PrivaKit – A zero-upload AI workspace in the browser via WebGPU
WebGPU AI workspace for sensitive docs that actually passes the airplane mode test.

This ships a tiny 22MB CNN that runs entirely in the browser using WebGPU/ONNX — impressive engineering that delivers instant, private 4x upscaling with a WebAssembly fallback. The UI is minimal but functional (drag-and-drop, GPU/CPU toggle, quantized mode), though the feature set is thin: max 1024px, no batch processing, and little info about model trade-offs or sample comparisons.
Photographers, designers, hobbyists, web developers and privacy-conscious users who need quick, local image upscaling
WebGPU AI workspace for sensitive docs that actually passes the airplane mode test.
Runs 208MB LaMa model locally via WebGPU, keeping images off servers.
772 MB model runs entirely in-browser with no backend, API calls, or telemetry whatsoever.
Explicit kernel control over TVM-style black boxes, but benchmarks show mixed wins vs Transformers.js.
5.6x realtime on CPU with voice cloning beats most local TTS options.
Policy.toml loaded before every action prevents the context compaction failures that broke OpenClaw.