Inline AI safety analysis for ChatGPT. Aegis injects a compact panel below each ChatGPT response with risk assessment, confidence vs evidence calibration, and structured analysis (assumptions, bias, failure modes).

1. Backend – see backend/README.md

   cd backend && pip install -r requirements.txt
   # Set OPENAI_API_KEY in .env
   uvicorn app.main:app --reload

2. Frontend – see frontend/README.md

   cd frontend
   cp .env.example .env   # Set API_BASE_URL
   npm install && npm run build

   Load frontend/dist as an unpacked extension in Chrome.

aegis/
├── backend/    # FastAPI API (analyze text, recalibrate)
├── frontend/   # Chrome extension (content script + popup)
└── README.md   # this file

• backend – Python FastAPI service. Uses heuristics + GPT-4o-mini for analysis; sentence-transformers for semantic confidence.
• frontend – Chrome extension that runs on chat.openai.com and chatgpt.com, sends text to the backend, and renders analysis panels.

MIT