fragmede/llamatop

LlamaTop is a terminal dashboard for Mac computers with Apple Silicon chips that answers one question: is llama.cpp, a tool for running AI language models locally, actually doing any work right now. It runs like the classic Unix top command, but focused specifically on finding and tracking llama.cpp processes rather than every program on the system. Once running, it shows how much CPU and memory the matching llama.cpp processes are using, breaks down every CPU core individually, and reads Apple's GPU activity information to show how busy the graphics chip is. It also shows a full picture of system memory, including how much is wired, active, compressed, or free, since Apple Silicon Macs share one pool of memory between the CPU and GPU. All of this works without installing a background service, a kernel extension, or needing administrator access. The dashboard starts with a simple overview and expands with keyboard shortcuts, similar to pressing keys in top, to show more detail on CPU cores, GPU pipelines, or the full memory layout. Users can change the refresh rate, sort the process list by CPU or memory usage, and filter out idle processes. Unlike top, LlamaTop deliberately leaves out commands that could kill or change the priority of a running process, since the author did not want a monitoring typo to accidentally interrupt an AI model that is generating output. The README is careful to explain what each number actually measures and where it can be misleading, such as noting that GPU activity numbers include other apps using the graphics chip, not just llama.cpp, and that the GPU core display only shows how many cores exist rather than their real time activity. Building it requires a Mac with Apple Silicon, macOS 13 or newer, and Xcode's command line tools, after which it can be built and installed with a couple of terminal commands.