llama.cpp Update Lets More Devices Run AI Without the Cloud

There's a free piece of software called llama.cpp that does something quietly revolutionary: it lets a normal computer run an AI chatbot all by itself, with no internet connection and no monthly bill. Most AI assistants live on giant company servers — you type, their computers think, and you pay for the privilege. This project, with about 128,000 "stars" (think likes) on GitHub, flips that around. Download it, download an AI model, and the thinking happens on your own machine.

On September 12, the team published a new build called b10933. The interesting part isn't any single feature — it's the sheer spread of ready-to-install versions. There are downloads for Windows, Mac (both Apple Silicon and Intel), Linux, Android, and iPhone, with extra versions tuned for Nvidia, AMD, Intel, and Qualcomm graphics chips. In plain terms, whatever device you own, there's probably a version built for it. You pick your device, download, and run.

The actual code change in this release is tiny and technical: better support for "jinja" templates. A template is the instruction sheet that tells an AI model how to format a conversation — where your question ends and its answer begins. The fix handles numbers written a certain way inside those instruction sheets, so fewer models break or produce garbled answers when you load them. It's plumbing, but plumbing that keeps the lights on.

The catch is real. This is a "pre-release," meaning beta software that may still have bugs. It's built for people comfortable with a command line, not a friendly app with a big blue button. You need a reasonably modern computer with 8–16GB of memory, and you have to find and download the AI models yourself. The answers won't match the best paid chatbots either. But it's free, it's private, and nobody is logging your questions.