what is llmctl fr?

llmctl is a terminal based tool for managing local llama-server instances, the server component of llama.cpp that runs large language models on your own machine. It gives you both a full screen text interface and a set of direct command line commands, so you can start, stop, and inspect model servers without retyping a long list of settings every time. Normally, running llama-server by hand means reconstructing a long command with the port, context size, GPU layers, sampling settings, and other flags every time you want to load a model, with nothing keeping track of what was used last time or why a server crashed. llmctl solves this by letting you save named profiles for each model, where a profile holds every llama-server setting you care about. From then on you can run, inspect, or stop any of these profiles from one screen instead of typing commands from scratch, and the whole configuration is stored as plain YAML so it can be versioned or shared with others. The interface is organized into tabs for browsing imported models and their profiles, quickly rerunning recently used setups, adjusting settings like model folders and the llama-server executable path, and inspecting currently running servers with their live health status and token generation rate. There is also a network tab on Linux for managing connections used in distributing model computation across machines, when that feature is turned on. Adding a model involves pointing llmctl at a folder containing GGUF model files, the common file format for these local models, after which llmctl scans the folder and lets you import the ones it finds, automatically creating a default profile so a model can be run right away. Profiles can also be exported as ready to run shell scripts or copied directly for use elsewhere. Each running server shows a colored status indicator, gets its output logged automatically to a file, and exposes an OpenAI compatible web address that other tools can be pointed at to talk to it. llmctl requires llama-server itself to already be installed, along with GGUF model files or a Hugging Face based setup, and runs on Windows, macOS, and Linux, with the executable simply needing to be placed on the system path.