PGVector

Yboard ships with pgvector — a PostgreSQL extension for storing vector embeddings and performing similarity search.

Overview

PGVector enables AI-powered features like semantic search, recommendation engines, and RAG (Retrieval Augmented Generation) pipelines directly in your database.

Enabling the Extension

PGVector is available on all Yboard projects. Enable it by running:

CREATE EXTENSION IF NOT EXISTS vector;

Creating Vector Columns

Add a vector column to any table:

ALTER TABLE documents ADD COLUMN embedding vector(1536);

The dimension (e.g., 1536) must match the output dimension of your embedding model.

Generating Embeddings with Yboard AI

Use Yboard’s built-in AI service to generate embeddings:

POST /api/ai/embeddings
{
  "input": "Your text to embed",
  "model": "text-embedding-3-small"
}

Available Embedding Models

Model	Dimensions	Best For
`text-embedding-3-small`	1536	General purpose, cost-effective
`text-embedding-3-large`	3072	Higher accuracy
`text-embedding-ada-002`	1536	Legacy support

Storing Embeddings

INSERT INTO documents (content, embedding)
VALUES ('Document text', '[0.1, 0.2, ...]'::vector);

Querying Vectors

Distance Operators

Operator	Distance Type	Use Case
`<->`	L2 (Euclidean)	Default similarity
`<=>`	Cosine	Normalized embeddings
`<#>`	Inner product	Dot product similarity

Similarity Search with RPC

Create a helper function for similarity search:

CREATE OR REPLACE FUNCTION match_documents(
  query_embedding vector(1536),
  match_threshold float DEFAULT 0.8,
  match_count int DEFAULT 10
)
RETURNS TABLE (id uuid, content text, similarity float)
AS $$
  SELECT id, content, 1 - (embedding <=> query_embedding) AS similarity
  FROM documents
  WHERE 1 - (embedding <=> query_embedding) > match_threshold
  ORDER BY embedding <=> query_embedding
  LIMIT match_count;
$$ LANGUAGE sql;

Building a RAG Pipeline

Basic RAG Flow

Embed the query — Generate an embedding for the user’s question
Search — Find similar documents using vector similarity
Generate — Send matched context + question to an LLM

Production RAG

For production use, consider:

Chunking large documents before embedding
Using metadata filters alongside vector search
Caching frequently accessed embeddings

Indexing

HNSW (Recommended)

CREATE INDEX ON documents
USING hnsw (embedding vector_cosine_ops)
WITH (m = 16, ef_construction = 64);

IVFFlat

CREATE INDEX ON documents
USING ivfflat (embedding vector_cosine_ops)
WITH (lists = 100);

Operator Classes

Class	Operator	Index Type
`vector_l2_ops`	`<->`	HNSW, IVFFlat
`vector_cosine_ops`	`<=>`	HNSW, IVFFlat
`vector_ip_ops`	`<#>`	HNSW, IVFFlat

Best Practices

Choose the right dimensions — Match your embedding model’s output dimensions
Add indexes once you have >1000 rows for faster queries
Use cosine distance (<=>) for normalized embeddings
Chunk documents into smaller pieces (512–1024 tokens) before embedding
Cache embeddings for frequently repeated queries

Overview

Core Features

App Management

Integrations

Overview

Enabling the Extension

Creating Vector Columns

Generating Embeddings with Yboard AI

Available Embedding Models

Storing Embeddings

Querying Vectors

Distance Operators

Similarity Search with RPC

Building a RAG Pipeline

Basic RAG Flow

Production RAG

Indexing

HNSW (Recommended)

IVFFlat

Operator Classes

Best Practices

​Overview

​Enabling the Extension

​Creating Vector Columns

​Generating Embeddings with Yboard AI

​Available Embedding Models

​Storing Embeddings

​Querying Vectors

​Distance Operators

​Similarity Search with RPC

​Building a RAG Pipeline

​Basic RAG Flow

​Production RAG

​Indexing

​HNSW (Recommended)

​IVFFlat

​Operator Classes

​Best Practices

Overview

Enabling the Extension

Creating Vector Columns

Generating Embeddings with Yboard AI

Available Embedding Models

Storing Embeddings

Querying Vectors

Distance Operators

Similarity Search with RPC

Building a RAG Pipeline

Basic RAG Flow

Production RAG

Indexing

HNSW (Recommended)

IVFFlat

Operator Classes

Best Practices