Intermediate12 min

Vector Database Selection

Comparing Pinecone, Weaviate, pgvector, Qdrant, and Chroma for production RAG. Features, pricing, scaling characteristics, and when to use each.

Quick Reference

→pgvector: simplest ops — runs in your existing Postgres, perfect for <1M vectors
→Pinecone: fully managed serverless, scales to billions, zero ops burden
→Qdrant: high-performance filtering + vector search, open-source with managed option
→Weaviate: hybrid search built-in, GraphQL API, good for complex query patterns
→Chroma: local development and prototyping, lightweight, embeds in your app
→For most teams: start with pgvector, migrate to a dedicated vector DB when you hit scale limits

pgvector — Vector Search in Postgres

pgvector adds vector similarity search to PostgreSQL. If your application already uses Postgres, pgvector is the lowest-friction path to production RAG. No new infrastructure, no new ops skills, no new billing. You get ACID transactions, familiar SQL, and your vectors live alongside your relational data. The tradeoff is performance at scale — pgvector is slower than purpose-built vector databases beyond ~1M vectors without careful index tuning.

RAG pipeline with pgvector via LangChain

Index tuning matters

pgvector supports HNSW and IVFFlat indexes. For <100K vectors, a sequential scan is fast enough. For 100K-1M vectors, create an HNSW index: CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops) WITH (m = 16, ef_construction = 64). This brings query time from seconds to milliseconds.

Vectors	Without Index	With HNSW Index	Recall@10
10K	5ms	2ms	100%
100K	50ms	5ms	99.5%
1M	500ms	10ms	98%
10M	5s+	20ms	95%

Pinecone — Fully Managed Serverless

Pinecone is a fully managed vector database with a serverless pricing model. You don't manage infrastructure, indexes, or scaling — Pinecone handles all of it. This makes it the fastest path to production for teams that don't want to manage vector DB infrastructure. The serverless tier charges per query and storage, making it cost-effective for low-to-medium query volumes.

Qdrant — High-Performance Filtering

Qdrant is an open-source vector database written in Rust, optimized for high-performance vector search with rich filtering. Its key differentiator is the ability to apply complex metadata filters during (not after) the vector search, which is critical when you need to scope searches by tenant, date range, or category. Qdrant offers both self-hosted and managed cloud options.

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.