benbaker76/llmfit ? reverse-engineered prompt

Build me a terminal app called llmfit that helps me figure out which local LLM models will actually run well on my computer.

I want it to detect my RAM, CPU, GPU, and VRAM, then rank a big catalog of models by how well they fit, how fast they should run, the quality I can expect, and how much context they support. Make the default experience an interactive TUI, but also include simple command line output for scripts and automation, like getting a ranked list, looking up one model, and printing JSON.

It should support common local runtimes like Ollama, llama.cpp, MLX, Docker Model Runner, and LM Studio, and it should handle things like multi GPU setups and different quantization choices. If possible, include a way to benchmark real speed on my machine and save the results locally first. Look up current docs online if you need to, and make the output feel polished and easy to use.