Vector Stores for RAG

What to Compare

Choose a store by filters, ANN index, persistence, operational load and cost. Start simple; move to a heavier engine only when data or traffic demands it.

ChromaDB

Open source, embedded by default, documents-plus-metadata API. Ideal for prototypes and small applications where a tiny footprint beats raw performance.

pgvector

A PostgreSQL extension that puts vectors next to the rows they describe. Perfect when your data already lives in Postgres and you want joins, transactions and familiar SQL.

Qdrant and Dedicated Engines

Dedicated vector engines such as Qdrant and Milvus bring payload filters, sharding, replication and REST/gRPC APIs. They scale past what embedded stores handle and fit production RAG.

Quick Decision Guide

• Prototype today - ChromaDB or SQLite-based tools.
• Already on Postgres - pgvector.
• Large collections, filters, HA - Qdrant-style engine.
• Cloud managed - The hosted service of your engine.

Key Points

• Embedded stores win for speed of development.
• pgvector keeps data and vectors in one system.
• Dedicated engines add scale, filters and replication.
• Re-evaluate the store when data or traffic grows 10x.