GEO / AI search

Vector database

Q: What is a vector database?

A vector database is a database system that stores vector embeddings and searches them quickly by similarity (approximate nearest neighbour, ANN). It is the foundation of modern RAG systems, semantic search and LLM knowledge bases.

Q: Which vector databases are available?

Four categories: (1) specialised (Pinecone, Weaviate, Qdrant, Milvus), (2) extensions (PostgreSQL pgvector, MongoDB Atlas Vector), (3) cloud-native (OpenAI Vector Stores, Google Vertex AI), (4) self-hosted (Chroma, FAISS). Rankmio uses pgvector for seamless integration with its MySQL backend.

Q: Does every RAG system need a vector database?

For production use, yes. A prototype RAG with fewer than 10,000 chunks can run on a simple NumPy search. From 100,000 chunks onwards you need genuine ANN indexes (HNSW, IVF), otherwise the search becomes unusably slow.

Also known as: Vector Database, Vector Store, Embedding index

A vector database (vector store) is a specialised database for high-dimensional vectors — typically embeddings derived from text, images or audio. Its core task is nearest-neighbour search: given a query vector, find the K most similar vectors in the index. For classic table-based databases this would be extremely slow because of the high dimensionality (often 768–3,072) — vector databases use dedicated indexing methods such as HNSW, IVF or Annoy to deliver sub-millisecond searches across millions or billions of vectors.

Common vector databases

pgvector — a PostgreSQL extension, integrated into a standard relational database. The pragmatic standard for medium-sized applications.
Pinecone — a managed service, easy to get started with, commercial.
Weaviate — open source, with built-in modules for hybrid search.
Milvus — open source, highly scalable (billions of vectors).
Qdrant — open source, Rust-based, fast and resource-efficient.
Elasticsearch / OpenSearch with vector search — anyone who already has full-text search in their stack can run vector search alongside it.

Where vector databases sit in the AI pipeline

In a typical RAG architecture: (1) sources are cut into chunks (for example, one per H2 section). (2) An embedding is generated for each chunk and written to the vector database, together with metadata (URL, date, author). (3) When a user asks a question, the question itself is converted into a vector and the K nearest chunks are retrieved from the vector database. (4) These chunks are passed to the LLM, which composes an answer from them and references the original URLs as citations.

Vector databases in Rankmio

Rankmio uses pgvector in a separate PostgreSQL database for the chat agent's RAG pipeline: glossary entries, manual sections, FAQ answers and user project data are fed in as chunks, so that the chat agent can retrieve the most thematically relevant pieces of knowledge for every request — without having to send the full content along with every LLM prompt. The effect: faster, more precise and cheaper answers than without RAG.

Example from practice

Example: A company website with 1,200 knowledge articles builds a RAG-based site search. Without a vector database, the search index would have to rely on classic full text — the user has to know the exact terms. With pgvector and embeddings, the search finds the right article even when the user phrases things quite differently: ”Why is my page loading slowly?” automatically surfaces the PageSpeed article and the Core Web Vitals documentation — even if none of those terms appear in the search text.

Frequently asked questions

What is a vector database?

A vector database is a database system that stores vector embeddings and searches them quickly by similarity (approximate nearest neighbour, ANN). It is the foundation of modern RAG systems, semantic search and LLM knowledge bases.

Which vector databases are available?

Four categories: (1) specialised (Pinecone, Weaviate, Qdrant, Milvus), (2) extensions (PostgreSQL pgvector, MongoDB Atlas Vector), (3) cloud-native (OpenAI Vector Stores, Google Vertex AI), (4) self-hosted (Chroma, FAISS). Rankmio uses pgvector for seamless integration with its MySQL backend.

Does every RAG system need a vector database?

For production use, yes. A prototype RAG with fewer than 10,000 chunks can run on a simple NumPy search. From 100,000 chunks onwards you need genuine ANN indexes (HNSW, IVF), otherwise the search becomes unusably slow.

How does vector search differ from full-text search?

Full text searches for word matches, vectors search for similarity in meaning. Full text finds ”car” only where the word ”car” appears. A vector search also finds ”vehicle”, ”automobile”, ”motor”. Hybrid search (full text plus vectors) is the current state of the art.

Used in Rankmio for

Chat agent with RAG pipeline in Rankmio

Go to the feature →

Last updated: 2026-06-17 · Browse all glossary entries