Vector Database

What is a vector database?

A regular database excels at exact keyword matches: you ask for "apple" and it returns records containing the word "apple." But AI needs to search by meaning — say "a big red fruit" and it should still think of an apple. A vector database is built specifically for finding things by similarity of meaning.

What does it actually store?

Vectors (embeddings)
AI turns a piece of text, an image or a clip of audio into a list of numbers called a vector. Content with similar meaning ends up close together in vector space.
Similarity search
On a query, the database doesn't compare words — it measures how close your query vector is to the stored vectors. Closer means more relevant.

What problems does it solve?

Semantic search
Find the same result even when you phrase it differently, no keyword-matching needed.
RAG (retrieval-augmented generation)
Before a large model answers, it pulls the most relevant snippets from a vector store, which cuts down on hallucination.
Recommendations and dedup
"You might like" and "find similar images or articles" all run on vector similarity.

How is it different from a normal database?

Traditional databases optimize equality and range queries; vector databases optimize nearest-neighbor search — fast hunting for the closest items in a sea of high-dimensional vectors. That needs special indexes like HNSW to stay fast, otherwise a few million vectors would take forever.

Bottom line: a vector database is the search engine underneath the AI's ability to find people, images and documents by meaning.