amberlin.ai
Amberlin's own page on the web: a small model says hello right there in your browser, on WebGPU where the browser has it, and nothing you type leaves your device.
- Status
- Live at amberlin.ai
- Licence
- Business Source License, GPL-3.0 from 1 May 2033
- Built with
- Cloudflare Workerstransformers.jsWebGPU
- Part of
- Amberlin AI
- Source
- amberlin-worker
amberlin.ai is where Amberlin introduces herself. Press Wake Amberlin and a small language model, Qwen2.5-0.5B-Instruct, loads into the page and runs there: on the graphics card through WebGPU where the browser has it, on the CPU where it does not. Say hello and she answers.
It is a hello, not the assistant itself: the real Amberlin lives on the Linux Mint desktop, with speech, tools and a bigger model, and the page links there.
Everything the page loads comes from amberlin.ai itself, with no outside scripts or fonts, and a strict security policy holds it to that. The model’s files come through the site from Hugging Face, cached for a month, and only that one model can be fetched that way.
Install
Nothing to install: open https://amberlin.ai in a browser. The first visit downloads the model, about 500 MB, which the browser then keeps.