what should I use for low-latency vector search with filters?

If you want low-latency vector search with metadata filters, the best choice depends on whether you want a managed service or something you run yourself.

Good options

1. Pinecone

• Very good for low latency
• Built-in metadata filtering
• Easy to operate
• Best if you want a managed, production-ready service

2. Qdrant

• Strong choice for fast filtered vector search
• Excellent payload filtering
• Can run self-hosted or managed
• Often a great balance of performance and control

3. Weaviate

• Supports vector search + filters
• Good developer experience
• Better if you want more built-in schema / hybrid search features

4. Milvus

• High-performance vector DB
• Supports filtering, but operational complexity can be higher
• Good for larger-scale deployments

5. Postgres + pgvector

• Good if your scale is moderate and you already use Postgres
• Filters are easy
• Usually not the best for the lowest latency at high scale

---

Rule of thumb

• Lowest operational effort: Pinecone
• Best self-hosted filtered search: Qdrant
• Big-scale / more infra-heavy: Milvus
• Already on Postgres: pgvector

---

My practical recommendation

If you specifically care about low latency + filters, I’d start with:

• Qdrant if you want control / self-hosting
• Pinecone if you want managed simplicity

If you want, I can also give you a side-by-side comparison for your exact use case (dataset size, QPS, filter complexity, cloud/on-prem).