Vector Database Selector

A vector database stores high-dimensional numerical vectors (embeddings) and enables fast similarity search. Instead of exact keyword matching, it finds semantically similar items using approximate nearest neighbor (ANN) algorithms. Used in RAG pipelines, semantic search, recommendation systems, and image search.

HNSW (Hierarchical Navigable Small World) is the most popular ANN algorithm. It builds a multi-layer graph structure where each layer contains a subset of nodes connected by edges. Search starts at the top (sparse) layer and navigates down to the bottom (dense) layer. It offers excellent recall (>99%) at high query speeds.

Chroma is open-source and runs locally — perfect for development, local RAG prototypes, and projects where data privacy is critical. Pinecone is fully managed cloud infrastructure — handles scaling, replication, and uptime automatically, but costs money and requires sending data to a third party.

Use pgvector when you already have PostgreSQL and your vector collection is under ~1M rows. It avoids running a separate service and supports transactional guarantees between vector data and relational data. For pure vector workloads at scale (10M+ vectors), dedicated systems like Qdrant or Weaviate offer better performance.

Faiss (Facebook AI Similarity Search) is a library for efficient similarity search, not a database. It doesn't persist data, support metadata filtering, or provide a server — it's an in-memory search engine. Most production systems wrap Faiss for search but use a traditional database for storage and metadata.

Retrieval-Augmented Generation — a technique where relevant documents are retrieved from a vector database and injected into an LLM's context window before it generates a response. This allows LLMs to answer questions about documents that weren't in their training data, and reduces hallucination.