What's My Model?

Which local (GGUF) model actually fits your machine — no backend, drop into any page.

Pick a model to see it here.

Detection is seeded to a 16 GB GPU for the demo — edit the VRAM/RAM fields for your machine. Browsers can't read VRAM, so a real host injects an exact probe.

How it works

Pure frontendThe fit engine + widget are dependency-free JavaScript. It carries no server — you drop <whats-my-model> into any page.

Hosts inject hardwareA HardwareProvider supplies { gpu, ram }: a browser best-effort default, manual entry, or (on desktop) an exact native probe.

Real catalogA bundled Hugging Face snapshot renders instantly; a live provider folds in fresh results, cached in IndexedDB.

Honest fitWeights + KV cache (context-aware) + buffers vs your VRAM — so "32K fits, 128K doesn't", and quantizing the KV cache changes the answer.