The Vector DB's Role
A dedicated database that fast-searches top similars from a large number of embedding vectors in RAG or recommendation systems. It speeds up approximate nearest-neighbor (ANN) search with indexes like HNSW, IVF.
Comparison of Main Options
| Product | Feature | Suited for |
|---|---|---|
| Pinecone | Fully managed, easy | Want to start quickly |
| Weaviate | OSS+cloud, hybrid | Keyword+vector together |
| pgvector | PostgreSQL extension | When you have existing PG |
| Qdrant | Rust-built, fast, OSS | Large-scale/speed-focused |
| Milvus | OSS, super-large-scale | Over 1B vectors |
| Chroma | Lightweight, for dev | Prototype |
| Elastic / OpenSearch | Add to existing search infra | Existing ES environment |
Selection Points
1. Data Scale
- Under 100k vectors: Chroma, SQLite + local
- 1M-100M: pgvector, Pinecone, Weaviate, Qdrant
- Over 1B: Milvus, Pinecone Enterprise
2. Existing Stack
- Have PostgreSQL → pgvector is the minimal cost
