Free Tool Makes Google's AI Run Faster on Your Own Laptop

There's a free, open-source program called llama.cpp that lets you download an AI model and run it right on your own computer. Think of it as the opposite of ChatGPT: instead of your words traveling to a company's giant server farm, everything stays on your machine. It's hugely popular among tinkerers — roughly 129,000 people have starred it on GitHub, the site where code projects live.

The update in question, labeled "b11065," is small but practical. It tunes how Google's Gemma AI models run on Nvidia graphics cards — the chips that do the heavy math behind AI — from the Ampere generation (roughly 2020) onward. In plain terms: the same model, on the same hardware, now answers you faster. The release also ships ready-made versions for Windows, Mac, Linux, Android and iOS, so it works on almost anything you already own.

Why should you care if you don't code? Two words: privacy and money. Running AI locally means no subscription fee, no internet connection required, and no chance your private notes, medical questions or work documents get sent to a corporate server. If you already own a decent gaming PC or a recent Mac, you may already have enough horsepower to run a capable assistant for free.

The honest catch: this is one tiny tweak among hundreds of releases, and it only helps a slice of users — those with newer Nvidia cards who already use this tool. Setup is still fiddly, and local models are generally less clever than the big cloud ones. But the direction is clear: more of AI is quietly moving onto the device in your bag.