I built a CLI to stop local AI models from eating my disk twice — lmm

Every tool (LM Studio, Ollama, llama.cpp) downloads models to its own directory. Same 8GB model × 3 tools = 24GB wasted.

lmm uses HF Cache as a single store and symlinks models to each tool. Download once, use everywhere.

https://reddit.com/link/1t934vi/video/zpx3dakzca0h1/player

• brew tap holotherapper/tap && brew install lmm
• Interactive search + install from HF
• Supports MLX, GGUF, safetensors
• Works with LM Studio, llama.cpp, Jan, ComfyUI, etc.
• Adopt existing HF Cache models without re-downloading

GitHub: https://github.com/holotherapper/lmm

Built in Rust, Apple Silicon only. Feedback welcome.