DocType is which kind of knowledge matched: kb-page (a wiki page), kb-memory (a unit of agent memory) or kb-source (a document a connector ingested). Those three are the whole indexed set, and searchIn.DocTypes filters on them.
Name is the document's name in the framework store — the id to read or open it with. Unique per (org, doctype), so it identifies the document with DocType and not alone.
Project is the project scope the document was saved under. Absent for a document saved with none, which is also why a project-scoped query cannot reach it.
Provider is the connector that ingested the document — github, slack, google or notion. Absent for a page or memory written in the product, which came from no connector.
Score is the cosine similarity between the query's embedding and the document's, from -1 to 1, higher being closer — the collection is created with Cosine distance. Hits arrive ordered by it, descending. There is no absolute cutoff: what counts as a good score moves with the query and the embedding model, so compare scores within one response and not across queries.
URL is the document's link back into the app it was ingested from. Absent when the indexed payload carries none, which is the normal case for pages and memories.