voice coinbase agent
This is a browser extension, which uses transformers.js and webgpu, to run the AI 100% locally.
Demo
About
This is a transformers.js project, which has automatic speech recognition, text to speech, and a language model inside of a chrome extension, which runs entirely on the client device, so they can ask common questions like "what is my balance" and "what is the price of ethereum" as a spoken query.
How it's made
it is made using the transformers.js and coinbase agentkit transformers.js has backends for webgpu , web assembly, and webnn, which are libraries that sandbox the computation in the browser, but ask the operating system to perform the matrix multiplications, and is supposed to be cross platform but in reality i had some problems getting vulcan to work in linux, but it should work fine in windows.