Thinking Tiny Thoughts

Thinking Tiny Thoughts

Thinking Tiny Thoughts is an interactive Proof of Concept (PoC) demonstrating that modern Large Language Models can run entirely inside your web browser. Zero installations, zero plugins, zero cloud dependencies, and zero API costs.

Everything you type is processed 100% locally on your computer's CPU. No account required, no tracking, and no data ever leaves your machine.

What Makes This Useful

• Zero Installation: Click and run directly in your browser. No Python environments, no CUDA setup, no gigabyte downloads.
• 100% Private and Client-Side: All neural network computations happen strictly in your browser's local memory. You can even disconnect your internet after loading and it will continue to generate text.
• Smooth Interactive UI: Built with high-performance raylib graphics featuring real-time token streaming, a live performance metrics dashboard (Tokens/Sec, Latency, First-Token time), and smooth dark/light themes.
• Demonstrating What is Possible: A lightweight showcase of what self-contained, client-side AI integration in web apps and games can look like today.

How to Try It

1. Load the Page: Allow the browser to stream the model weights (SmolLM2-135M) into local memory.
2. Start: Press Enter or click to begin.
3. Ask Anything: Type a question, prompt, or greeting in the input box and press Enter.
4. Watch It Generate: See the model stream words in real time alongside live hardware telemetry.
5. Customize: Click the theme toggle button in the bottom right corner to switch between dark and light themes.