wtf is browserai?

BrowserAI is a TypeScript toolkit that lets developers run small AI models directly inside a user's web browser, using the browser's graphics hardware through a technology called WebGPU, instead of sending requests to a remote AI server. It comes with no built-in visual interface. Instead it is meant to be dropped into an existing website so a developer can build their own chat or voice interface on top of it. The toolkit ships with a curated list of 47 supported models and can handle text conversation, image understanding, speech to text transcription, and text to speech, using three different underlying engines depending on the type of model. It manages downloading and caching these models in the browser so they only need to be fetched once, tracks performance details like generation speed, and can constrain a model's output to match a specific data structure so responses come back as clean, structured JSON rather than free-form text. One of its more complete features is a ready-made voice assistant pipeline: a developer can wire up a microphone, have speech transcribed, get a reply from a text based AI model, and have that reply spoken back out loud, all chained together as a single reusable component with events a developer can hook into for building a custom interface. Some voice models can even start speaking a reply before the AI has finished generating the rest of it. To use BrowserAI, a developer installs it as an npm package into a JavaScript or TypeScript project. It requires a browser that supports WebGPU for most models, though a few lightweight speech models can fall back to running without it. Setting it up involves writing code to load a chosen model and call functions like generating text, transcribing audio, or synthesizing speech, rather than any manual installation step for an end user.