core
Vector database
The infrastructure that stores billions of document embeddings so engines can run nearest-neighbor search in milliseconds.
Definition
A vector database (Pinecone, Weaviate, pgvector, Vespa, Qdrant) stores high-dimensional embeddings and supports approximate-nearest-neighbor search at scale. Engines like Perplexity maintain vector databases of indexed web content; a query is embedded then compared against the index. ANN algorithms (HNSW, IVF) trade exactness for speed — searching a billion vectors in <50ms.
Example
Perplexity's index contains embeddings of ~10B web pages in a vector database. Each query takes ~40ms to retrieve top-50 candidates from the database.How to optimize
Get crawled and indexed. A page that GPTBot/ClaudeBot/PerplexityBot can fetch + parse cleanly ends up in the vector database; a page they can't fetch never enters retrieval.