Catalog concepts

A collection is the unit you open in Vector Catalog. The other words on this page are things inside it, or records that point at it.

Pieces

A document is a file the collection holds. An extracted field is a value taken from that file. Global fields repeat across documents. Local fields belong to one document. A contract is the schema and roles those fields follow. A Smart Bite is a stored passage with its own id. A vector is the embedding of a bite. An index is what makes those vectors searchable. A result is a stored answer or extract. A publication is a catalog record of a result. Creating a result or a publication is described in the workflow handbook, not here.

What has to be true first

Catalog browsing needs a stored workspace. A non-retaining workspace does not keep this library. Reading metadata and reading file content are different permissions. Running extraction or indexing needs processing authority. Deleting a collection needs collection management. A role that can open the page cannot be assumed to do all of those.

Stages that are not the same

Upload stores a file. Discovery proposes a structure. Approval accepts that proposal. Extraction writes field values. Materialization makes those values available to browse. Indexing writes vectors. Finishing one stage does not finish the next. A saved approval is not an indexed collection.