April 26, 2026
Your browser just became an AI diva
The Prompt API
Google puts mini AI in Chrome — fans cheer, others yell “22GB?!”
TLDR: Chrome’s Prompt API lets websites use a local AI on your computer for search, summaries, and more—after a hefty download and strict hardware needs. The community is split between cheering privacy-friendly, free local AI and slamming bloat, security risks, and weak performance, with test‑phase questions still swirling.
Google’s new Prompt API turns Chrome into a pocket-sized, on‑device AI studio that websites can tap for search, summaries, content filters, calendars, and more. It runs locally after an initial download, which means no data leaves your device, a big privacy win — in theory. But the crowd? Immediately split. One top joke summed up the vibe: “Sorry, to use our website, you must have 22GB free.” People roasted the hardware diet: desktop‑only, 16GB RAM, and more than 4GB of graphics memory, plus a big first download that auto‑removes itself if your storage drops under 10GB. Supporters fired back, calling it a free, privacy‑preserving “poor person’s ollama” that gives non‑tech users local AI without scary installs. Then came the paranoia: what if a shady site quietly uses your browser as an AI worker bee? Commenters imagined rogue scripts offloading work to visitors, while others dreamed up crowdsourced “mini‑cluster” compute — equal parts clever and creepy. Performance skeptics piled on, claiming the built‑in model fizzles after a couple of chat turns and pushing alternatives. Meanwhile, sharp eyes asked if this is still in a test phase and spotted a new “temperature” creativity dial on Chrome’s status page. Verdict: ambitious, brawny, and already a meme magnet.
Key Points
- •Prompt API enables natural-language requests to Gemini Nano running locally in Chrome for tasks like search, classification, and content processing.
- •Supported platforms include Windows 10/11, macOS 13+, Linux, and ChromeOS on Chromebook Plus (Platform 16389.0.0+); mobile Chrome (Android/iOS) is not supported.
- •Hardware requirements include ≥22 GB free storage, GPU with >4 GB VRAM or CPU with ≥16 GB RAM and ≥4 cores; audio input requires a GPU.
- •Initial model download requires an unmetered network connection; subsequent use is offline and sends no data to Google or third parties.
- •Developers should use LanguageModel.availability() and LanguageModel.create() (after user activation), and can use @types/dom-chromium-ai for TypeScript typings.