Chrome Prompt API brings Gemini Nano to the browser

Google has introduced the Prompt API, a browser feature that lets developers send natural-language requests to Gemini Nano directly inside Chrome. The company is positioning it as a way to build AI-powered browser features without sending every prompt to a separate cloud service first.

The API is part of Google’s broader effort to bring on-device AI into the browser. Gemini Nano is the smaller, local version of Google’s Gemini model family, designed to run on supported hardware rather than in a remote data center. In practice, that means a web app or extension can ask the browser to perform a task such as extracting structured data, classifying text, or generating a short response.

Google says the Prompt API is aimed at features like AI search, content filtering, calendar extraction, and contact capture. Those are all tasks that fit a common pattern: a page contains messy text, and the browser helps turn it into something more useful. A shopping site could identify product details, a scheduling app could pull dates and times from a message, and a form could detect names, phone numbers, or email addresses.

The appeal here is not just convenience. Running model inference in the browser can reduce latency because the request does not have to travel to an external API and back. It can also reduce the amount of raw user data that needs to leave the device, which matters for privacy-sensitive workflows and for teams trying to keep some processing local.

Chrome has already been adding APIs for on-device AI, and the Prompt API extends that direction into a more general text-in, text-out interface. That makes it easier for developers to experiment with browser-native AI features without building their own model-serving stack. For many teams, the browser is already the place where users input text, read documents, and interact with forms, so putting inference there lowers the integration cost.

The trade-off is that browser-based AI depends on what the local device can support. Google has tied these features to specific Chrome capabilities and hardware requirements, which means availability will vary by machine and browser version. That is a different deployment model from a cloud endpoint, where the server does the heavy lifting and the client just sends requests.

Google’s documentation also points developers to permissions and testing details, including localhost workflows. That matters because browser features that touch AI, user text, or local resources usually need tighter controls than a standard JavaScript API. Security teams will care about where the model runs, what data is sent, and how much access the page gets before any production rollout.

The Prompt API also signals a shift in how web apps may be built. Instead of treating AI as something that lives only behind a vendor API, Chrome is making it possible to call a local model from the browser itself, with Gemini Nano handling the response inside the user’s session.