Real AI in a single .html file — no key, no server, no bill

Here's the idea in one line: you make one index.html — a text box and a few buttons, "simpler / softer / shorter". You paste an awkward message, hit "softer", and get a cleaned-up version back. And the thinking isn't done by a server somewhere — it's done by the visitor's own browser. No key, no server, not a cent per request.

And here's what's new. "An app with AI" used to mean a backend: get a key, hide it, route requests through a server, pay for tokens. Now Chrome has a built-in model — Gemini Nano — with a direct call from JavaScript: LanguageModel. It runs on the user's device: the network is needed once to download the model, then everything is local. A year ago this just wasn't this easy — you needed a server. Now it's simply in the browser.

Why this one

We all hit the moment when text needs "tidying": soften a sharp email, simplify a dense paragraph, cut a wall of text. Routing that through a paid API just for yourself feels wasteful and fiddly with keys. Here the request never leaves the browser: free, private, and it even works offline after the first load.

And there's less "magic" than it looks. Your page is a pipe: it takes the text from the box, hands the built-in model one prompt, and shows the answer. All the difficulty lives in one clean instruction.

What you'll learn

• AI with no backend at all. For the first time you call a model not over the network with a key, but through one global object right in the page. No key — nothing to leak; no server — nothing to stand up.
• On-device vs cloud. You'll see why a local model is free, private, and offline — and where its limit is: it's small, so don't hand it a heavy job. That's a basic choice in every project.
• "The prompt is the feature." "Soften the text" isn't a separate technology. It's an instruction to the model: "rewrite it softer, keep the meaning." A good prompt is your main feature.

A ready starter prompt

Don't ask the agent to "make a site with AI" — out of habit it'll reach for a cloud key and stand up a server. Say it plainly: the model is built into the browser, no key, and describe the fallback:

A strong prompt leaves no room for guessing: you can see the model is built in, you can see the availability check, you can see the fallback for other browsers. The first result lands closer to what you wanted.

What you end up with

You wrote a slightly sharp message to a coworker. You open your little page, paste the text, hit "softer" — and a second later you get the same meaning, but human. You copy it, send it. Then you drop the link to a friend: they open it in Chrome and it just works for them too — no signup, no key, no bill. You didn't pay for the request. You never even reached for a key — there isn't one.

One bit of honesty to close: the in-browser model is small and, for now, lives in Chrome on desktop. For long texts and other browsers you'll wire in a cloud model later — but the first free, private version ships as a single file.

Source: Chrome for Developers — Prompt API: Gemini Nano right in the browser, no key