Browser AI
A browser-native AI platform: pick a model, download it once, and run inference entirely on-device with WebGPU.
Stack TypeScript · Next.js · WebGPU · Transformers.js · IndexedDB
Most AI products are a form at the front and somebody else’s GPU at the back. Browser AI asks what changes when the model lives on the device: no API keys, no per-token bill, and a prompt that never leaves the laptop.
Local by construction
The app downloads a quantised model once, caches it in the browser, and runs inference through WebGPU. A worker owns the model so the interface stays responsive while tokens stream. There is no server-side inference path; if the model is not downloaded, the app has nothing to send.
That constraint changes the product. Model choice becomes a real trade-off between size, speed, and quality, so the picker explains the cost of each download instead of hiding it behind a friendly label. A progress bar is not decoration; it is the honest state of the product.
What it costs
Local inference is slower than a hosted endpoint and the first load is measured in hundreds of megabytes. The app says so. The result is a tool that is private by default, works offline after the first download, and makes the compute visible instead of pretending it is free.