Vector Database
A store for embeddings that can quickly find the vectors most similar to a query among millions, the retrieval engine behind semantic search and RAG.
Also vector search · approximate nearest neighbor · ANN · HNSW · FAISS · semantic search · embedding index
A library where books are shelved by meaning, not title. Ask for something, and the librarian walks straight to the shelf of books most like your question.
Where it fits
+All relationships as a list
Is a kind of / part of
Siblings
- AI Agents and Tool Use
- BERT
- Direct Preference Optimization
- Embedding
- Fine-Tuning
- Google Translate
- GPT
- Hallucination
- In-Context Learning
- Keyboard Autocomplete
- Large Language Model
- LLM Evaluation and Benchmarks
- Perplexity
- Pretraining
- Prompting
- Retrieval-Augmented Generation
- RLHF
- Sampling and Temperature
- Scaling Laws
- Tokenization
- word2vec
Uses / built on
How it works
Embeddings turn text, images or products into vectors where similar meaning means nearby. Finding “the most similar items” is then a nearest-neighbor search. With a few thousand items you can compare against all of them; with hundreds of millions you cannot.
Vector databases build an index that finds near neighbors without checking everything:
- Graph indexes (HNSW): each vector links to some neighbors; a search hops through the graph toward the query, like asking for directions from person to person.
- Clustered indexes: group vectors into buckets (k-means style) and search only the closest buckets, often with compressed vectors.
These are approximate: they occasionally miss a true neighbor, in exchange for speedups of hundreds or thousands of times. On top come practical features: filtering by metadata (only documents from this team), updates, and combining with keyword search. In a RAG system, this is the step that fetches the passages the model then reads.
Everyday use cases
- Case 01
RAG over company documents
Find the passages most relevant to an employee's question.
- Case 02
Similar products and images
"More like this" recommendations from image or product embeddings.
Strengths, weaknesses and when to use it
Strengths
- Millisecond search over millions of items by meaning.
Weaknesses
- Approximate: can miss some true nearest neighbors.
- Results are only as good as the embeddings.
When to use itWhen you need to search large collections by similarity rather than exact keywords.
Compare with
Check yourself
- Q01
Why do vector databases use approximate search?
Show answer
A: Comparing a query with every stored vector is too slow at large scale