do i need pgvector or is postgres enough for embeddings?

Usually Postgres is enough at first — you only need pgvector if you want fast, native similarity search inside Postgres.

Use plain Postgres if:

• You’re just storing embeddings as arrays/JSON
• You have small volume or low search frequency
• You’ll do search elsewhere (app code, external vector DB, etc.)
• You don’t need efficient nearest-neighbor queries yet

Use pgvector if:

• You want to run queries like “find the top 10 most similar items” directly in SQL
• You have many embeddings and need performance
• You want indexes for approximate nearest-neighbor search
• You want to keep everything in one database

Key difference

Plain Postgres can store embeddings, but it’s not good at vector similarity search by itself.
pgvector adds:

• a vector column type
• similarity operators / distance functions
• indexes like HNSW or IVFFlat

Practical rule

• Prototype / small scale: plain Postgres is fine
• Production semantic search / RAG / recommendations: use pgvector

If you want, I can also show:

1. a plain Postgres schema for embeddings, and
2. the same schema using pgvector.