KIT AI is an offline-first emergency health assistant designed for situations where internet access is unreliable or unavailable—such as airplanes, hiking trails, or low-connectivity regions.
It provides objective, widely accepted first-aid information using a locally running AI model, without requiring network access at the time of use.

Disclaimer: KIT AI is not a doctor, does not diagnose, and does not replace professional medical care. It shares general first-aid guidance only. In emergencies, seek professional help whenever possible.

Most AI health tools require constant internet access. In real emergencies, that's often impossible.

KIT AI is built to:

• Work fully offline
• Run directly on the user's device
• Share clear, factual first-aid steps
• Avoid hallucinations, diagnoses, or personalized medical advice
• Update medical content safely when internet is available

• Uses a local LLM (WebLLM) running entirely on-device via WebGPU
• Stores vetted first-aid knowledge in IndexedDB (syncs from backend when online)
• Provides a chat-style interface for health questions
• Never prescribes medication or gives diagnoses
• Syncs updated medical guidelines only when online

• No user accounts or logins
• No real-time cloud AI calls during inference
• No diagnosis or treatment plans
• No medication recommendations or dosages
• No replacement for emergency services

• React + Vite
• Progressive Web App with service worker
• IndexedDB for:
  • LLM model weights (WebLLM cache)
  • Medical knowledge (synced from backend when online)

• @mlc-ai/web-llm in a Web Worker
• WebGPU / WASM inference
• Model: Llama-3.2-3B-Instruct (configurable)

• Node.js + Express
• MongoDB Atlas for medical content
• GET /api/medical — frontend fetches and caches when online
• POST /api/medical — webscraper or ingest script updates content

• Node.js 18+
• MongoDB Atlas account (or local MongoDB)
• Browser with WebGPU (Chrome 113+, Edge 113+, Safari 26+, Firefox 141+)

cd backend
cp .env.example .env
# Edit .env with your MONGODB_URI, PORT, FRONTEND_ORIGIN
npm install
npm run dev

Ingest initial medical content (with backend running):

npm run ingest
# Or: npm run ingest path/to/medical-knowledge.json

cd frontend
cp .env.example .env
# For API mode: set VITE_MEDICAL_SOURCE=api and VITE_MEDICAL_API_URL
npm install
npm run dev

Visit http://localhost:5173. On first load (while online), the model downloads and medical content syncs. After that, the app works offline.

{
  "version": 1,
  "entries": [
    { "id": "topic-id", "content": "Medical content..." }
  ]
}

The webscraper outputs this format. Use npm run ingest in the backend to push it to MongoDB.

• In-flight medical situations
• Hiking or camping emergencies
• Regions with limited internet access
• Disaster or outage scenarios

KIT AI is intentionally conservative:

• When uncertain, it escalates to "seek professional help"
• It avoids personalized or speculative advice
• It prioritizes clarity, calmness, and safety

Active development. Current features:

• WebLLM-based offline chat
• Medical knowledge cache with backend sync
• PWA with IndexedDB caching