cong-cu Intermediate

What is a Vector Database?

A type of database specialized in storing and quickly searching vector embeddings — the foundation for RAG and semantic search.

Updated: May 5, 2026 · 2 min read

A Vector Database is a database type optimized for storing and searching millions of embedding vectors in just a few milliseconds.

Why does this need its own kind of DB?

Traditional databases (PostgreSQL, MySQL) search by exact match (WHERE name = 'X'). Vector DBs search by similarity (“which vector is most similar to this one?”).

If you have 10 million embedded documents, finding the nearest vector with brute force takes seconds. Vector DBs use ANN (Approximate Nearest Neighbor) algorithms like HNSW and IVF to bring it down to a few ms — at the cost of 1-2% accuracy.

Popular vector databases (2026)

Cloud / managed

Name	Highlights	Starting price
Pinecone	Easiest to use, scales well	Small free tier, $70/month+
Weaviate Cloud	Open source + cloud	Free tier, $25+
Qdrant Cloud	Fast, Rust core	Free tier, $25+
Turbopuffer	New, cheap, serverless	$0.10/GB/month

Self-host

Name	Highlights
Qdrant	Fast, easy to deploy
Weaviate	Has a GraphQL API
Milvus	Large scale, by Zilliz
Chroma	Lightweight, great for prototypes
pgvector	Postgres extension — no new DB needed!

When you DON’T need a Vector DB

< 100k vectors → use pgvector or even just a file
You need absolute search precision → brute force is slow but correct

Typical workflow with RAG

1. Ingest:  Documents → split into chunks → embed → store in vector DB
2. Query:   User asks → embed the question → query top-k chunks
3. Augment: Inject the chunks into the prompt → LLM answers

How to choose

Latency matters → pick Qdrant, Turbopuffer
Already on Postgres → pgvector, no extra DB needed
Need complex filters (geo, time, tag) → Weaviate
Most generous free tier right now → Qdrant Cloud