TL;DR
A CLI tool that automatically recommends the best-performing local LLM for your hardware based on real benchmark performance.
Key features
Hardware auto-detection: Reads GPU/CPU/RAM to filter executable models.
Real-time ranking: Merges latest benchmarks from HuggingFace (LiveBench, Artificial Analysis, Aider, Chatbot Arena ELO, etc.) to rank models.
Recency weighting: Older leaderboards are downweighted based on model lineage to prevent outdated models from outperforming newer ones.
Simulation: Use --gpu option to get recommendations assuming a specific GPU.
Additional workflows: Supports upgrade comparison (upgrade), finding required GPU for a model (plan), running a model (run), outputting Python code snippets (snippet), etc.
When to use it
When you want to run a local LLM but don't know which model to choose.
When you want to check which models will become faster before upgrading your GPU.
When you want to find a model with good actual performance, not just one that fits in VRAM.