Free AI That Runs on Your Laptop Just Got a Small Fix

Most AI chatbots live on a company's servers: you type, your words travel across the internet, and someone else's computer answers. llama.cpp is the opposite. It is free, open-source software that lets a full AI model run directly on hardware you already own — a Mac, a Windows laptop, an Android phone, even a newer Snapdragon-powered handset. Nothing you type has to leave your device, and there is no subscription.

This release, numbered b11263, is a tiny housekeeping fix. It concerns PLaMo-2 and PLaMo-3, AI models built by the Japanese company Preferred Networks. Every AI model has a private shorthand for punctuation and instructions, and one of those symbols accidentally looked like the signal that means 'stop generating text.' The result: the model could cut itself off early or behave strangely. The fix teaches the software to look for the proper end-of-message marker, spelled <|plamo:eos|>, so ordinary sentences are no longer confused with 'I'm done.'

What is more interesting than the fix is the packaging. The same release ships ready-to-run versions for macOS, Windows, Linux, iPhone and Android, plus builds tuned for Nvidia, AMD and Intel graphics chips. That matters because it means the same free engine works whether you have a $3,000 gaming PC or a $600 phone. You download a file, pick a model, and chat — entirely offline, on a plane or in a basement with no Wi-Fi.

Here is the honest catch: this is maintenance, not a headline. Unless you are running a PLaMo model, you will notice nothing at all. But small fixes like this are exactly how free, private, on-your-own-machine AI quietly gets reliable enough for normal people to use. The big paid labs grab the attention; projects like this one quietly hand you the same capability for nothing.