Show HN: PrivaKit – A zero-upload AI workspace in the browser via WebGPU

Category: ai-ml

Tags: privacy, client-side-ai, webgpu, transformers, local-processing

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 4, Utility: 7)

PrivaKit is a browser-based AI workspace that runs models like Whisper and transformers.js locally via WebGPU, enabling sensitive document and audio processing with zero data upload. It's interesting because it addresses privacy concerns in AI workflows by keeping all processing client-side, with the creator providing a technical audit to verify the data flow. The project demonstrates the practical application of emerging browser capabilities for on-device machine learning.

Target audience: data engineers, backend devs, security professionals, privacy-conscious users

View on Hacker News