sinatra app that streams openai chat responses

sse.rb

require 'bundler/setup'
require 'dotenv'
require 'sinatra'
require 'sinatra/streaming'
require 'json'
require 'typhoeus'

set :server, 'thin' # not sure if puma also has streaming nowadays?

Dotenv.load
API_BASE_URL = "https://api.openai.com/v1"
ACCESS_TOKEN = ENV["OPENAI_ACCESS_TOKEN"]

def streaming_completions_request(prompt)
  headers = {
    "Content-Type" => "application/json",
    "Authorization" => "Bearer #{ACCESS_TOKEN}"
  }

  request = Typhoeus::Request.new(
    "https://api.openai.com/v1/completions",
    method: :post,
    headers: headers,
    body: {
      prompt: prompt,
      model: "text-davinci-003",
      stream: true,
      max_tokens: 400
    }.to_json
  )

  request
end

get '/stream_completions' do
  content_type 'text/event-stream'
  cache_control :no_cache

  stream(:keep_open) do |out|
    request = streaming_completions_request("Once upon a time...")

    request.on_body do |chunk|
      data_line = chunk.lines.find { |line| line.start_with?('data:') }
      next unless data_line

      json_str = data_line.sub('data:', '').strip
      json = JSON.parse(json_str)
      out << "data: #{json.to_json}\n\n"
    end

    request.on_complete do |response|
      puts response.inspect
      out << "event: close\n"
      out << "data: {}\n\n"
      out.close
    end

    request.run
  end
end