A collection is a named group of vectors that share the same schema and index configuration within a vector database. It is the rough equivalent of a table in a relational database: a logical container that holds related records and defines how they are structured and indexed.
Organising data into collections keeps different kinds of content separate and independently configurable. You might keep product embeddings in one collection, support-ticket embeddings in another, and documentation in a third — each with its own vector dimension, distance metric, and metadata fields. Queries run against a specific collection, so results never bleed across unrelated data.
Collections are also a common building block for multi-tenancy. In the collection-per-tenant pattern, each customer’s data lives in its own collection, providing clean isolation while still sharing the underlying database instance. The exact terminology varies between systems — some call the same concept a class, an index, or a namespace — but the role is the same.