Llama.cpp Update Fixes Memory Drain for Mac Users Running AI Locally

There's a quiet but welcome fix for people who run AI directly on their Apple devices. Llama.cpp, a free open-source tool that lets you run AI models like ChatGPT on your own Mac, iPhone, or PC, just released a new update called b10819. The change is simple: it fixes a memory leak on Apple's Metal graphics system. In plain terms, earlier versions of the tool would slowly use up your device's memory and never give it back, causing slowdowns or crashes after long chats. Now that memory is released properly, especially when the app is quitting or switching tasks.

Why should you care? Because running AI locally has real benefits: it's free, it works offline, and your conversations stay private since nothing goes to a cloud server. For Mac and iPhone users, this update makes that experience more reliable, especially if you like to keep an AI assistant running in the background for hours. It's like fixing a dripping faucet — not a new feature, but it stops a slow leak that could ruin your setup.

The update comes with ready-to-use builds for macOS on Apple Silicon and Intel chips, iOS, and even Android and Windows. Developers and power users can grab it right away from the project's site. For the rest of us, it means the apps that use llama.cpp — like local chat assistants you download — will gradually become more stable on Apple devices.

The catch: this is a very technical, incremental fix, not a flashy upgrade. If you're just using cloud-based AI like ChatGPT, you won't notice anything. But if you value privacy or want AI on an old Mac without subscription fees, these steady improvements are what make that possible.