Goku is a new WASM (wllama)-powered LLM inference and model manager showcased on Hacker News. Built by userfrom1995, the project enables running large language models directly in the browser using WebAssembly, eliminating the need for server-side infrastructure. The demo site at github.io/goku provides a model manager interface where users can download, configure, and interact with various LLMs locally.

This client-side approach leverages wllama, a WebAssembly port of llama.cpp, to perform inference entirely within the browser sandbox. Users benefit from privacy-preserving execution since no data leaves their device, and offline capability once models are cached. The manager supports multiple model formats and quantization levels, allowing trade-offs between performance and quality.

The Show HN submission received 4 points with 0 comments at time of posting, indicating early-stage visibility. As a pure client-side solution, Goku represents a growing trend toward edge AI deployment, making LLM capabilities accessible without GPU servers or API keys.