities, and generate submission-ready reports.
SanMuzZzZz/LuaN1aoAgent
LuaN1aoAgent v2 is an autonomous security research agent that splits work between Planner, Executor and Observer roles and records every conclusion as evidence-backed nodes in a causal reasoning graph.
ASCIT31/Dark-Moon
Open-source autonomous penetration testing platform where specialised agents plan and run real offensive operations through a controlled MCP execution layer, with local tokenization that hides sensitive values from the model.
hackerai-tech/hackerai
An open-source penetration testing assistant with a chat interface and an agent mode that executes security tooling inside isolated cloud sandboxes on a durable task runtime.