Google opened Prompt API testing in Chrome Canary this week: web pages can directly call the browser's built-in Gemini Nano model, with no API keys and no server requests required. We judge this to be the most noteworthy infrastructure upgrade in the browser space this year.
What this is
Prompt API is a new set of JavaScript interfaces in the Chrome browser that allows web pages to directly call the device's local AI model (currently Gemini Nano, a small model Google designed specifically for the device side). Developers write a line of window.ai to run inference in the user's browser and return results. The entire process requires no requests to the cloud, and data never leaves the user's device.
This is not "adding a chat window to the browser." This is turning AI inference into a native browser capability—just like when browsers first started supporting video playback and WebGL 3D rendering. Google calls it "Built-in AI," with a clear intent: AI should be browser infrastructure, not an external service.
Currently, this feature is only available in Chrome Canary (experimental version), requires manually enabling a flag, and is still in its early stages. But Google has already released complete developer documentation, sending a clear signal: this is not a tech demo, it's a product roadmap.
Industry view
Supporters see three clear benefits: first, privacy—data stays on the device, which is significant for sensitive scenarios like finance and healthcare; second, latency—local inference eliminates network round-trips, enabling millisecond-level interaction responses; third, cost—not calling an API means spending nothing, allowing small teams to add AI features to their products at zero cost.
But opposition is equally clear. Mozilla and WebKit have yet to express support, and W3C has no related standard discussions. Critics argue Google is using Chrome's market share advantage to unilaterally push a de facto standard, making the Gemini model the browser's "default brain"—the same path IE took by binding to Windows. A more practical concern is that Gemini Nano is a small model with limited capabilities; complex tasks will still require calling cloud APIs. If developers sacrifice "capability" for "local," the trade-off may not be worth it.
Our judgment: The value of the Prompt API lies not in what it can do today, but in defining the "browser-level AI" category. The standards war and model capability issues do exist, but the direction is right—on-device inference is a deterministic trend.
Impact on regular people
For enterprise IT: AI features for intranet applications have a new option—no need to procure external API services or worry about data compliance; simply let the browser do the inference. However, in the short term, the Chrome-only reality limits deployment choices, and enterprises won't easily accept a tech solution tied to a single browser.
For the individual workplace: Front-end developers need to start paying attention to the "on-device AI" tech track. In the future, "knowing how to call cloud APIs" and "knowing how to make the browser run models locally" will be two different skills; the former has a low barrier but severe homogenization, while the latter is temporarily niche but clearly differentiated.
For the consumer market: Regular users won't directly perceive the existence of the Prompt API, but they will gradually notice that AI features on certain websites are faster, and they no longer pop up prompts saying "data will be sent to the server." The experience improvement is implicit but real.