RAVEN

Red-teaming AI Validation & Exploitation Network

RAVEN is an automated AI red teamer that tests attacks for LLM prompt injection, information leakage, and content policy bypassing.

Setup

You will need to run both the backend Python FastAPI server and the Vite front-end.

API Key

RAVEN uses an LLM to generate adversarial prompts. To set up this LLM, get an OpenAI API key and create a .env file in the backend directory with the following format:

# backend/.env
OPENAI_CHAT_ENDPOINT=https://your.openai.chat.endpoint
OPENAI_CHAT_MODEL=gpt-4o
OPENAI_CHAT_KEY=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Backend

cd backend
pip install -r requirements.txt

Frontend

cd frontend
npm install --force

Usage

The backend and frontend each need to be running in their own terminal, at the same time.

Backend

cd backend
fastapi dev main.py

Frontend

cd frontend
npm run dev

After starting the servers, open the browser to the localhost address indicated by the frontend Vite terminal. You may need to wait a moment for the backend to finish importing PyRIT.