Wllama
WebAssembly binding for llama.cpp that runs LLMs entirely in the browser.
About
Wllama compiles llama.cpp to WebAssembly so large language models can run directly in the browser with no backend and no GPU required. Inference happens inside a web worker using WebAssembly SIMD, with optional WebGPU acceleration in the newer V3 release.
It ships as a zero-dependency npm package with an OpenAI-compatible API, TypeScript support, multimodal input, and tool calling. A community-maintained fork by Reese Levine adds WebGPU and multimodal support on top of the original project by Xuan Son Nguyen.
What it can do
- llama.cpp compiled to WebAssembly
- OpenAI-compatible completions API
- WebGPU acceleration and tool calling in V3
- Zero runtime dependencies
Setup
Install the npm package @wllama/wllama, load a GGUF model from the Hugging Face Hub, and run completions inside a web worker.
Prerequisites
A modern browser; for multithreaded builds, COOP/COEP response headers on the hosting page.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 116; Trending score 70/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.