full_search

SQLite FTS5 full-text search for Rails/ActiveRecord. A lightweight, self-contained alternative to pg_search for apps already running on SQLite.

Installation

Add to your Gemfile:

gem "full_search"

Run:

bundle install
bin/rails generate full_search:install

Usage

class Customer < ApplicationRecord
  full_search do
    field :first_name, weight: 5
    field :last_name,  weight: 5
    filter :account_id, required: true
  end
end

Customer.full_search("sam", filters: { account_id: 1 }).page(params[:page])

Features

• Declarative full_search DSL
• SQLite FTS5 backed
• Required-filter enforcement for multi-tenant apps
• Exact-match queries for encrypted identifiers
• Soft-delete awareness
• Automatic table creation and schema-drift detection
• Phrase, exclusion, and OR query operators
• FTS5 highlight() support for result snippets
• Opt-in trigram typo/substring fallback (requires SQLite >= 3.34)
• Rake tasks: full_search:rebuild, full_search:optimize, full_search:status

Query operators

Customer.full_search('"Sam Smith"', filters: { account_id: 1 })      # phrase
Customer.full_search("honda -civic", filters: { account_id: 1 })     # exclusion
Customer.full_search("honda OR toyota", filters: { account_id: 1 })  # OR

Highlighting

class Customer < ApplicationRecord
  full_search do
    field :first_name, weight: 5
    highlight open_tag: "<mark>", close_tag: "</mark>"
  end
end

Customer.full_search("sam", filters: { account_id: 1 }, highlight: true)
# each result has #full_search_snippet

Typo tolerance

class Customer < ApplicationRecord
  full_search do
    field :first_name, weight: 5
    typo_tolerance
  end
end

typo_tolerance uses an FTS5 trigram shadow table as a fallback when the primary index returns no results. It is substring matching, not edit-distance correction, and requires SQLite >= 3.34.

Known limitations

• Queries run with highlight: true return an Array of records, not an ActiveRecord::Relation. No further chaining (.where, .order, .limit) is possible after highlighting is applied.
• Per-model only; no built-in multi-model aggregator