llmfit answers the only local-model question that matters: what actually runs on your machine.

AlexsJones/llmfit (30,610 stars, +1,066 this week, MIT, Rust) detects your hardware then ranks hundreds of models and providers on quality, speed, fit and context. Integrates with Ollama, llama.cpp, MLX, Docker Model Runner and LM Studio, handling multi-GPU, MoE architectures, dynamic quantization and speed estimation, with an interactive TUI and a JSON output mode. The JSON mode is the part worth wiring into a pipeline that picks models programmatically instead of by hand.