Introducing Sparse and Hybrid Indexes

Upstash Vector has introduced support for sparse and hybrid indexes, enhancing its capability beyond the previously solely supported dense indexes for semantic similarity searches. Dense indexes create a comprehensive map of data in a high-dimensional vector space, capturing overall similarity but potentially struggling with out-of-domain queries or those with rare words. In contrast, sparse indexes focus on efficiency by storing only significant non-zero values, making them ideal for data with variable-length features such as text documents or product catalogs. Hybrid indexes combine the strengths of both dense and sparse indexing, catering to data with both semantic meaning and specific keywords. These new index types, available through Upstash's console, allow users to create and query with either custom or Upstash-hosted embedding models, offering flexibility and enhanced capability for AI and RAG workflows.