Let’s write some code

Imports and Setup:

import 'dotenv/config'
import { openai } from './openai.js'
import math from 'advanced-calculator'
const QUESTION = process.argv[2] || 'hi'

• The code starts by importing necessary modules.
  • dotenv/config is imported to load environment variables (e.g., API keys) from a .env file.
  • An openai module is imported, which likely contains functions to interact with the OpenAI API.
  • The advanced-calculator module is imported as math, providing math calculation capabilities.
  • QUESTION is set to the first command-line argument, or defaults to 'hi' if none is provided.

Messages and Functions:

const messages = [
  {
    role: 'user',
    content: QUESTION,
  },
]

const functions = {
  calculate: async ({ expression }) => {
    return math.evaluate(expression)
  },
}

• messages: Initializes an array with the user's question.
• functions: Defines an object where the keys are function names, and the values are the corresponding functions. Here, there's only one function, calculate, which evaluates a mathematical expression.

Get Completion Function:

const getCompletion = async (messages) => {
  const response = await openai.chat.completions.create({
    model: 'gpt-3.5-turbo-0613',
    messages,
    functions: [
      {
        name: 'calculate',
        description: 'Run a math expression',
        parameters: {
          type: 'object',
          properties: {
            expression: {
              type: 'string',
              description:
                'Then math expression to evaluate like "2 * 3 + (21 / 2) ^ 2"',
            },
          },
          required: ['expression'],
        },
      },
    ],
    temperature: 0,
  })

  return response
}

• This function communicates with the OpenAI API to get a completion based on the provided messages. It specifies that the model can use a function named calculate to evaluate math expressions.

Main Loop:

let response
while (true) {
  response = await getCompletion(messages)

  if (response.choices[0].finish_reason === 'stop') {
    console.log(response.choices[0].message.content)
    break
  } else if (response.choices[0].finish_reason === 'function_call') {
    const fnName = response.choices[0].message.function_call.name
    const args = response.choices[0].message.function_call.arguments

    const functionToCall = functions[fnName]
    const params = JSON.parse(args)

    const result = functionToCall(params)

    messages.push({
      role: 'assistant',
      content: null,
      function_call: {
        name: fnName,
        arguments: args,
      },
    })

    messages.push({
      role: 'function',
      name: fnName,
      content: JSON.stringify({ result: result }),
    })
  }
}

• This is the main loop where interactions with the OpenAI model take place.
• If the model's response has a finish_reason of 'stop', it prints the response and exits the loop.