Red Team LLM Lab
github.com
- Free plan
- Windows
- Self-hosted
7.1 Club score Visit github.com
What the maker publishes
- Analysis modules
- Its modules include a prompt analyzer with risk scoring, a jailbreak simulator, a honeypot orchestrator, a backdoor tester, and a master dashboard. source
- Attack testing
- It includes prompt injection testing, jailbreak simulation with 14+ techniques, honeypot agents, backdoor detection, prompt analysis, and RAG poisoning tests. source
- Backdoor detection
- Its backdoor tester checks for hidden triggers, including Unicode triggers. source
- Defense testing
- The framework tests input validation and sanitization, pattern-based injection detection, and output filtering for information leaks. source
- Honeypots
- It provides intentionally vulnerable honeypot agents for studying attack patterns. source
- Integrations
- Support for OpenAI and Anthropic API testing is listed on the project roadmap. source
- Intended users
- The project names security researchers, secure AI application developers, educators, and organizations conducting internal security audits as intended users. source
- Jailbreak simulation
- It tests guardrail effectiveness using 14 or more jailbreak techniques. source
- License
- The project states that it is licensed under the MIT License. source
- Model setup
- The installation instructions download a quantized Phi-3 Mini model, and the project is built with llama-cpp-python. source
- Notable limits
- The roadmap also lists a web UI, Docker support, GPU acceleration, function-calling tests, and CI/CD examples as planned items. source
- Platforms
- The setup instructions show a virtual environment command for Windows and describe running the lab with Python scripts; they do not list supported operating systems as a formal platform matrix. source
- Prompt injection
- It tests for vulnerabilities to prompt manipulation and includes static prompt analysis with risk scoring. source
- Purpose
- Red Team LLM Lab is an open-source framework for testing large language model robustness against security attacks. source
- RAG security
- It can test retrieval-augmented generation poisoning attacks. source
- Reporting
- It saves JSON logs with metadata and provides security posture scoring, category-based vulnerability analysis, and automated recommendations. source
- Requirements
- The stated prerequisites are Python 3.8 or higher, 4GB RAM minimum (8GB recommended), and about 4GB disk space for models. source
- Security guidance
- The project recommends isolated or offline test environments, authorization, careful handling of results, and responsible vulnerability disclosure. source
- Support
- The repository directs users to GitHub Issues for bugs or feature requests and Discussions for conversation. source
Checked against the maker's pages on 7 Oct 2026.