whichllm

Why we included this project

Choosing a local LLM usually comes down to one boring question: will the model I want actually run on this machine? whichllm answers that with a single command. It checks your GPU, RAM, and CPU, pulls candidate text-generation models from HuggingFace, and ranks them by what your hardware can realistically handle instead of by parameter count. The ranking blends recency-aware results from leaderboards like LiveBench and Aider with runtime-fit checks, so you get a useful sense of tokens per second and memory use before downloading something large. You can even simulate a future GPU to plan a hardware purchase. If you run open-weight models locally with Ollama, llama.cpp, or LM Studio, this saves you from testing models one by one to find one that works.