What Vector DBs Do

Store high-dimensional vectors from embeddings.

Enable fast similarity search – find items closest to a query vector.

Foundation of RAG, semantic search, recommendation.

Pinecone

Managed service. Zero ops.

Fastest for large scale (100M+ vectors).

Expensive: $70/mo starter, scales up.

Best for: production systems that need reliability.

Weaviate

Open source, can self-host or use cloud.

Rich features: hybrid search, filtering, multi-tenancy.

Learning curve steeper than Pinecone.

Best for: teams wanting control + features.

pgvector

Postgres extension. Uses existing DB.

Cheapest option – no separate service.

Slower at scale (millions of vectors).

Best for: starting small, already using Postgres.

Qdrant

Open source, Rust-based.

Very fast, good filtering.

Growing ecosystem.

Best for: performance-critical self-hosted.

Chroma

Simple, developer-friendly.

Great for prototypes and small projects.

Not production-scale yet.

Best for: learning and MVPs.

When to Use Managed

Team focused on product, not infra.

Traffic unpredictable.

Budget for $200-2000/mo.

Compliance simpler with managed.

When to Self-Host

Cost sensitive at scale.

Data residency requirements.

DevOps capability in team.

Already running similar infrastructure.

Performance Considerations

Index type matters: HNSW is fastest for most.

Dimensionality: 1536 (OpenAI) vs 768 (open source models).

Batch operations for bulk inserts.

Filter pushdown reduces search space.

Cost Modeling

Storage: usually small cost.

Query cost: scales with QPS.

Pinecone: pay per pod running.

Self-hosted: server costs.

Ballpark: $50-500/mo for typical SaaS.

Migration Path

Start with pgvector if using Postgres.

Move to dedicated when scaling issues.

Migration usually straightforward – re-index data.

Our Recommendation

MVP stage: pgvector.

Growth stage under 10M vectors: Weaviate or Qdrant.

Scale/enterprise: Pinecone or Weaviate Cloud.

AI-specific: Chroma for very early experimentation.

Based on Real Projects

This guide is based on our work with:

• RepuAI
• SmartSearch

Further Reading

If this guide helped you, you might also want to read our comprehensive guide on AI Solutions.