What is a vector database?
A vector database stores embeddings, the numeric vectors that represent the meaning of text, images or other data, and finds the stored vectors nearest to a query vector, usually with an approximate nearest neighbour index.
Checked
Semantic search, retrieval-augmented generation, recommendations and agent memory all need the same operation: given a query vector, find the closest stored vectors fast, often with filters such as tenant or date. A vector database is built around that operation.
The category has blurred. In 2026 vector databases added keyword search, search engines and Postgres added vectors, and object storage became a common backend. The question is less which vector database to buy and more where vectors should live.
What a vector database does
Approximate nearest neighbour search
Indexes such as HNSW, IVF or DiskANN-style graphs find close vectors without comparing every one.
Metadata filtering
Restricting results by fields such as customer, permission or date, which is where many real-world recall problems appear.
Hybrid search
BM25 keyword ranking alongside vectors, now shipped by most engines.
Compression
Quantization to cut memory, with rescoring to recover accuracy.
The shapes it takes in 2026
Dedicated vector databases
Pinecone, Weaviate, Qdrant, Milvus and the open-source Chroma, built around vector search first.
Postgres extensions
pgvector adds a vector type with HNSW and IVFFlat indexes, so vectors sit next to relational data with transactions and joins.
Search engines
Elasticsearch, OpenSearch and Vespa combine mature keyword search with vectors and ranking.
Object-storage engines
turbopuffer, Amazon S3 Vectors (generally available on 2025-12-02) and Milvus 3.0's lake-native design keep indexes on object storage with a cache.
How the market moved
Chroma announced on 2026-10-23 that it is joining Replit and winding down Chroma Cloud over 60 days, while open-source Chroma continues under Apache 2.0. Pinecone made full-text search generally available on 2026-09-09, and Qdrant added BM25 in 1.20. Model labs also sell managed retrieval, such as OpenAI's file search tool and the Gemini API File Search tool, which compete with a standalone store for smaller corpora.
How Motiontic's engineers choose one
Motiontic's engineers start from where the client's data already lives and what the eval set needs: corpus size, update rate, tenancy, filtering, keyword needs and the team that will run it. If Postgres already holds the data and the corpus fits, pgvector is often enough. A dedicated or object-storage engine earns its place when scale, tenancy or cost demands it.
Where vectors can live, compared
| Fits | Watch for | |
|---|---|---|
| Dedicated vector database | Large or fast-growing corpora, heavy query load | Another system to sync and pay for |
| Postgres extension | Data already in Postgres, moderate scale | Index memory and filtered-search tuning |
| Search engine | Keyword-heavy search with ranking needs | Operational weight and licence terms |
| Object-storage engine | Very large, cold or per-tenant corpora | Slower cold queries, feature limits |
| Managed file search from a lab | Small to medium document sets, fast start | Less control over chunking and ranking |
Common questions
Does RAG need a vector database?
Not always. A knowledge base under about 200,000 tokens can go straight into the prompt, per Anthropic's Contextual Retrieval post, and many teams use pgvector in an existing Postgres database rather than a separate system.
Is pgvector a vector database?
pgvector is a Postgres extension that adds a vector type and HNSW and IVFFlat indexes. It turns Postgres into a capable vector store, without a separate database to run.
What is happening to Chroma Cloud?
Chroma announced on 2026-10-23 that it is joining Replit and winding down Chroma Cloud over the following 60 days. Open-source Chroma continues under the Apache 2.0 licence.
What is the difference between a vector database and a search engine?
The line has mostly gone. Vector databases now ship BM25 keyword search, and search engines such as Elasticsearch and OpenSearch ship vector indexes. The differences are in scale model, operations, licence and ranking flexibility.
Related
HNSW
The layered graph index behind most vector search.
Embeddings
Meaning turned into vectors, so similar things can be found by distance.
pgvector vs Pinecone
A Postgres extension compared with a managed serverless vector database.
pgvector vs Qdrant
Vectors inside Postgres compared with an open-source Rust vector engine.
AI product development
New products and major AI features, built from scratch alongside the client's team.
GraphRAG
RAG over a model-built knowledge graph, for questions about a whole corpus.
Sources
- Chroma, Chroma joins Replit (2026-10-23)
- Anthropic, Introducing Contextual Retrieval (2024-09-19)
- Pinecone, Full-text search generally available (2026-09-09)
- Qdrant, Qdrant 1.20 release (2026-10-22)
- AWS, Amazon S3 Vectors generally available (2025-12-02)
- Zilliz, Milvus 3.0 lake-native vector database (2026-07-16)
- pgvector README on GitHub
- OpenAI, File search tool guide
- Google, File Search tool in the Gemini API (2025-11-06)
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.