Learning from data
in real time

A pre-trained foundation model for event streams.

It makes predictions directly from a stream of events, such as purchases, interactions in a temporal graph, machine logs, or financial transactions, with the stream's own history as context. At any event it predicts any missing part: which entity comes next, which category a feature takes, what value a number holds. One model covers every prediction task, with nothing to train per dataset or per task.