Private on-device mental health journal powered by local AI.
Zero cloud. Zero data leaks. Your thoughts never leave your device.
Built for QVAC Hackathon I — Psy Models Track
MindMirror is a fully local, privacy-first mental health journal that uses multiple AI agents running entirely on your device to analyze your mood, detect stress patterns, and provide weekly reflections — without ever sending your data to the cloud.
| Agent | Model | Purpose |
|---|---|---|
| Agent 1 — MoodAnalyzer | Llama 3.2 1B Instruct | Analyzes journal entries for mood & stress scores |
| Agent 2 — MemoryAgent | GTE Large FP16 | RAG semantic search across all journal history |
| Agent 3 — ReflectionAgent | MedGemma 4B | Deep weekly mental health insight generation |
All models run via the QVAC SDK — 100% on-device inference.
No API keys. No internet required after first model download.
- 📓 Journal — Write daily thoughts, auto-analyzed by AI
- 📊 Analytics — Mood & stress trend charts from your data
- 🔍 RAG Chatbot — Ask anything about your journal history
- 🧘 Reflections — Weekly AI mental health insights via MedGemma
- 🔒 100% Private — All inference on-device, zero cloud
- Node.js v18+
- QVAC SDK installed
- 8GB RAM minimum
git clone https://github.com/Prithiv04/MindMirror.git
cd MindMirror
npm install
npm run devOpen http://localhost:3000 in your browser.
⚠️ AI models download automatically on first run (~2-4GB total).
Subsequent runs use cached models — fully offline.
User ↔ UI (Journal / Analytics / Reflections / Chatbot)
↓
Express Server (server.js)
↓
┌─────────────────────────────────────────┐
│ /api/analytics → Agent 1 (Llama 1B) │
│ /api/reflection → Agent 3 (MedGemma 4B)│
│ /api/chat → Agent 2 (GTE RAG) │
└─────────────────────────────────────────┘
| Layer | Technology |
|---|---|
| Runtime | Node.js v24 + Express |
| AI Inference | QVAC SDK (local) |
| LLM | Llama 3.2 1B Instruct Q4 |
| Medical AI | MedGemma 4B IT Q4 |
| Embeddings | GTE Large FP16 |
| Vector Store | QVAC RAG |
| Frontend | HTML + CSS + Chart.js |
- ✅ All inference runs locally via QVAC SDK
- ✅ No API keys required
- ✅ No data ever leaves your device
- ✅ Journal entries stored only in local filesystem
- ✅ Works fully offline after initial model download
Prithiv — @Prithiv04
Built solo in 11 days on an 8GB laptop. No GPU. No cloud. 🔥
MIT