rubyllm-embeddings

name	rubyllm-embeddings
description	Generate vector embeddings with RubyLLM. Use this skill for semantic search, recommendations, content similarity, RAG applications, and any task requiring numerical text representations.

RubyLLM Embeddings

Transform text into numerical vectors for semantic search, recommendations, and content similarity.

Basic Usage

# Single text
embedding = RubyLLM.embed("Ruby is elegant")
vector = embedding.vectors  # Array of floats
puts "Dimension: #{vector.length}"

# Multiple texts (batched, more efficient)
embeddings = RubyLLM.embed(["Ruby", "Python", "JavaScript"])
puts "Vectors: #{embeddings.vectors.length}"  # => 3

# Specific model
embedding = RubyLLM.embed("text", model: 'text-embedding-3-large')

# Access metadata
puts "Model: #{embedding.model}"
puts "Input tokens: #{embedding.input_tokens}"

Models

Model	Provider	Dimensions	Input/$1M
text-embedding-3-small	OpenAI	1536	$0.02
text-embedding-3-large	OpenAI	3072	$0.13
gemini-embedding-3	Google	768	$0.025

# OpenAI
RubyLLM.embed("text", model: 'text-embedding-3-small')

# Google
RubyLLM.embed("text", model: 'gemini-embedding-3')

Similarity Calculation

def cosine_similarity(a, b)
  dot = a.zip(b).sum { |x, y| x * y }
  dot / (Math.sqrt(a.sum { |x| x**2 }) * Math.sqrt(b.sum { |y| y**2 }))
end

vec1 = RubyLLM.embed("cat").vectors
vec2 = RubyLLM.embed("dog").vectors
vec3 = RubyLLM.embed("car").vectors

puts cosine_similarity(vec1, vec2)  # => ~0.85 (similar concepts)
puts cosine_similarity(vec1, vec3)  # => ~0.15 (different concepts)

RAG Example

class DocumentSearch
  def initialize(documents)
    @documents = documents
    @index = build_index
  end
  
  def search(query, limit: 5)
    # 1. Embed query
    query_embedding = RubyLLM.embed(query).vectors
    
    # 2. Find similar documents
    @documents.map do |doc|
      {
        doc: doc,
        similarity: cosine_similarity(query_embedding, doc.embedding)
      }
    end.sort_by { |d| -d[:similarity] }.first(limit)
  end
  
  private
  
  def build_index
    @documents.map do |doc|
      {
        doc: doc,
        embedding: RubyLLM.embed(doc.content).vectors
      }
    end
  end
  
  def cosine_similarity(a, b)
    dot = a.zip(b).sum { |x, y| x * y }
    dot / (Math.sqrt(a.sum { |x| x**2 }) * Math.sqrt(b.sum { |y| y**2 }))
  end
end

# Usage
docs = Document.all
searcher = DocumentSearch.new(docs)
results = searcher.search("Ruby programming", limit: 3)

Rails Integration

With pgvector

# db/migrate/create_documents.rb
class CreateDocuments < ActiveRecord::Migration[7.2]
  def change
    create_table :documents do |t|
      t.text :content
      t.vector :embedding, limit: 1536  # pgvector
    end
    
    add_index :documents, :embedding, using: :ivfflat, opclass: :vector_cosine_ops
  end
end

# app/models/document.rb
class Document < ApplicationRecord
  before_save :generate_embedding, if: :content_changed?
  
  def self.search_similar(query, limit: 5)
    query_embedding = RubyLLM.embed(query).vectors
    where("embedding <=> :embedding < 0.5", embedding: query_embedding.to_s)
      .order("embedding <=> :embedding", embedding: query_embedding)
      .limit(limit)
  end
  
  private
  
  def generate_embedding
    self.embedding = RubyLLM.embed(content).vectors
  end
end

Batch Processing

# Efficient batch embedding
texts = Document.pluck(:content)
batches = texts.each_slice(100)

batches.each do |batch|
  embeddings = RubyLLM.embed(batch)
  # Process embeddings
end

Async Embeddings

require 'async'

Async do
  documents.map do |doc|
    Async do
      doc.update(embedding: RubyLLM.embed(doc.content).vectors)
    end
  end.map(&:wait)
end

See Also

• Main RubyLLM: rubyllm
• RAG Pattern: See rubyllm-agents for workflow examples