1. What is the difference between a vector database and a regular database?

A regular database stores structured data and retrieves records using exact queries like filters or keyword matches. A vector database stores high-dimensional numerical vectors and retrieves results based on mathematical similarity. The two serve different purposes, and many production systems use both together.

2. Can a vector database replace Elasticsearch for search?

Not entirely. Elasticsearch is built for full-text search, aggregations, and log analysis. A vector database handles semantic similarity. In 2026, most teams use hybrid search, which combines both. You can run this natively in Elasticsearch with its built-in dense vector support or with tools like Weaviate and Pinecone.

3. How much does it cost to run a vector database in production?

Costs vary by provider and scale. Pinecone charges based on pod type and the number of vectors stored. Qdrant and Weaviate have free open-source tiers with paid cloud options. Self-hosting reduces vendor costs but adds infrastructure management. For most mid-scale applications.

4. What is the maximum number of vectors a vector database can handle?

Modern vector databases handle billions of vectors with the right setup. Milvus and Pinecone are built for large-scale enterprise use. For most applications, a few million vectors is well within the comfortable range of tools like Qdrant or Weaviate without special configuration.

5. Do I need a GPU to run a vector database?

No. The vector database itself does not require a GPU. GPUs are needed for generating embeddings using models like OpenAI's or a local model from HuggingFace. Once vectors are generated, the database stores and retrieves them using CPU-based ANN algorithms.

6. What is HNSW and why is it used in vector databases?

HNSW stands for Hierarchical Navigable Small World. It is a graph-based indexing algorithm that organizes vectors into a multi-layer structure for fast approximate nearest neighbor search. It delivers high recall with low query latency, which is why it is the default index type in Qdrant, Weaviate, Milvus, and most other modern vector databases.

7. How is a vector database used in a RAG application?

In a RAG setup, documents are split into chunks, converted into embeddings, and stored in a vector database. When a user asks a question, the query is embedded and the database returns the most relevant chunks. Those chunks are passed to an LLM as context, allowing it to generate answers grounded in your data rather than just its training knowledge.

8. Is pgvector good enough for production, or do I need a dedicated vector database?

pgvector works well for smaller datasets and teams already running Postgres. It handles a few million vectors comfortably. For applications requiring high-throughput queries, advanced filtering, or support for billions of vectors, a dedicated vector database like Qdrant or Pinecone will serve better.

9. What is the role of metadata filtering in a vector database?

Metadata filtering lets you narrow vector search results by structured attributes. For example, you might search for "customer reviews about delivery" and filter only reviews from a specific product category or date range. Without filtering, you get the most semantically similar results from your entire dataset, which may include irrelevant data.

10. How do I evaluate the quality of a vector database's search results?

Use standard information retrieval metrics. Recall@k measures how many of the true nearest neighbors appear in the top k results. Mean Reciprocal Rank (MRR) and NDCG evaluate ranking quality. Benchmarks like ANN-Benchmarks compare different databases on speed and recall trade-offs across multiple dataset types.

11. Can vector databases store text and images together?

Yes, if you use a multimodal embedding model. Models like CLIP encode both text and images into the same vector space, so you can store and query across modalities. A text query like "red sports car" can retrieve matching images. Weaviate's multi2vec module and Marqo support this natively.