The Free Tool That Runs AI on Your Own Laptop Just Got an Update

Most AI chatbots you've used — ChatGPT, Gemini, Claude — don't actually live on your computer. Your questions travel to a company's data center, get answered there, and come back. That's why they cost money each month and why privacy-minded people worry about what happens to their conversations. llama.cpp is the free, open-source alternative. It's a small engine that squeezes big AI models down enough to run on ordinary hardware: a laptop, a desktop, even a recent phone. No internet connection required, no account, no bill.

The release flagged here is build b11027, a pre-release update published on GitHub on 17 September. Most of the news is plumbing rather than fireworks. The team ships ready-made versions for a long list of setups: Windows on standard and ARM chips, Macs with Apple Silicon, Intel Macs, Linux on several processor types, Android, and even Chinese-made openEuler servers. There are also builds that tap into graphics cards from Nvidia (CUDA), AMD (ROCm) and Intel (OpenVINO, SYCL), plus Vulkan, a general graphics standard — meaning your gaming GPU can help answer questions faster.

The one actual code change in this build is a fix to how the software handles 'RPC' mode — a way of spreading one AI model across several computers on your home or office network, like splitting a big job between coworkers. The fix tells it to skip certain accelerator devices that were causing trouble. Unless you're running AI across multiple machines, you probably won't notice anything.

The honest catch: this is a developer release, not a polished app. There's no friendly installer, no customer support line, and the notes are written for programmers. If you're curious, the easiest route is a free front-end app that bundles this engine for you. Otherwise, this update is mostly good news for the growing crowd who'd rather own their AI than rent it.