Hi all, I'm Furkan. aidekin is an AI chat widget you add with one script tag. The catch, and the whole point, is that it runs entirely in the visitor's browser: a ~1.7B model (Bonsai) on WebGPU, no server, no API keys, $0 per message no matter how much it gets used. Whatever a visitor types stays on their machine, and after the first load it works offline. To make a model that size usable in a browser I wrote my own inference engine, bitgpu, a 1-bit WebGPU runtime built from scratch. It keeps the weights packed in ~0.5 GB of VRAM instead of blowing up to several GB, and it's the part I'm most proud of. It's open source too (MIT, both repos). Honest limits: it's a ~290 MB download the first time (cached after, offline once you have it), so it's for places a visitor is already committed (docs, help panels, internal tools etc.) rather than a landing-page popup. It needs WebGPU, so desktop Chrome or Safari is the safe bet and mobile is effectively out for now. The model is small, which is exactly why it answers from a knowledge file you give it instead of pretending to be a general chatbot. There's an on-device voice mode too, but it's beta and pulls ~1.6 GB more. No signup. Click the bubble on aidekin.com, or the cafe demo where it answers from that shop's own info. Tear it apart and tell me what breaks. Thank you :)

aidekin

An AI chat widget that runs in the visitor's browser

An AI chat widget that runs in the visitor's browser