BrowserAI

Why we included this project

BrowserAI is a JavaScript SDK that pushes inference off the server entirely, loading small models like Llama-3.2, Qwen3, Gemma, and DeepSeek-R1 distill checkpoints straight into a web page and running them locally on WebGPU. Teams building privacy-conscious web apps get chat, structured JSON output, speech recognition, and text-to-speech without standing up GPU infrastructure or paying per token. The SDK wraps several runtimes (MLC, Transformers.js, a WASM-based GGUF engine, and Demucs) behind one API, so you can swap engines and move between text, audio, and embedding models without rewriting your integration. Pages work offline after the first model download and include a built-in database for conversations, which makes the whole thing an easy playground for hobbyists and researchers building no-code tools. The one caveat is that the curated model list is fairly small, so confirm your target model runs acceptably on the devices you care about before committing.