The Lab · Live · runs on your machine
Real AI, running live in your own browser.
Not a video. Not a mockup. A small open model downloads once to your device and runs on your GPU — zero cloud, nothing sent to us. It's the whole thesis made literal: intelligence you own, on your machine, private by construction. Watch a council of AIs debate your question, generated token-by-token, locally.
A council of AIs, debating — running on your own machine.
Ask a question. Four voices — The Architect, The Adversary, The Optimist, The Pragmatist — each answer in turn, streaming live, then The Chair synthesises. Every token is generated on your GPU. Nothing is sent to us.
How this actually works
No server. No API key. No trust required — you can watch the network go quiet.
Opt in
Nothing downloads until you press the button. Then the model weights (a few hundred MB) come once, from a public CDN straight to your browser's cache.
Runs on your GPU
WebLLM compiles the model to WebGPU and runs it in a Web Worker on your own hardware. Your prompt and every token stay on your device.
Offline after
Once cached it works with the network off. We never see your question or the answer — there is no request to see. That's the point.
Being honest about it: this is a deliberately tiny open model (Llama 3.2 / Qwen 2.5, ~0.5–1.5B parameters) chosen so it fits and runs on a laptop. It is not our frontier system — it’s here to prove that useful AI can run locally, privately, at zero cost to us and to you. Expect small-model quality: quick, sometimes rough. It needs WebGPU — a recent Chrome or Edge on a laptop, desktop, or Android. On other browsers you’ll see a worked-through fallback instead.
The reusable pattern
The same engine, in one line — so any experiment can come alive.
The Council and this chat share one <LiveDemo> component. Historic lab reports can embed it to add a small, live, on-device demo of their own — the WebGPU gate, the opt-in download, the privacy note and the streaming all come for free.
Talk to a tiny model that lives on your machine.
The same in-browser engine as the Council, in the simplest possible form — a chat. This is the reusable pattern any lab report can embed to come alive.