🧠 Overthinker

🤖 Model

Powered by nvidia/nemotron-3-nano-30b-a3b via OpenRouter. Optimized for structured decision tree generation with dual-prompt architecture (options vs outcomes).

🌍 Framework

Built on Gradio.Server with SQLite per-session persistence. Fully client-side rendering with D3.js for interactive tree visualization. Browser memory fix applied: zoom handler cleanup, addOptions replaces children, D3 keyed exit/remove.

✨ Key Features

• 🌳 Generate trees from any decision prompt
• 🧠 Dual prompt design: Options vs Outcomes
• 🔍 Interactive explore / regenerate / add options
• 📍 Path sidebar with depth tracking & tips
• 🖼 Export: SVG, JSON, Markdown, PNG
• 🌓 Dark/Light theme toggle
• ⌨ Keyboard shortcuts (E, R, A, F, C)
• 🔄 Regenerate with user context
• 💾 Multi-user session isolation via SQLite
• 📤 Upload trace to HF dataset
• 🧩 Browser memory optimized: data arrays cleared, zoom listener cleanup, addOptions replaces children

💬 OpenRouter

API calls go through openrouter.ai/api/v1/chat/completions with model nvidia/nemotron-3-nano-30b-a3b. Temperature: 0.8, Max tokens: 2048.

1. Enter a Decision

Type or paste a decision into the text box (e.g., "Should I quit my job to start a startup?") and click Generate or press Enter.

2. Explore the Tree

Click any node to see its details. Use the popup to Explore (generate more branches), Regenerate (with your feedback), or Add Options.

3. Navigate

• Click node = select + details popup
• Right-click node = context menu
• Drag canvas = pan view
• Scroll = zoom in/out
• E = Explore, R = Regenerate, A = Add Options
• F = Fit view, C = Clear
• Esc = Close popups

4. Track Your Path

Click the Path tab on the right to see your current decision path with depth and tips. Export path as SVG, JSON, Markdown, or PNG.

5. Export

Use the toolbar buttons to export the entire tree or just the selected path in your preferred format: SVG, JSON, Markdown, or PNG.

6. Upload Trace

Click the Upload Trace button in the toolbar to upload your decision tree to a shared HuggingFace dataset for public analysis.