ActivePrompt

ActivePrompt is a gem that helps you structure and manage LLM prompts in your rails apps. View the gem on RubyGems.

Installation

Add the following line to your Gemfile:

gem "active_prompt_rails", "~> 0.1.22", require: "active_prompt"

and run bundle install

Usage

When you want to create a new LLM prompt, you can generate one using the following command:

rails g active_prompt:prompt moderate_user_comment

This will create the following files in your rails project:

app/prompts/application_prompt.rb
app/prompts/moderate_user_comment_prompt.rb
app/prompts/templates/moderate_user_comment_prompt/system.liquid
app/prompts/templates/moderate_user_comment_prompt/user.liquid

You can add your system message and user message to the corresponding Liquid template files and include any dynamic variables you'd like, e.g.:

# /app/prompts/moderate_user_comment_prompt.rb

class ModerateUserCommentPrompt < ApplicationPrompt
  variable :comment
end

# /app/prompts/templates/moderate_user_comment_prompt/system.liquid

You are a comment moderator agent that is responsible for reviewing and moderating user comments on a social media platform. You are assigned to reviewing comments and returning your decision on whether to approve or flag them for further review. Check for bullying, hate speech, and threatening language. Swearing is allowed as long as it's not harrassing. You return your decision in JSON format with the following structure: 
{
  "result": "approve" (or "flag"),
  "reason": "optional reason for flagging the comment"
}

# /app/prompts/templates/moderate_user_comment_prompt/user.liquid

Review this comment and return your decision in the specified JSON format:

{{ comment }}

Then within your application, you can use this prompt when calling an LLM as such:

prompt = ModerateUserCommentPrompt.new(comment: user_comment)

client = OpenAI::Client.new
response = client.chat(
    parameters: {
        model: "gpt-4o",
        response_format: { type: "json_object" },
        messages: prompt.render_messages, # ActivePrompt::Base.render_messages returns OpenAI formatted message data
        temperature: 1.0
    }
)
content = response.dig("choices", 0, "message", "content")
data = JSON.parse(content)
puts data

You can also override the run() method in ApplicationPrompt to integrate your LLM layer with ActivePrompt and execute your prompts from the prompt object. Here's an example using OpenAI:

# frozen_string_literal: true

class ApplicationPrompt < ActivePrompt::Base
  def run(model: 'gpt-3.5-turbo-0613', json_output: false, temperature: 1.0)
    client = OpenAI::Client.new
    parameters = {
      model: model,
      messages: render_messages,
      temperature: temperature
    }
    parameters[:response_format] = { type: 'json_object' } if json_output
    client.chat(parameters: parameters)
  end
end

Using this, we can simplify the code shown earlier:

prompt = ModerateUserCommentPrompt.new(comment: user_comment)
response = prompt.run(model: 'gpt-4o')
output = response.dig("choices", 0, "message", "content")
data = JSON.parse(output)
puts data

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and the created tag, and push the .gem file to rubygems.org.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/Shaneprrlt/active_prompt.