damn-vulnerable-llm-agent
11commits
4contributors
2languages
Tech stack & purpose
Damn Vulnerable LLM Agent is an educational chatbot built with Langchain that implements a Large Language Model ReAct agent designed to teach security researchers and developers about prompt injection attacks. The project focuses specifically on Thought/Action/Observation injection techniques and was adapted from a challenge created by WithSecure Labs for the BSides London 2023 CTF competition. It supports multiple LLM backends including OpenAI, HuggingFace models, and locally-run models via Ollama, and can be deployed using either Pipenv or Docker with a Streamlit interface.
Community & reference links
Languages
Python
95.3%
Dockerfile
4.7%
Contributors
- Donato Capitella5 commits
- Deepthi3 commits
- kyuz02 commits
- Nick Jones1 commits