ts-chatgpt

A library that is created to receive pure responses that are typed using the official ChatGPT API by OpenAI.

Install

npm install ts-chatgpt

It has been confirmed to work with Remix's loader function.

Usage

| Function | Description | Parameters | Return | | -------- | ------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | -------------------------- | | prompt | Get a response from ChatGPT API | props - The props contains an model name you want to use for the ChatGPT API, an array of Message type and an options object. | Promise<ChatGPTResponse> |

When calling this function, be sure to set the OPENAI_API_KEY environment variable to the API key you received from OpenAI.

import { prompt } from "ts-chatgpt";

const response = await prompt({
  model: "gpt-4",
  messages: [
    {
      role: "user",
      content:
        "In the style of Nicholas Sparks, please summarize the following introductory You are limited to 140 characters. 'I love Android and I develop applications using Kotlin and Jetpack Compose.'",
    },
  ],
  options: {
    temperature: 0.1,
  },
});

Since dotenv.config() is automatically called internally, developers do not need to install dotenv to load OPENAI_API_KEY themselves.

Props

When calling prompt(), you must pass an object containing the following as an argument:

| Key | Description | Type | Required | | ---------- | --------------------------------------------------- | --------------- | -------- | | model | The model name you want to use for the ChatGPT API. | string | ✅ | | messages | An array of Message type. | Message[] | ✅ | | options | An object containing options. | PromptOptions | ❌ |

The following values are currently available for the model More will be added in the future.

| Model | Description | Available | | -------------------- | -------------------------------------------- | --------- | | gpt-3.5-turbo-0301 | The default model. | ✅ | | gpt-3.5-turbo | - | ✅ | | gpt-4 | GPT-4 is the latest and most powerful model. | ✅ |

The following values can be specified by the user as messages to be passed to the prompt function.

| Key | Description | Type | Required | | --------- | --------------------------- | ------------------------------------- | -------- | | role | The role of the message. | "system", "assistant" or "user" | ✅ | | content | The content of the message. | string | ✅ |

The following values can be specified by the user as options to be passed to the prompt function.

| Key | Description | Type | | ------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------- | | apiKey | API key that can be obtained from the OpenAI configuration page. You can omit this value by setting the OPENAI_API_KEY environment variable. | string | | temperature | The lower the temperature, the more accurate the results. API temperatures set to 0 or close to 0 (e.g. 0.1 or 0.2) tend to give better results in most cases; with GPT-3, the higher the temperature, the more creative and random the results, while with Codex, the higher the temperature, the more truly random and erratic the response can be. | number |

For detailed specifications of the ChatGPT API, please refer to this document.

Response Type

There are two types of return values for the prompt function: ChatGPT and ChatGPTError.

| Type | Description | | -------------- | ------------------------------------------------------- | | ChatGPT | The response from the ChatGPT API. | | ChatGPTError | The response from the ChatGPT API when an error occurs. |

ChatGPT by type is as follows:

type ChatGPT = {
  choices?:
    | {
        message: {
          role: string;
          content: string;
        };
        finish_reason: string;
        index: number;
      }[]
    | undefined;
  object: string;
  id: string;
  created: number;
  model: string;
  usage: {
    prompt_tokens: number;
    completion_tokens: number;
    total_tokens: number;
  };
};

Next, ChatGPTError as a type is as follows:

type ChatGPTError = {
  error: {
    message: string;
    type: string;
    param: string | null;
    code: string | null;
  };
};

Team

| | | --------------------------------------------------------- | | Keisuke Takagi |

License

This project is licensed under the terms of the MIT license.

MIT