A modern, privacy-first AI chatbot app for Android and iOS, running Google's Gemma 3N model entirely on-device. Supports both text and image (multimodal) conversations, with a beautiful, accessible UI and robust offline capabilities.
- Local AI Processing: All inference runs on your device—no cloud required, ensuring privacy and offline use.
- Gemma 3N E2B IT Model: 1.5B parameter, instruction-tuned, multimodal (text + image) model from Google, optimized for mobile.
- Multimodal Chat: Send text, images, or both. Get AI insights on photos, screenshots, and more.
- Smart Model Management: Guided download, resumable progress, and update checks for the 3.1GB model file.
- Persistent Conversation History: Your chats are stored locally and can be exported/imported as JSON.
- Rich UI/UX: Clean, accessible chat interface with light/dark mode, error handling, and conversation stats.
- Privacy by Design: No data leaves your device. All processing and storage is local.
- Flutter SDK: >=3.8.1
- Android: API 24+ (Android 7.0+), 4GB+ RAM, 4GB+ free storage
- iOS: iOS 16.0+, 4GB+ RAM, 4GB+ free storage
git clone https://github.com/yourusername/gemma3n_app.git
cd gemma3n_appflutter pub get- Create a
.envand add your Hugging Face token asHF_TOKEN=your_token_here. - Or, enter your token in the app during first launch.
- Android:
flutter run -d android
- iOS:
flutter run -d ios
- On first launch, you'll be guided to download the Gemma 3N model (~3.1GB).
- Requires a Hugging Face account and token (free for most users).
- Download progress, errors, and status are shown in-app.
- Model is stored securely in your app's documents directory.
- Chat: Type messages or attach images to interact with the AI.
- Multimodal: Combine text and images for richer queries (e.g., "Describe this photo").
- Conversation Management: View stats, clear history, or export/import conversations.
- Model Management: Check model status, update, or re-download as needed.
- ModelManager: Handles download, initialization, and lifecycle of the Gemma model.
- ChatService: Manages conversation flow, message persistence, and AI inference.
- ImageService: Handles image picking, compression, and cleanup for multimodal queries.
- UI Layer: Modern, accessible chat interface with theming and error handling.
- Light & Dark Mode: Auto-detects system theme, with custom color palettes for both.
- Typography: Large, readable fonts with proper contrast.
- Accessibility: Screen reader support, large touch targets, and high-contrast options.
- Style Guide: See STYLE_GUIDE.md for design tokens, layout, and component specs.
- All AI processing is local—no user data or queries are sent to external servers.
- Conversations and images are stored only on your device.
- Clear all data: Option to delete conversation history and model files at any time.
We welcome contributions! Please:
- Follow the STYLE_GUIDE.md for code and UI consistency.
- Write clear commit messages and document your code.
- Add tests for new features where possible.
- Open issues or pull requests for bugs, ideas, or improvements.
- Model download fails: Check your Hugging Face token and internet connection. Resume is supported.
- App crashes on startup: Ensure your device meets the RAM/storage requirements.
- Image upload issues: Grant storage/camera permissions and use supported formats (JPEG, PNG, WebP).
- Other issues: See PROJECT_REQUIREMENTS.md for known risks and mitigations.
For more details, see PROJECT_REQUIREMENTS.md and STYLE_GUIDE.md.