🧠 active_record-vector

Native AI vector embeddings, semantic search & RAG for Rails ActiveRecord

active_record-vector gives any Rails ActiveRecord model native AI vector embedding generation, semantic similarity search, and RAG (Retrieval-Augmented Generation) document chunking capabilities in 2 lines of code.

Works out-of-the-box with OpenAI, Ollama (100% free local AI), Cohere, pgvector (PostgreSQL), and SQLite/MySQL.

• 🤖 has_vector Macro: Automatically generates & updates AI vector embeddings on model save callbacks.
• 🔍 Native ActiveRecord Scopes: Chain .semantic_search("query") and .nearest_to(vector) directly with standard Rails queries (where, limit, order).
• 🆓 Free Local AI via Ollama: Generate embeddings 100% offline, locally, and free using nomic-embed-text or all-minilm.
• ⚡ Multi-Provider Support: OpenAI (text-embedding-3-small), Ollama, Cohere, or custom Procs/Lambdas.
• 📑 RAG Text Chunker: Built-in document text splitting utility (ActiveRecordVector::Chunker) with token-aware overlap.
• 🗄️ PostgreSQL pgvector Integration: Migration DSL extensions (add_vector_column, add_vector_index) supporting HNSW and IVFFlat indexes with fallback Ruby distance algorithms for SQLite/MySQL.

Add to your Rails application's Gemfile:

gem "active_record-vector"

And execute:

bundle install

Add has_vector to your ActiveRecord model:

class Article < ApplicationRecord
  has_vector :embedding,
             provider: :openai,                 # :openai, :ollama, :cohere, or custom proc
             model: "text-embedding-3-small",
             from: [:title, :body],              # concatenated automatically
             auto_generate: true                # before_save callback
end

Perform vector similarity searches using standard Rails scopes:

# Semantic search by query text
Article.semantic_search("Ruby on Rails 8 performance", limit: 5)

# Chain with standard ActiveRecord queries
Article.where(published: true)
       .semantic_search("AI integration", limit: 10)

Generate embeddings 100% offline, privately, and for free using Ollama:

class Article < ApplicationRecord
  has_vector :embedding,
             provider: :ollama,
             model: "nomic-embed-text",  # or "all-minilm"
             host: "http://localhost:11434",
             from: :body
end

Split long documents into overlapping chunks for embedding generation in RAG pipelines:

# Split long document text
chunks = ActiveRecordVector::Chunker.split(long_text, chunk_size: 1000, chunk_overlap: 200)

chunks.each do |chunk_text|
  article.chunks.create!(content: chunk_text) # auto-generates vector embedding
end

class AddEmbeddingToArticles < ActiveRecord::Migration[7.2]
  def change
    # Adds pgvector column (or text column fallback on SQLite)
    add_vector_column :articles, :embedding, dimensions: 1536

    # Adds HNSW vector index for high-speed similarity queries
    add_vector_index :articles, :embedding, type: :hnsw, distance: :cosine
  end
end

git clone https://github.com/aditya-8108/active_record-vector.git
cd active_record-vector

bundle config set --local path 'vendor/bundle'
bundle install

# Run test suite
bundle exec rspec

Distributed under the MIT License.