About
WhatsMyLLM answers one question: which language model wrote this reply? It does so with a method that is public, a bank that is versioned, and numbers that were measured under the site's own conditions and published with their limits.
Why it exists
Model labels are cheap. A relay, a wrapper, or a product can call its model whatever it likes, and the text it returns carries no signature. What it does carry is habit: asked for random numbers, every model leans in its own way, and that lean is stable enough to recognise. This site turns that observation into a tool anyone can run against any chat window, with nothing sent anywhere.
What it is built on
The fingerprint method and the scoring core are from ModelTrace, MIT licensed and vendored unchanged. This project adds the input gates that refuse counted or sorted lists, the open-set verdict that can say "no model here fits", the bilingual site, and its own collected fingerprints. Every claim on the methodology page traces to a file in the project's evidence directory.
What it is not for
A verdict here is statistical evidence about number-choice habits. It is not a certificate, not a legal finding, and not a basis for a commercial dispute. Treat a single confident result the way you would treat a single witness: useful, and worth checking twice.