OpenAI offers official Node.js and Python client libraries for their API but does not currently package a gem for Rubyists looking to interact with their platform. A number of third party gems exist to simplify the integration, but integrating directly with OpenAI’s API is fairly straightforward thanks to their well documented with curl examples. This article looks into the steps needed to make API calls using HTTP.rb for both regular and streamed completions.

Requesting with HTTP.rb

To get started an API key is required. OpenAI offers the ability to generate a an API key here in their dashboard. While Ruby offers basic HTTP functionality through Net::HTTP, using an HTTP client library greatly simplifies HTTP usage. While many options exist, HTTP.rb is a basic client that offers a terse and easy to understand API. It can be installed using gem install http. Once installed, it is time to make a ChatGPT API call:

require 'http'

OPENAI_API_KEY = ENV.fetch('OPENAI_API_KEY', 'sk-...')

response = HTTP.auth("Bearer #{OPENAI_API_KEY}").post(
  "https://api.openai.com/v1/chat/completions",
  json: {
    model: 'gpt-4o',
    messages: [{
      role: "user",
      content: "Tell me a joke!",
    }],
  }
)

content = response.parse["choices"][0]["message"]["content"]
puts content # "Why don't skeletons fight each other? They don't have the guts!"

That’s it! We’ve made an API call to OpenAI’s ChatGPT. The HTTP.rb library automatically handles the parsing of the JSON response by calling .parse.

Streaming with HTTP.rb

While our API works, it does not yet provide key visual flair of real-time typing. An important UI / UX integration that many LLM providers offer through their portals via streaming back responses. OpenAI is no exception. To stream back content via the API two changes are required:

1. Introduce the stream: true key to our request.
2. Handle the server sent event (SEE) streaming payload that HTTP.rb gives back.

To do this a new class is required that can properly handle the SEE paylods. It processes data in chunks using a buffer that it constantly flushes lines matching data: ...\n\n. Within the lines it parses JSON to grab the 'choices', then those are printed back to the user:

require 'http'

OPENAI_API_KEY = ENV.fetch('OPENAI_API_KEY', 'sk-...')

class OpenAIEventStream
  LINE_REGEX = /data:\s*(?<data>.*)\n\n/

  def initialize
    @buffer = ""
  end

  # @yield [data] a parsed hash
  # @param [String] chunk
  def process!(chunk)
    @buffer << chunk

    while (line = @buffer.slice!(LINE_REGEX))
      match = LINE_REGEX.match(line)
      data = match[:data]
      break if data.eql?('[DONE]')
      yield JSON.parse(data)
    end
  end
end

response = HTTP.auth("Bearer #{OPENAI_API_KEY}").post(
  "https://api.openai.com/v1/chat/completions",
  json: {
    model: 'gpt-4o',
    messages: [{
      role: "user",
      content: "Tell me a joke!",
    }],
    stream: true,
  }
)

stream = OpenAIEventStream.new
response.body.each do |chunk|
  stream.process!(chunk) do |data|
    data["choices"].each do |choice|
      print(choice["delta"]["content"])
    end
  end
end

With this OpenAIEventStream class the data from OpenAI now prints in realtime! The text is streamed back to the user as it is generated via the LLM.

This article originally appeared on https://workflow.ing/blog/articles/using-chat-gpt-with-ruby.