pgvector with LangChain: Build a RAG Pipeline on PostgreSQL

LangChain has a vectorstore abstraction that lets you swap out the underlying vector database without rewriting your application logic. Swap Chroma for Pinecone, Pinecone for Weaviate, whatever. In theory, it's clean. In practice, most teams end up staying with whatever they picked first, because migration is never as simple as swapping a class name. So the decision matters. And if you're building on PostgreSQL, the answer is almost always: use pgvector. Your embeddings live in the same database as your users, documents, and application state. No sync pipeline. No eventual consistency. Full SQL. This guide walks through the LangChain PGVector integration from scratch, including document loading, embedding, similarity search, metadata filtering, and wiring it into a working retrieval chain. Why pgvector Over a Dedicated Vectorstore Before the code, let's be direct about the tradeoff. Dedicated vectorstores like Pinecone are fast and scale to billions of vectors without you thinking abou