BrowserAI

BrowserAI is an open source JavaScript library for running small language models locally inside a web browser. It uses WebGPU and WebAssembly to load and run models on the user's own device, so an application can generate text, hold a chat or process speech without calling a cloud API, holding API keys or maintaining servers. It is built for web developers adding AI features to their applications, companies that need privacy conscious AI because data never leaves the browser, and researchers experimenting with browser based inference. Installation is an npm package, and loading a model and generating text takes a few lines of code. The library switches between several engines, including MLC, Transformers and Flare for GGUF models through WebAssembly, and ships preconfigured models. Other capabilities include Web Worker support to keep interfaces responsive, structured output against JSON schemas, speech recognition, text to speech, audio source separation that isolates vocals, drums and bass, and a built in database for conversations and embeddings. After the initial model download it works offline. Live demos cover chat, voice chat and text to speech. The library is free under the MIT licence, and the team describes it as having zero operational cost because inference runs on the user's hardware. The pricing page on the site reads "Coming Soon" with a contact form, and no paid plans are published. The same team also runs Browseragent, a separate no code platform for building AI agent workflows on top of the BrowserAI library.