no local AI required
Read the box, if you want.
You do not need Ollama, llama.cpp, or an agent. A readout is GPU / VRAM / RAM / OS — not a score, not a certificate, not a silent scan. Nothing leaves the machine unless you paste it somewhere.
Fastest
Type the card on the recipe page if you want an estimate. “4090” is enough. The arena is the board.
Optional local readout
There is no silent browser scan. On the box:
mycolo-measure bench --model qwen3.8:27b --handle you --submit
Hostname and username never leave. GPU class + engine tok/s do, if you pass --submit.
Then what
- Match a recipe (primary / stretch / don’t).
- If you later run a model, sign a receipt with your agent. MyColo asks. We do not certify.
- Do not upload a full dump that fingerprints a one-of-one desk. A profile id is enough.