Wavu AI v1
Offline AI assistant powered by the SmolLM 135M ONNX model and adapted for the Wavu AI v1 mobile application.
🧠 What is Wavu AI?
Wavu AI is a lightweight offline AI assistant designed for:
- 📱 Mobile devices
- 📴 Offline usage
- ⚡ Fast local inference
- 🧩 JSON-based knowledge customization
- 🤝 Collaborative exploration and guidance
Wavu AI is designed around the idea of:
“Let’s explore and figure things out together.”
Instead of acting as a perfect authority, Wavu AI works best as a lightweight thinking partner for brainstorming, explanations, learning, and guided exploration.
⚙️ Model Information
This project uses:
- SmolLM 135M ONNX model
- Adapted and optimized for the Wavu AI v1 mobile inference pipeline
The model was integrated specifically for:
- mobile inference
- reduced memory usage
- offline execution
- JSON-grounded responses
- lightweight conversational assistance
Original model source:
✨ Features
- ✅ Fully offline AI
- ✅ Mobile optimized inference
- ✅ ONNX Runtime compatible
- ✅ Lightweight memory footprint
- ✅ Fast startup and responses
- ✅ JSON knowledge grounding
- ✅ User-customizable knowledge files
📚 Knowledge Customization
Wavu AI supports lightweight JSON-based grounding using files stored locally.
Example structure:
1Documents/Wavu/
2├── soul.json
3└── general_knowledge.json
🧠 soul.json
Defines Wavu AI’s core identity and behavioral foundation.
Example:
1{
2 "identity": [
3 "I am Wavu AI",
4 "I run on SmolLM 135M"
5 ],
6 "rules": [
7 "Work with the user to explore and figure things out",
8 "Stay clear and practical"
9 ]
10}
📖 general_knowledge.json
Provides lightweight grounding and capability awareness.
Example:
1[
2 {
3 "Capabilities": "Wavu AI helps explain ideas and explore solutions.",
4 "Limitations": "Wavu AI may be incorrect or incomplete."
5 }
6]
📱 Intended Use Cases
Wavu AI works best for:
- Idea exploration
- Learning assistance
- Brainstorming
- Offline help
- Lightweight guidance
- Mobile AI experimentation
⚠️ Limitations
Wavu AI is built on a small language model.
It may:
- produce incorrect answers
- misunderstand context
- hallucinate information
- struggle with advanced reasoning
This project prioritizes:
- speed
- offline access
- accessibility
- lightweight deployment
over large-scale model intelligence.
🔧 Technical Notes
- Base Model: SmolLM 135M ONNX
- Runtime Target: Mobile / Edge Devices
- Inference Type: Fully Local
- Integration: Wavu AI v1 mobile pipeline
- Grounding: JSON-based contextual injection
🚀 Goals
Wavu AI explores how small offline language models can become useful everyday assistants through:
- practical design
- lightweight grounding
- mobile accessibility
- local-first AI experiences
The project focuses on making AI more accessible, customizable, and usable without requiring constant internet access.
📜 License
This project is released under the Apache License 2.0.
You are free to:
- use
- modify
- distribute
- build upon
the project under the terms of the Apache 2.0 license.
Please also review the original SmolLM model license and attribution requirements where applicable.
Apache License 2.0:
🤝 Contributions
Experiments, optimizations, mobile inference improvements, and lightweight grounding systems are welcome.
