Share my computer

Let your classmates use your computer’s graphics chip to run a small AI model, for as long as this tab is open.

What this does

Your browser downloads a small AI model (Qwen2.5 1.5B Instruct, 870 MB, once) and runs it on your computer’s graphics chip with WebLLM.
While you share, classmates who ask for pool/qwen2.5-1.5b get their answer written here, and the words go back to them as they are written.
Each job shows in the list above while it runs: the model, how many tokens, how many seconds.
Every job is also on the class usage page as served by you, at no cost.
One job at a time. If you stop or close the tab in the middle of one, the cloud finishes it.

It never opens your files, runs programs, or installs anything. A job is only “write text for this prompt”.
It doesn’t show or keep the prompts. They pass through this tab’s memory to the model and are gone.
It doesn’t tell you who asked, and it doesn’t tell them it was you (the class usage page does, as a pseudonymous id).
The model doesn’t learn from prompts; it is a fixed file.
Nothing runs when this tab is closed. No background service, nothing left running.
Your computer accepts no incoming connections. This tab connects out, to one address.

pool.visualia.space: this page, sign-in, and the job connection (one outgoing WebSocket)
cdn.jsdelivr.net: the WebLLM program, version 0.2.86
huggingface.co, *.hf.co: the model files, downloaded once
raw.githubusercontent.com: the model’s graphics code (mlc-ai/binary-mlc-llm-libs)

The browser blocks every other address for this page (its Content Security Policy).

The model files, in this browser’s storage for pool.visualia.space, and a sign-in cookie. Nothing else.
To stop: press Stop sharing, or close the tab.
To remove the model:
The teacher can revoke your key; school IT can block pool.visualia.space.

The code is in the Visualia repository, apps/pool.