What should I use for vector search with strict metadata filtering?

If you need vector search with strict metadata filtering, the best choice depends on how hard “strict” is and where your data lives. Here’s the practical shortlist:

Best general-purpose options

• PostgreSQL + pgvector
  Best if you want strong, exact metadata filtering with SQL semantics.

  • Pros: very reliable filters, transactional, easy to combine with structured data
  • Cons: not the fastest at very large scale compared with dedicated vector DBs

• Pinecone Good if you want managed infrastructure and metadata filters that are easy to use.

  • Pros: simple API, scalable, good hybrid/vector search
  • Cons: less flexible than SQL for complex boolean filtering

• Weaviate Good for vector search plus filtering, with a strong feature set.

  • Pros: flexible metadata filters, hybrid search, self-host or managed
  • Cons: more moving parts than Postgres

• Qdrant A strong choice for strict filtering with vectors.

  • Pros: efficient payload filtering, good performance, easy to self-host
  • Cons: fewer “database-like” capabilities than Postgres

If “strict filtering” means:

• Exact, complex filtering, joins, business rules → PostgreSQL + pgvector
• Large-scale vector search with solid metadata filters → Qdrant or Pinecone
• Hybrid search + flexible schema → Weaviate

Quick recommendation

• If you’re building something new and need strict correctness on filters, start with Postgres + pgvector.
• If you need high-scale ANN search and filtering is important but not relationally complex, use Qdrant.

If you want, I can also give you a decision table by scale, latency, and filter complexity.