erdemozkan/kiln: WebGPU-native LLM inference for the browser

Kiln is a WebGPU-native JavaScript library that runs language, vision, and speech models directly in the browser using original weights and GGUF files without requiring conversion or a server. The author developed the library to repurpose idle hardware on modest devices like older laptops. Benchmarks indicate Kiln is faster end-to-end than transformers.js across tested tasks on standard hardware. Open article →