ToDoozAI：基于 Gemma-3 270M 的浏览器端任务与专注助手

⚡ ToDoozAI: In-Browser Smart Task & Focus Assistant

A privacy-first productivity dashboard with task-bound ToDoozie timers, local WebGPU AI breakdowns, and an integrated Spotify player.

💡 The Vision: Lightweight, Frictionless AI & Focus Tools

ToDoozAI explores a simple question:
"Can we build useful, modern productivity tools with local AI that run 100% inside the browser without backend servers, subscription costs, or exposed API keys?"

Most AI utilities rely on heavy cloud APIs or server infrastructure. ToDoozAI shifts everything to the client side. By pairing Task-Bound ToDoozie Timers, an in-browser LLM (@huggingface/transformers), LocalStorage persistence, and an embedded Spotify Focus widget, it offers a complete focus workspace hosted directly on GitHub Pages for $0.

✨ Key Features

⏱️ Task-Associated ToDoozie Timers: Each task features its own dedicated 25-minute countdown timer. Starting a timer automatically shifts task status to In Progress and auto-completes it upon session finish.
🤖 In-Browser AI Task Breakdown: Uses Gemma-3 270M (ONNX) running locally to decompose complex, daunting tasks into 3 actionable 25-minute steps.
🎵 Floating Spotify Focus Widget: Fixed in the upper-right corner for quick access to curated lo-fi and focus playlists while working.
📊 Three-State Task Lifecycle: Track progress across Not Started, In Progress, and Done with clickable status badges.
💾 LocalStorage Persistence: All tasks, active subtasks, and remaining timer states persist automatically across browser refreshes.
🔒 100% Local & Private: Prompts and task data never leave your browser. Zero backend telemetry, zero API key requirements.
⚡ WebGPU / WASM Accelerated: High-performance local AI execution powered by your browser's graphics processing or WASM fallback.
💸 $0 Cost to Host: Single-file static architecture native to GitHub Pages.

🛠️ How It Works Under the Hood

Client-Side Architecture: The entire app runs out of a single index.html file using standard web APIs and CDN-loaded ES modules.
Model Caching: On first execution, the quantized ~270M parameter model is cached locally in IndexedDB, allowing subsequent task breakdowns to generate quickly.
Smart State Management: Starting a focus timer on one task automatically pauses any other running timer to keep you focused on a single priority at a time.

🚀 Live Demo & Quickstart

1. View Live

Try the live app on GitHub Pages:
👉 https://<your-username>.github.io/<your-repo-name>/

2. Run Locally

No npm install, Node.js server, or build step required!

Clone the repository:

git clone [https://github.com/](https://github.com/)<your-username>/<your-repo-name>.git

📊 Tech Stack

Frontend: HTML5, CSS3 (CSS Variables & Flexbox), Vanilla JavaScript (ES6 Modules) AI Engine: @huggingface/transformers v3 Model: onnx-community/gemma-3-270m-it-ONNX Integrations: Spotify Embed iFrame API Persistence: HTML5 localStorage & IndexedDB Hosting: GitHub Pages

🔮 Future Roadmap

[ ] Custom ToDoozAI duration sliders per task [ ] Drag-and-drop task reordering [ ] Export completed tasks & session logs to Markdown/CSV [ ] Custom Spotify playlist URL input in widget settings

🤝 Contributing

Contributions, feature requests, and bug reports are welcome! Feel free to open an issue or submit a pull request.