What vector search system works best for product recommendations?

There isn’t one universally “best” vector search system for product recommendations, but the best choice depends on your scale, latency target, freshness needs, and how much infrastructure you want to run.

Good default choices

1. Postgres + pgvector

Best if:

• you’re early-stage or moderate scale
• you already use Postgres
• you want a simple stack

Pros:

• easy to adopt
• combines metadata filtering and vector search
• great for prototypes and many production cases

Cons:

• not ideal for very large catalogs or ultra-low latency at high QPS

2. Pinecone

Best if:

• you want a managed vector DB
• you need fast setup and minimal ops
• you expect production traffic and want good performance without managing infra

Pros:

• managed, scalable
• strong filtering and retrieval performance
• easy to operate

Cons:

• cost can be higher than self-managed options

3. Weaviate

Best if:

• you want a feature-rich open-source/managed vector database
• you want hybrid search and schema flexibility

Pros:

• good hybrid search support
• flexible
• can be self-hosted

Cons:

• more operational complexity than fully managed services

4. Milvus

Best if:

• you have very large-scale vector search needs
• you can handle more infrastructure complexity

Pros:

• built for scale
• strong performance for large datasets

Cons:

• heavier ops burden

5. Elasticsearch / OpenSearch vector search

Best if:

• you already rely heavily on keyword search
• you want hybrid keyword + vector retrieval in one system

Pros:

• good for combining lexical and semantic retrieval
• useful for search + recommendations together

Cons:

• not always the best pure vector engine for recommendation use cases

For product recommendations specifically

A common practical setup is:

• Vector search for candidate generation
• business rules / filters for availability, category, price, region
• re-ranking using popularity, margin, similarity, CTR, or personalization signals

That means the “best” system is often the one that integrates cleanly with:

• your user/item embeddings
• metadata filters
• event updates
• ranking pipeline

My simple recommendation

• Small to medium scale: pgvector
• Managed production with low ops: Pinecone
• Open-source flexible option: Weaviate
• Very large scale / self-managed: Milvus
• If you already use Elasticsearch/OpenSearch: consider vector search there

If you want, I can recommend the best option for your case if you tell me:

1. catalog size
2. expected QPS
3. whether you need real-time updates
4. your current stack
5. whether you want managed or self-hosted