Tools & retrieval
What is Vector Database?
A vector database stores embeddings and returns similar items.
Similarity is not truth. Retrieved neighbors still need checking.
Why it matters
Vector databases store embeddings and return the most similar items for a query. They are the storage layer that makes semantic search and RAG possible at scale.
A vector database is not a knowledge base by itself — it needs to be loaded with well-chunked, well-embedded documents, and the retrieval results need to be validated before use.
Choosing a vector database
Key factors are: query latency, filtering capabilities (metadata filters alongside vector search), scale (millions vs. billions of vectors), and hosting model (managed vs. self-hosted). For most agent applications, a managed vector database with metadata filtering is sufficient.
Key takeaways
- 1Vector databases store embeddings for similarity search.
- 2Similarity is not truth — results need validation.
- 3Metadata filtering alongside vector search improves precision.
Common mistakes
- ✕Treating vector search results as authoritative without verification.
- ✕Indexing documents without metadata, losing filtering capability.
Related terms
Concept neighborhood
Terms linked from Vector Database in the glossary graph.
- Vector Database
- Embeddings
- Retrieval
- RAG