Glossary
Vector Search
A retrieval method that ranks documents by semantic similarity — typically cosine distance between embedding vectors — rather than keyword overlap.
Vector search retrieves documents based on meaning rather than literal text match. Each document chunk is represented as a high-dimensional vector — its embedding — produced by an embedding model. A query is embedded the same way, and the database returns the chunks whose vectors are closest, usually by cosine similarity or inner product.
Vector search is what makes RAG work on natural-language queries. A user asks "how do we handle audit log retention?" and the system finds chunks discussing log retention, audit trails, and CloudTrail policies even when none of them contain the exact phrase.
In regulated industries, vector search introduces concerns that keyword search does not. Embeddings can leak information across tenants if the index is shared. Embedding models trained on the open web may carry license or PII concerns. And a vector index is not a substitute for an access control layer — you still need to filter results by the user's permissions before returning them, otherwise vector search will happily surface documents the user is not authorized to see.
Most production systems combine vector search with metadata filters and a re-ranking stage. The vector recall is wide; the filter is strict; the re-ranker improves precision.
Related terms
RAG (Retrieval-Augmented Generation)
An LLM pattern that retrieves relevant documents at query time and feeds them to the model as context, instead of relying on the model's training data alone.
Embeddings
Numerical vector representations of text (or images) that place semantically similar inputs near each other in vector space.