← All explainers
Learn
Searching video in plain language
A sentence is read into objects, attributes, places and times. Only what the parser is sure of becomes a filter; the rest ranks results, and every result says why it matched.
From a sentence to a query
Query understanding turns free text into entities, attributes, relations and constraints, with no preset categories to choose from.
Several signals, one ranking
Dense embeddings, structured predicates and OCR are fused, a vision-language model reranks the candidates, and frames are grouped into occurrences.
Scope before results
Which cameras and hours a search may read is decided by policy before anything is read, and what was excluded is disclosed with the results.