Vector Database
A database that stores meaning, not just text — finding similar content by concept rather than keyword.
A vector database stores data as high-dimensional numerical vectors (embeddings) and enables fast similarity search — finding items that are semantically similar even if they share no keywords.
How it works:
- Pass text (or images, audio) through an embedding model — it outputs a list of ~1,500 numbers representing meaning
- Store those numbers in the vector DB alongside the original content
- At query time, embed the user's question, then find stored vectors with the smallest geometric distance
The result: search that understands meaning, not just string matching. "How do I cancel my subscription?" will surface relevant results even if no document uses the word "cancel."
Popular vector databases:
- Pinecone — managed, production-grade
- Weaviate — open-source, hybrid search
- Qdrant — open-source, fast, Rust-based
- pgvector — PostgreSQL extension (add vectors to your existing DB)
- Chroma — lightweight, great for prototyping
Primary use case: The retrieval layer in RAG (Retrieval-Augmented Generation) pipelines. The vector DB finds relevant document chunks; the LLM reads those chunks and generates an answer.
In plain terms
A librarian who organizes books not by title or author, but by what they're *about* — so asking for 'something like Sapiens but more technical' returns the right shelf.