This path is for you if every retrieval demo names a different database and they sound interchangeable. It separates an in-memory library, a dedicated engine, pgvector, and a managed platform. No part is live on this page yet. Start with Practical AI for what retrieval is, then come back to pick a product.
- 1 Do You Need a Vector Database or Just a Library? Choose an embedded library for batch corpora that fit in memory. Choose a vector database when you need live CRUD, metadata filters, sharding, or multi-tenant access. Scheduled · December 9, 2026
- 2 pgvector and Use the Database You Have: Pragmatic Vector Search Enable pgvector, store embeddings beside rows, build HNSW, and filter tenants in the same SQL. Prefer this until tens of millions of vectors or extreme write rates force a specialist store. Scheduled · December 10, 2026
- 3 The Managed Vector Stores Landscape: Pinecone, Weaviate, Qdrant, Chroma, and Milvus Map managed vector engines to workload: serverless ease, Rust performance, modular vectorizers, distributed scale, or lightweight local demos. Match features to real filters and write rates. Scheduled · December 11, 2026
- 4 Hybrid Search and Metadata Filtering: The Secret to Production Retrieval Combine dense and sparse retrieval with RRF, and apply metadata filters early. Exact IDs and tenant ACLs are not optional polish. Scheduled · December 12, 2026
- 5 Vector Database Cost, Sizing, and Infrastructure for Beginners Plan vector capacity from dimensions, count, and index overhead. Use quantization when RAM is the constraint. Treat infra sizing as a first-class design step. Scheduled · December 13, 2026
- 6 Choosing the Right Vector Stack: Personal Side Project vs Enterprise Production Climb the vector stack by corpus size, write pattern, filter needs, and team capacity. Measure retrieval quality before you migrate. Scheduled · December 14, 2026