llm-spell is both a library and command-line utility that corrects spelling mistakes using a Large Language Model (LLM). It is powered by llm.rb.

• ✨ LLM-powered corrections – smarter spelling fixes than traditional tools
• 🤖 Fewer false positives – avoids flagging uncommon but valid words
• 🌐 Broad provider support – use OpenAI, Gemini, or xAI (Grok) out of the box
• 💻 Offline ready – run locally with Ollama and LlamaCpp, no cloud required
• 🔒 Privacy – keep sensitive text local with offline models
• 🛠️ Easy to use – provides an easy to use library and command-line utility

#!/usr/bin/env ruby
require "llm"
require "llm/spell"

##
# Text
llm  = LLM.openai(key: ENV["OPENAI_SECRET"])
text = LLM::Spell::Text.new("Ths is a smple txt with sme speling erors.", llm)
print "mistakes: ", text.mistakes, "\n"
print "corrections: ", text.corrections, "\n"

##
# PDF
llm  = LLM.openai(key: ENV["OPENAI_SECRET"])
file = LLM::Spell::Document.new("typos.pdf", llm)
print "mistakes: ", file.mistakes, "\n"
print "corrections: ", file.corrections, "\n"

The command line interface can be configured through the configuration file located at $XDG_CONFIG_HOME/llm-spell.yml or ~/.config/llm-spell.yml. This step is required to use llm-spell from the command line but it is not required when using llm-shell as a library:

# ~/.config/llm-spell.yml
openai:
  key: YOURKEY
gemini:
  key: YOURKEY
xai:
  key: YOURKEY
ollama:
  host: localhost
  model: gpt-oss
llamacpp:
  host: localhost

Usage: llm-spell [OPTIONS]
    -p, --provider NAME              Required. Options: gemini, openai, xai, ollama or llamacpp.
    -f, --file FILE                  Required. The file to check.
    -v, --version                    Optional. Print the version and exit.

llm-spell requires Ruby 3.2+ and can be installed via RubyGems:

gem install llm-spell

This project was born while I was working on the documentation for a friend's open source project. After realizing how much manual effort was involved with traditional spell checkers I decided to see if I could leverage LLMs to make the process easier and also faster.

Compared to traditional spell checkers like aspell and hunspell, llm-spell provides significantly more accurate suggestions with far fewer false positives – eliminating the need to manually ignore irrelevant corrections, and that often reduces the overall time spent on correcting spelling mistakes.

I would call the experiment a success but I also realize this approach is not for everyone, or every situation. For example, my friend preferred to not use AI for this and instead we opted to stick with hunspell – even though it meant more manual work.

BSD Zero Clause
See LICENSE