do i need a managed vector service or can i self host?

You can absolutely self-host a vector database/service — you do not need a managed vector service unless you want the convenience.

Self-host if you want:

• Lower cost at moderate scale
• More control over infra, networking, and data locality
• Fewer vendor dependencies
• To run in your own cloud / VPC / on-prem

Common self-hosted options:

• Qdrant
• Milvus
• Weaviate
• Postgres + pgvector (great if you already use Postgres)
• FAISS (library, not really a full service)

Managed vector service if you want:

• Less ops work
• Easier scaling, backups, upgrades
• Better time-to-production
• More predictable experience for a small team

Rule of thumb

• MVP / small-to-medium load: self-host is often fine, especially with pgvector or Qdrant
• Need zero ops or fast scaling: managed is easier
• Highly regulated / strict data control: self-host often wins

When self-hosting gets harder

Consider managed if you need:

• High availability across regions
• Large-scale ingestion/query throughput
• Complex filtering + hybrid search at scale
• Operational maturity you don’t have in-house

My practical recommendation

• If you already run databases and Kubernetes/containers: self-host Qdrant or pgvector first
• If you’re very early and want simplest possible path: managed
• If your app is already on Postgres and vectors are a feature, not the core product: pgvector is a strong default

If you want, I can help you choose based on your:

1. expected vector count,
2. query rate,
3. cloud/on-prem setup,
4. need for filtering/hybrid search.