What it is: Atlas Vector Search now supports $vectorSearch queries for vector embeddings stored within arrays of subdocuments. Developers can query nested content— such as chunks within a document, reviews within a product, or messages within a conversation — and retrieve the parent document ranked by the best or average match among its children.

Who it's for: Developers building RAG applications and semantic search systems on top of hierarchically structured data who need to run vector similarity search against nested content without restructuring their data model.

Why it matters: Applications can now perform vector similarity search directly against nested document structures, preserving MongoDB's embedded document pattern without duplicating parent metadata or flattening hierarchical collections.

How to get started: Add a nestedRoot field to a vector search index definition pointing to the target array, then query with $vectorSearch using the nested vector path. Review the documentation below for a step-by-step example.