Nova Chat UI + AI Studio v1.0.0 – On-Device Gemma AI for Flutter

🚀 Nova Chat UI + AI Studio v1.0.0 is a Flutter AI chat application built to demonstrate on-device AI using Gemma models.

This release includes support for running Gemma 3 1B and Gemma 3 270M directly on supported devices, allowing developers to experiment with private, local AI inference without sending every prompt to a cloud API.

✨ Key Features
🧠 Gemma 3 1B On-Device AI
⚡ Lightweight Gemma 3 270M model support
📥 Hugging Face model download and installation
💬 Real-time streaming AI responses
📝 Markdown-rendered AI messages
🔄 Switch between Local Gemma, OpenAI, Gemini and Groq
🎙️ Voice input with speech-to-text
🔊 Optional text-to-speech support
🔐 Secure API key and Hugging Face token storage
🧩 Riverpod state management
📱 Responsive Flutter UI for mobile, desktop and web
🌙 Light and dark theme support
🌐 Experimental browser/WebGPU support
🔒 Local AI inference for better privacy
💰 Reduced dependency on paid AI API requests
How On-Device AI Works

The first time a local Gemma model is selected, the application downloads the compatible model file.

After installation, supported prompts can be processed directly on the user's device:

User Prompt → Flutter App → Gemma Model on Device → Streaming AI Response

This makes the project useful for experimenting with:

Offline AI assistants
Private AI chatbots
Local productivity tools
Educational AI apps
AI features with lower cloud API usage
Hybrid local + cloud AI applications
Supported Local Models
Gemma 3 270M IT – lightweight and faster
Gemma 3 1B IT – better response quality
Gemma 3n E2B support
LiteRT-LM compatible Gemma model support
Technology Stack

Flutter • Dart • Riverpod • flutter_gemma • MediaPipe • LiteRT-LM • Hugging Face • Markdown • Speech-to-Text • Text-to-Speech

Getting Started
flutter pub get
flutter run

Before downloading gated Gemma models, make sure you have accepted the required Gemma model terms on Hugging Face and configured a valid read-access token where required.

Recommended Model

For lightweight testing:

Gemma 3 270M

For better local AI responses:

Gemma 3 1B

Performance depends on device RAM, CPU/GPU capability, operating system and selected inference backend.

Important Note

On-device AI does not always mean zero internet usage. An internet connection may be required for the first model download. Once a compatible model is installed, supported inference can run locally without sending each prompt to a cloud AI provider.

Built with Flutter for developers exploring practical on-device and hybrid AI applications.