cloudflare-vectorize

Semantic search uses vector embeddings to find meaning-based matches instead of keyword matches. Cloudflare Vectorize indexes and queries embeddings, enabling you to search documents, products, or content by semantic similarity rather than exact text.

Yes. Vectorize supports retrieval-augmented generation by storing document embeddings with metadata, then retrieving relevant chunks during queries. This powers knowledge-base Q&A, product recommendations, and content discovery workflows.

Vectorize supports up to 10 metadata indexes per index, enabling precise filtering on query results. You can filter vectors by metadata fields during retrieval, combine metadata with namespace isolation for multi-tenant workflows, and upsert vectors with attached metadata.

Extract text and images from PDFs using pypdf or pdfplumber, convert to embeddings via Workers AI or OpenAI, then insert vectors with metadata into a Vectorize index. This prepares documents for semantic retrieval and Q&A.

Yes. Vectorize integrates with Workers AI's bge-base-en-v1.5 model and OpenAI embeddings. Create a 768-dimension index for Workers AI, insert vectors, and query using the same embedding model for consistent semantic matching.

Vectorize requires fixed embedding dimensions at index creation, metadata indexes must be created before vector insertion, and indexes support up to 10 metadata filters. Namespace and metadata filtering enable multi-tenant isolation but require careful schema planning upfront.