What is a Vector Database?

What Is a Vector Database?

A vector database is a database designed to store vectors and efficiently retrieve the vectors that are most similar to a query vector. In Generative AI, those vectors commonly come from an embedding model.

Why Do We Need One?

Embeddings turn text, images, audio, or other content into vectors. Once you have thousands or millions of vectors, you need a practical way to store them, search them, filter them, update them, and retrieve the best matches.

What Does It Store?

A useful mental model is that one stored item contains more than just a vector. It usually combines the vector with an identifier, metadata, and often the original text or a reference to it.

Vector Database vs. Traditional Database

They are not necessarily competitors. Many production systems use both: a relational or document database for application data and a vector database or vector index for semantic retrieval.

How Similarity Search Works

The application first converts the user's query into an embedding. The vector database then compares that query vector with stored vectors and returns the closest matches according to the selected distance or similarity metric.

What Does “Top-k” Mean?

Most retrieval systems ask for the top-k nearest results. If k = 5, the search returns the five highest-ranked candidates after applying the selected search rules.

A high similarity score does not automatically mean the result is correct. Retrieval quality should be evaluated on your real questions and documents.

Metadata Makes Retrieval More Useful

Similarity is only one dimension of retrieval. Metadata lets you narrow the search before or during vector retrieval.

Indexing: How Search Becomes Fast

If a collection contains a huge number of vectors, comparing a query with every single vector can be expensive. Vector systems therefore use indexing and approximate nearest-neighbor techniques to search efficiently.

Python: A Tiny Vector Store Mental Model

This example is intentionally simple. It shows the data flow without depending on a particular vector database vendor.

documents = [
    {"id": "doc-1", "text": "Reset your password from account settings.", "vector": [0.90, 0.10]},
    {"id": "doc-2", "text": "Track your order from the shipping page.", "vector": [0.20, 0.85]},
    {"id": "doc-3", "text": "Update your billing address in your profile.", "vector": [0.65, 0.35]},
]

query_vector = [0.82, 0.18]

# In a real vector database, the index performs this search efficiently.
# Here we keep the example conceptual and focus on the stored structure.
for document in documents:
    print(document["id"], document["text"], document["vector"])

print("Query vector:", query_vector)

Python: Calculate a Simple Similarity Ranking

Before using a database, it helps to understand the operation it accelerates. This small example ranks two-dimensional vectors with cosine similarity.

import numpy as np

def cosine_similarity(a, b):
    a = np.array(a, dtype=float)
    b = np.array(b, dtype=float)
    denominator = np.linalg.norm(a) * np.linalg.norm(b)
    return float(np.dot(a, b) / denominator)

query = [0.82, 0.18]
candidates = {
    "password reset": [0.90, 0.10],
    "order tracking": [0.20, 0.85],
    "billing address": [0.65, 0.35],
}

ranked = sorted(
    ((name, cosine_similarity(query, vector))
     for name, vector in candidates.items()),
    key=lambda item: item[1],
    reverse=True,
)

for name, score in ranked:
    print(f"{score:.3f}  {name}")

Vector Database in a RAG System

RAG uses retrieval to bring relevant external information into the model's context. A vector database is commonly used as the retrieval layer for that workflow.

Vector Database vs. Vector Index

These terms are related but not identical. A vector index is the search structure; a vector database is a broader storage and retrieval system that can manage vectors, metadata, persistence, and APIs.

Common Vector Database Operations

When Should You Use a Vector Database?

What a Vector Database Does Not Do

How to Choose One Later

You do not need to choose a vendor in this lesson. When you reach that decision, compare the system against your actual workload.