Prompt Injection Protection Platform

An educational web app for exploring prompt‑injection vulnerabilities, understanding mitigations, and testing LLM systems against injection attacks. Built with React and FastAPI.

Architecture

graph LR
    Browser["React SPA (Vite)"] -->|REST / SSE| API["FastAPI backend"]
    API --> DB["SQLite"]
    API --> LLM["LLM providers (OpenAI, Anthropic, OpenAI‑compatible)"]

Loading

• Frontend: Single‑page React app.
  • Vulnerability catalogue and details.
  • Mitigations library.
  • Testing workbench for running prompts against LLMs.
  • Prompt enhancer tool.
  • Security knowledge assistant (chat).
• Backend: FastAPI service.
  • Stores tests and prompt enhancements in SQLite.
  • Routes all LLM calls through a provider router.
  • Analyses responses for injection risk.

Prerequisites

• Node.js ≥ 18 (npm or pnpm)
• Python ≥ 3.10

Quick Start

Backend Setup

1. Navigate to the backend

cd backend

2. Create a virtual environment

python -m venv .venv

3. Activate the virtual environment

macOS / Linux

source .venv/bin/activate

Windows

.venv\Scripts\activate

4. Install dependencies

pip install -e ".[dev]"

5. Configure environment variables

cp .env.example .env

Add your Azure AI Foundry endpoints and API keys to the .env file.

6. Start the backend server

PYTHONPATH=src uvicorn main:app --reload --port 8080

Frontend Setup (run in a separate terminal)

1. Navigate to the frontend

cd frontend

2. Install dependencies

npm install

3. Start the development server

npm run dev

Frontend will be available at: http://localhost:5173

• Backend REST + docs: http://localhost:8080 (Swagger at /docs)
• Frontend SPA: http://localhost:5173
• Default API base: http://localhost:8080/api (configurable via VITE_API_BASE).

Project structure

.
├── frontend/               → React + Vite + Tailwind + shadcn/ui
│   └── src/app/
│       ├── pages/          # One component per route
│       ├── data/           # Vulnerabilities, mitigations, test presets
│       ├── components/ui/  # shadcn/ui primitives
│       ├── api/            # API client functions
│       ├── assistant/      # Chat widget + streaming client
│       ├── types/          # Shared TypeScript types
│       └── lib/pdf/        # PDF export templates
│
├── backend/                → FastAPI + SQLite
│   └── src/
│       ├── main.py         # FastAPI app entry point
│       ├── container.py    # Dependency injection wiring
│       ├── api/            # Routes + Pydantic schemas
│       ├── app/            # Application services
│       ├── domain/         # Core domain (providers, tests, knowledge, prompt)
│       └── infra/          # Config, DB, providers, runners, analyzers
│
├── FEATURES.md             # Extra UX / feature highlights
└── backend/ARCHITECTURE.md # Backend internals and flows

Key flows

• Testing a prompt

  • User configures a test in the frontend testing page.
  • Frontend calls POST /api/tests and then POST /api/tests/:id/run.
  • Backend resolves model and runner, applies mitigations, calls the LLM, runs analysis, and stores results.
  • Frontend displays test runs and risk scores.

• Prompt enhancement

  • User submits a system prompt and selects mitigations.
  • Frontend calls POST /api/prompt‑enhancements.
  • Backend runs the three‑stage pipeline (improve → prepend mitigations → verify) and stores the final result.
  • Frontend shows original, improved, and enhanced prompts with verdict.

• Security assistant chat

  • User opens the chat and sends a question.
  • Frontend opens an SSE stream to POST /api/chat.
  • Backend searches the knowledge base, augments the message, calls the LLM, and streams tokens back.

For more detail:

• Frontend: see frontend/README.md.
• Backend: see backend/README.md and backend/ARCHITECTURE.md.