[panchamk]

Hi HN! I built this because I was tired of watching long videos when I just needed the information.

Technical highlights: - Runs OpenAI's Whisper entirely in the browser using ONNX/WebAssembly - Videos never leave your device - all processing is client-side - Smart YouTube caption extraction (faster than AI when available) - Works with YouTube, Twitter/X, direct URLs, and local files - Three model sizes to match your device's capabilities

The biggest challenge was fitting Whisper models into browser memory constraints. I used quantized ONNX models and implemented chunked processing to handle large videos without OOM errors.

Stack: Next.js, Transformers.js, FFmpeg.wasm, TypeScript

Would love to hear your thoughts on: - WebAssembly performance optimizations - Handling larger models in browser memory - The privacy-first approach vs server-side processing

Happy to answer any questions!