• Allows you to query by an entity ID, a “feature group,” and a timestamp
    • Here “feature group” refers to a collection of primitives
    • A -dimensional embedding is a group of floating point features
  • Behind the scenes, these “feature groups” are usually stored in two environments:
    • A database that stores the latest version of the “feature group” for low latency
    • Static storage with previous versions, by timestamp
  • At query time, by default you’ll get the latest copy
  • But if you supply a timestamp, you’ll get the most recent copy before that time