llama-buddy
A command-line wrapper around llama.cpp's llama-server that gives it an Ollama-like workflow: starting and stopping a background server, downloading and inspecting GGUF models interactively, and automatically applying a model's recommended sampling parameters.
llama-buddy manages a llama.cpp llama-server instance as a background daemon, adding the model management, download and inspection conveniences that llama.cpp itself leaves to the user. It supports multi-model routing through preset-based configuration with automatic load and unload, interactive Hugging Face search and download with resume support, and a terminal UI with tables and live search for browsing installed models.
A GGUF inspector shows a model's metadata, architecture and sampling parameters, and can automatically sync a preset's sampling settings to match what the GGUF file itself recommends. An idle-model watchdog unloads models after a configurable timeout, and the tool parses server logs to track VRAM usage per loaded model, along with per-model settings for context size, GPU layers and flash attention.
It installs with pipx or uv as the llb command and requires llama.cpp's llama-server to already be installed and on the system PATH.