Question 1

What is a vector database?

Accepted Answer

A vector database is a storage system optimised for high-dimensional embedding vectors and nearest-neighbour similarity search. It powers retrieval-augmented generation, semantic search, and recommendation systems by indexing millions to billions of vectors and returning the most similar results in sub-second query time.

Question 2

How much do vector databases cost?

Accepted Answer

Small applications run £40–250 per month on managed serverless tiers. Mid-scale production lands between £400–4000 per month. Large-corpus deployments reach £8000 to six figures monthly on managed plans, with self-hosted alternatives shifting the cost into platform-engineering rather than subscription fees.

Question 3

Which vector database is best for RAG?

Accepted Answer

Pick by native hybrid-search support, p99 latency at your real concurrency and filter complexity, and operational fit with your existing data stack. RAG retrieval quality depends more on chunking strategy, embedding model choice, and retrieval ranking logic than on raw vector store benchmark scores.

Question 4

Vector database versus traditional search engine — what is the difference?

Accepted Answer

Traditional search excels at keyword matching and structured queries with deterministic ranking. Vector search wins on semantic similarity, fuzzy intent matching, and multimodal retrieval. Most production retrieval systems combine both — keyword filters narrow the candidate set, then vector similarity ranks within it.

Question 5

Managed versus self-hosted vector database — which should I choose?

Accepted Answer

Managed services trade higher per-vector cost for zero operational overhead, automatic scaling, and managed upgrades. Self-hosted wins on extreme volume, strict data residency, and predictable cost ceilings — but requires platform engineering capacity. Total cost of ownership rarely matches either side's list price on its own.

Question 6

Can I use a general-purpose database as a vector store?

Accepted Answer

Several relational and document databases now support vector extensions with approximate nearest-neighbour indexing. These are reasonable for small corpora and lower query throughput. At tens of millions of vectors, high query concurrency, or complex hybrid-search requirements, purpose-built vector stores typically outperform extensions significantly on both latency and recall.

Vector Databases marketplace

Supabase

Snowflake

Databricks

Vectara GenAI Platform