@doc-gpt/chat NPM

Doc GPT - Chat SDK

Typed SDK for the OpenAI Chat Api

Use the new OpenAI Chat Api /v1/chat/completion with ease.

Stream request are supported

Note: OpenAI Chat Api is still in beta, and may change very quickly. If you notice some changes in the api that has not been implemented, please open an issue or propose a PR.

For more informations about the underlay api, please refer to OpenAI Chat Api Docs

Install

The library is avaliable both on NPM and GitHub Packages.

Install from NPM

npm i @doc-gpt/chat

Install from GitHub Packages

First of all, you need to add a .npmrc file, in the root of your project, with the following content:

@doc-gpt:registry=https://npm.pkg.github.com

Then install the package

npm install  @doc-gpt/chat@1.1.7

Getting started

DocGptChat offers a simple api to interact with OpenAI Chat Api. You just need to configure the instance and start chatting with the methods avaliable:

Chat(messages, options)
SimpleChat(messages, options)
ChatStream(messages, options, onMessageDelta, onDone, onError)

Note: If you use a static api key, it's better to use an environment variable.

Import

import DocGptChat from '@doc-gpt/chat';

Configure the instance

const gpt = new DocGptChat({
  // Required,
  apiKey: OPENAI_API_KEY, // If static, it's highly recommended to use environment variable
  // Optional
  defaultModel: GptModels['gpt-3.5-turbo-0301'], // or simply as string (default gpt-3.5-turbo)
  // Optional
  defaultSystemMessage: 'You are ChatGPT...',
});

Simple usage example:

If you just need the first message content from the response, it's avaliable the SimpleChat method, which wraps Chat with the same parameters and returns the first message string.

try {
  // Get a full response from the api
  const message = await gpt.SimpleChat([
    {
      role: 'user',
      content: prompt,
    },
  ]);

  // Use the message
} catch (err) {
  // Handle Api errors
}

Basic usage example:

If you just need the full response from the api, it's avaliable the Chat method.

try {
  // Get a full response from the api
  const res = await gpt.Chat([
    {
      role: 'user',
      content: prompt,
    },
  ]);

  // Handle the response

  // For example:
  // Get the first message object
  const firstMessage = res.choices[0].message;
} catch (err) {
  // Handle Api errors
}

Stream usage example:

If you want to use stream response, you can use ChatStream method.

// Message to increment
let message = '';

// Get a full response from the api
gpt
  .ChatStream([
    {
      role: 'user',
      content: prompt,
    },
  ])
  .onMessage((messageDelta, fullResponse) => {
    // Increment the response message
    message = message + messageDelta;
  })
  .onDone(() => {
    // Use the full message
    console.log(message);
  })
  .onError((err) => {
    // Handle errors
  });

Methods

Chat

  // Get a full response
  public async Chat(
    messages: GptMessage[],
    options?: GptChatOptions
  ): Promise<GptResponse>

SimpleChat

  // Get the first choice message content.
  public async SimpleChat(
    messages: GptMessage[],
    options?: GptChatOptions
  ): Promise<GptResponse>

Types

All the avaliable message roles

// Message Roles
type GptRole = 'system' | 'user' | 'assistant';

Chat Message

// Chat Message
interface GptMessage {
  content: string;
  role?: GptRole;
}

Response

// OpenAI `/v1/chat/completion` api response
interface GptResponse {
  id: string;
  object: string;
  created: number;
  choices: GptResponseChoices[];
  usage: GetResponceUsage;
}

// Response Choice
interface GptResponseChoices {
  message: GptMessage;
  index: number;
  finish_reason: string;
}

// Response Choice structure with stream mode
export interface GptResponseStreamChoices {
  delta: GptMessage; // delta instead of message
  index: number;
  finish_reason: string;
}

// Response Usage
interface GetResponceUsage {
  prompt_tokens: number;
  completion_tokens: number;
  total_tokens: number;
}

Utils

There is a const named GptModels you can import in order to refer to avaliable models. If a new model is release, and you want to try it, you can still pass any string to model option.

export const GptModels = {
  'gpt-3.5-turbo': 'gpt-3.5-turbo' as GptModel,
  'gpt-3.5-turbo-0301': 'gpt-3.5-turbo-0301' as GptModel,
};

// Example
const model = GptModels['gpt-3.5-turbo-0301'];