Ragify

RAG (Retrieval-Augmented Generation) for Rails. Add semantic search and AI chat to any ActiveRecord model using pgvector and OpenAI.

No external vector databases. No complex infrastructure. Just your PostgreSQL and a few lines of code.

Requirements

Ruby >= 3.2
Rails >= 7.0
PostgreSQL with pgvector extension
OpenAI API key

Installation

Add to your Gemfile:

gem "ragify"

Then run:

bundle install

Quick Start

1. Configure

# config/initializers/ragify.rb
Ragify.configure do |config|
  config.openai_api_key = ENV["OPENAI_API_KEY"]
  config.embedding_model = "text-embedding-3-small"  # default
  config.chat_model = "gpt-4o-mini"                  # default
  config.similarity_threshold = 0.2                   # default
end

2. Add embedding column

rails generate ragify:install FaqEntry
rails db:migrate

This creates a migration that adds a vector column with an IVFFlat index.

3. Include in your model

class FaqEntry < ApplicationRecord
  include Ragify::Embeddable

  ragify_content { |record| "#{record.question} #{record.answer}" }
end

4. Use it

# Semantic search
results = FaqEntry.semantic_search("how does cashback work?")
# => [{ record: #<FaqEntry>, content: "...", similarity: 0.87 }, ...]

# RAG chat (search + LLM)
response = Ragify.chat("how does cashback work?",
  context_model: FaqEntry,
  system_prompt: "You are a helpful assistant for a loyalty program."
)
# => { answer: "Cashback is automatically...", context: [...], tokens: 342 }

# Generate embeddings for existing records
FaqEntry.embed_all!

Features

Embeddable

Include Ragify::Embeddable in any ActiveRecord model. Embeddings are automatically generated on create/update.

class Article < ApplicationRecord
  include Ragify::Embeddable

  # Block form - full control over content
  ragify_content { |r| "#{r.title}\n#{r.body}" }

  # Or method form
  ragify_content :embedding_text
end

Semantic Search

Find records by meaning, not keywords.

Article.semantic_search("climate change effects",
  limit: 10,           # max results (default: 5)
  threshold: 0.3       # min similarity 0-1 (default: 0.2)
)

Returns an array of hashes with :record, :content, and :similarity.

RAG Chat

Combine semantic search with an LLM for context-aware answers.

Ragify.chat("what's your refund policy?",
  context_model: FaqEntry,
  system_prompt: "You are a customer support agent.",
  search_limit: 5,
  history: [
    { role: "user", content: "hi" },
    { role: "assistant", content: "Hello! How can I help?" }
  ]
)

Returns a hash:

{
  answer: "Our refund policy allows...",
  context: [{ record: ..., content: "...", similarity: 0.91 }],
  tokens: 487,
  had_context: true
}

Batch Embedding

Generate embeddings for all records (useful for initial setup):

FaqEntry.embed_all!

# Or embed specific records
Ragify.embed("some text")
Ragify.embed_batch(["text 1", "text 2", "text 3"])

Configuration

Option	Default	Description
`openai_api_key`	`ENV["OPENAI_API_KEY"]`	OpenAI API key
`embedding_model`	`text-embedding-3-small`	Embedding model
`embedding_dimensions`	`1536`	Vector dimensions
`chat_model`	`gpt-4o-mini`	Chat completion model
`max_tokens`	`1000`	Max response tokens
`temperature`	`0.3`	LLM temperature
`similarity_threshold`	`0.2`	Min cosine similarity
`search_limit`	`5`	Default search results
`ivfflat_probes`	`10`	IVFFlat index probes
`logger`	`Rails.logger`	Logger instance

How It Works

Embedding: When a record is saved, ragify_content extracts text and sends it to OpenAI's embedding API. The resulting vector is stored in a pgvector column.
Search: Queries are embedded the same way, then pgvector finds the nearest neighbors using cosine similarity with an IVFFlat index.
Chat: The search results are injected as context into an LLM prompt, which generates a grounded answer.

Contributing

Bug reports and pull requests are welcome at https://github.com/pashgo/ragify.

License

MIT License. See LICENSE for details.

ragify

Runtime