NemoAI

On-device AI chat — private, offline, always available.

Most AI chat apps send every message to a cloud server. NemoAI runs a full language model directly on your device — no internet required, no data leaves your phone.

• 100% offline — chat on airplanes, in tunnels, anywhere
• Zero data collection — nothing is sent to any server
• Instant responses — no network latency, no rate limits
• Open source — inspect, modify, and self-host everything

Runs LFM2-700M (Liquid Foundation Models, 700M parameters, Q4_0 quantized) locally using llama.rn — a React Native binding for llama.cpp. The model is bundled inside the app and ready on first launch.

• Full conversational UI with message history
• User messages in amber/orange bubbles, AI responses in blue
• Markdown rendering for formatted code blocks, lists, and emphasis
• Auto-scroll to the latest message

Fine-tune model behavior in the settings panel:

Changing context length triggers an automatic model reload.

• Radial gradient "Nemo AI" header with SVG mask
• Decorative glow orbs with Gaussian blur filters
• Frosted glass header and input bar (BlurView)
• Geist Mono custom font throughout
• Smooth linear gradient background

• Loading — shows "Loading LFM2-700M model..." while initializing
• Error — displays errors in-line with a red indicator
• Ready — enables the input bar and starts accepting messages

A separate Express.js server (hf_api_server.js) lets you ask questions about local files using Mistral-7B-Instruct-v0.2 via the Hugging Face Inference API.

curl -X POST http://localhost:3000/api/ask \
  -H "Content-Type: application/json" \
  -d '{"filePath": "./notes.txt", "question": "Summarize this file"}'

This is an optional companion server — the core chat app requires zero cloud connectivity.

• Node.js 18+
• Expo CLI (npm install -g expo-cli)
• Xcode (for iOS builds) or Android Studio

git clone https://github.com/<your-username>/NemoAI.git
cd NemoAI
npm install

# Start Expo dev server
npm start

# Run on iOS (requires native build for on-device inference)
npx expo prebuild
npx expo run:ios

# Run on Android
npx expo run:android

# Run on web
npm run web

Important: On-device inference requires a native build (npx expo prebuild && npx expo run:ios). The Expo Go client does not support native modules like llama.rn.

If you already have the app installed on a simulator, you can copy the model directly:

./copy-model-ios.sh com.anonymous.NemoAI

cp .env.example .env
# Edit .env and add your HF_TOKEN from https://huggingface.co/settings/tokens
node hf_api_server.js

NemoAI/
├── App.jsx                    # Main app — chat UI + on-device LLM inference
├── index.jsx                  # Expo entry point
├── app.json                   # Expo config (plugins, icons, splash)
├── hf_api_server.js           # Optional Express server for cloud-based HF inference
├── simulate_ai.js             # Standalone Node.js inference module
├── copy-model-ios.sh          # Script to copy model to iOS simulator
├── plugins/
│   └── withModelAsset.js      # Expo config plugin — bundles GGUF into iOS app
├── models/
│   └── LFM2-700M-Q4_0.gguf   # Quantized LLM model (~425 MB)
├── assets/
│   ├── fonts/                 # Custom fonts
│   └── Images/                # App icons & splash screens
└── .env.example               # Environment variable template

All parameters are adjustable at runtime via the in-app settings panel. See the Configurable Inference Parameters table above.

• Upload — attach files and images as context for the model
• Voice — speech-to-text input
• Video — multimodal input support
• Android optimization — tune performance for Android devices
• Model swapping — download and switch between multiple GGUF models

┌─────────────────────────────────────────────┐
│                   App.jsx                    │
│                                             │
│  ┌─────────┐  ┌──────────┐  ┌────────────┐ │
│  │ Chat UI │  │ Settings │  │ Model Init  │ │
│  │ (React  │  │ Panel    │  │ (llama.rn)  │ │
│  │ Native) │  │          │  │             │ │
│  └────┬────┘  └────┬─────┘  └──────┬──────┘ │
│       │            │               │        │
│       └────────────┴───────────────┘        │
│                    │                        │
│         ┌──────────▼──────────┐             │
│         │  LFM2-700M-Q4_0.gguf│             │
│         │  (bundled on-device) │             │
│         └──────────────────────┘             │
└─────────────────────────────────────────────┘

┌─────────────────────────────────────────────┐
│           hf_api_server.js (optional)       │
│                                             │
│  POST /api/ask                              │
│    ├── Read local file                      │
│    ├── Send to Mistral-7B via HF API        │
│    └── Return AI-generated answer           │
└─────────────────────────────────────────────┘

MIT

Built with ❤️ and llama.cpp