Vector Stores

Storing and querying embeddings, similarity_search(), Pinecone, Chroma, pgvector, FAISS. When to use which.

Quick Reference

→VectorStore is the base interface — add_documents(), similarity_search(), similarity_search_with_score()
→FAISS is great for local development and prototyping — no server needed, pure in-memory
→Chroma is the lightweight persistent option — embedded mode with SQLite backing
→pgvector integrates with existing PostgreSQL infrastructure — ideal if you already run Postgres
→Pinecone is the managed cloud option — zero ops, scales automatically, but adds vendor lock-in

The VectorStore Interface

VectorStore

A vector store holds document embeddings and supports similarity search. Given a query, it finds the most semantically similar documents. Every vector store in LangChain implements the same interface: add_documents(), similarity_search(), and as_retriever().

Core VectorStore operations

FAISS (Local Development)

Quick setup for development

Chroma / pgvector / Pinecone

Store	Type	Persistence	Scale	Best For
FAISS	In-memory	Save/load files	Small	Prototyping, testing
Chroma	Embedded	SQLite	Small-Medium	Local apps, simple RAG
pgvector	PostgreSQL extension	PostgreSQL	Medium-Large	Already using Postgres
Pinecone	Managed cloud	Automatic	Large	Production, no infra management
Weaviate	Self-hosted/cloud	Automatic	Large	Multimodal, hybrid search

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.