Vector Database
Definition
A vector database is a data system designed to store, index and retrieve vector embeddings efficiently, often using approximate nearest-neighbour search or related indexing methods.
Why it matters
Vector databases make large-scale semantic retrieval practical by allowing a system to search across many embeddings without comparing every query vector with every stored vector exhaustively.
How it works
Storage
- Embeddings are stored together with identifiers and often metadata linking them back to documents, passages, products or entities.
Vector indexing
- Specialised indexes organise embeddings for efficient similarity search.
- Different implementations use different indexing algorithms and trade-offs.
Similarity retrieval
- A query embedding is compared against the index to return nearby or high-similarity candidates.
- Metadata filters can also narrow the search space where supported.
How Netsleek uses the term
Netsleek uses Vector Database as the infrastructure node connecting Vector Embeddings, Vector Search and Dense Retrieval. It is treated as one possible implementation layer rather than a requirement for every semantic retrieval system.
Comparisons
- Vector Database vs Vector Search: The database stores and indexes embeddings. Vector Search is the retrieval operation performed over those representations.
- Vector Database vs Knowledge Base: A knowledge base organises information for knowledge access. A vector database specifically manages vector representations and similarity retrieval.
- Vector Database vs Inverted Index: Vector databases index numerical embeddings. Inverted indexes map terms to documents for lexical retrieval.
How Vector Database Connects Across AI Systems
Vector Database does not operate in isolation. It connects to other established concepts across the AI search and retrieval process:
- Stores embeddings: Vector Embeddings are the primary searchable representations.
- Supports dense retrieval: Dense Retrieval can query vector indexes for semantically related candidates.
- Can support RAG: RAG systems commonly use vector retrieval over document or passage embeddings.
- Complements lexical indexes: Hybrid Search can combine vector and lexical retrieval infrastructure.
Related glossary concepts
- Vector Embedding
- Vector Search
- Dense Retrieval
- Embedding Models
- Retrieval-Augmented Generation (RAG)
- Hybrid Search
- Inverted Index
- Knowledge Base
Common misinterpretations
- A vector database is not the only way to implement vector search.
- Storing embeddings does not guarantee useful retrieval quality.
- Vector databases do not replace knowledge graphs, lexical indexes or structured databases in every task.
Summary
A Vector Database stores and indexes embeddings for efficient similarity retrieval. It is a common infrastructure component for dense retrieval, semantic search and retrieval-augmented generation.