Choosing a local model usually means downloading several large files and testing them blindly until one fits the machine.
Decision
I combined hardware detection, a cached Ollama catalog and an explicit heuristic that classifies each model as Good, Tight or No.
Ollama-Fit
Challenge
Normalizing VRAM and unified memory across platforms while maintaining the catalog without an official public API.
Result
Before downloading anything you know which models fit your machine: each one is rated Good, Tight or No, presented as an estimate rather than a benchmark.
