Free Tool That Runs AI on Your Own Laptop Just Got More Stable

Most people meet AI through ChatGPT or Gemini, where your questions travel to a company's servers and you pay a monthly fee for heavy use. llama.cpp is the opposite approach: a free, volunteer-built program that downloads an AI model onto your own computer and runs it there. Nothing is sent anywhere. No subscription. It works on a plane, in a basement, or on a laptop with the Wi-Fi switched off.

The update released September 13, labeled b10938, is a maintenance release rather than a flashy new feature — and that is normal and healthy for software like this. Its headline fix addresses a bug in Vulkan, the graphics-layer software that lets AI calculations run on your computer's graphics card instead of its main processor. In plain terms: when two AI jobs tried to run at once, certain graphics drivers could lose track of each other and freeze or crash the app. The patch forces those jobs to take turns, a small slowdown in exchange for reliability.

Why should a non-engineer care? Because this is the plumbing that determines whether local AI feels like a real product or a science experiment. Every crash fixed is one less reason to give up and go back to a paid cloud service. The release also includes pre-built packages for Windows (with Nvidia and AMD graphics support), Mac, Linux, Android, and iPhone and iPad — meaning you can usually install it without touching a command line.

The catch: llama.cpp is not a consumer app with a support line and a friendly setup wizard. It is a tool built by volunteers, and getting a model running still takes some patience, a decent amount of free disk space, and ideally a modern graphics card. If you want zero friction, ChatGPT still wins. If you want privacy, no bills, and full control, this is the free option quietly getting better every week.