Identity, done carefully.
ZenithSearch started as a question: can face matching be both genuinely fast and genuinely respectful of the people in the photos? Everything we've built since answers that question.
Mission
Give teams a search engine for faces that is as accurate as it is accountable — clear about what it found, and just as clear about what it didn't.
We work with security teams, platforms, and investigators who need a dependable answer, not a plausible-looking guess.
Technology
Three components, one pipeline.
YuNet
A lightweight, real-time face detector that locates faces in a frame before anything else runs.
SFace
Turns each detected face into a compact embedding vector that captures its identity.
Faiss
Facebook AI's similarity search library, indexing embeddings so match returns in milliseconds.
Privacy
Your search photo is not our product.
Photos submitted for a search are processed in memory to compute a match and are not written to disk or added to the index unless you explicitly enroll them.
Confidence scores are shown as returned by the model — we don't round a low-confidence match up to look more certain than it is.