ChatGPT assistant workflow model support

With this extension you won't need to pay anything to third party except OpenAI for contextual search based on your personal data. Streaming is also supported!

Demo

You can see ChatGPT chat running on https://doc.livehelperchat.com/ it can answer questions only related to Live Helper Chat documentation. Demo also supports streaming.

We are using for this demo https://livehelperchat.com/order/now if you want easily setup your own bot.

For streaming demo look at https://youtu.be/9clxMjnrGsM

How it works?

• Account at https://platform.openai.com is created
• Background worker is used https://github.com/LiveHelperChat/lhc-php-resque
• Vector store is created and attached to assistant
• To vector store files from https://github.com/LiveHelperChat/doc were uploaded

Present release

• Now you can use streaming workflow. For streaming to work you have to have those extensions up and running.
• NodeJS (https://github.com/LiveHelperChat/NodeJS-Helper)
• PHP-Resque (https://github.com/LiveHelperChat/lhc-php-resque)
• Before sending Rest API call in your bot make sure. Last message is a Send typing message. We use typing message indicator and fill its content with streaming content.
• Most recent versions of Live Helper Chat and NodeJS is required
  • https://github.com/LiveHelperChat/livehelperchat/commit/dfcb0597179d961eef1b922093db5f7679475511
  • https://github.com/LiveHelperChat/NodeJS-Helper/commit/582a273acb1222e7bde6f50e2bfd58079c74b85d
• Bot and Rest API json files you will find here https://github.com/LiveHelperChat/chatGPT/tree/main/doc/assistant_stream
  • Import first Rest API after that import bot and choose just imported Rest API
  • Remember to change where you see {CHAT_GPT_TOKEN} and {CHATGPT_ASSISTANT_ID}

Documentation section regarding how streaming works

https://doc.livehelperchat.com/docs/bot/rest-api#streaming

v1.2

• Added support for empty first visitor message on chat start. Rest API call now uses {not_emtpy_* feature.
• Meta data for a Run is stored within chat, because there can be no message during run. E.g first run on chat start
  • Because of this chat Bot and Rest API calls were modified.

Required version since this - https://github.com/LiveHelperChat/livehelperchat/commit/5b2f0b25404dfe8bf56a15d5031086de3622c496

v1.1

Updates

• Multiple tools call support. ScheduleRun Rest API call will use "parallel_tool_calls": false, argument
• Bot now has example how to have multiple tools calls with arguments

Make sure you have most recent Rest API and Bot versions also version with this commit https://github.com/LiveHelperChat/livehelperchat/commit/6e9ecaa573902adc082204799fec98f7796d81d9

V1.0

Initial release

Requirements

• Min 4.51v Live Helper Chat version.
• At the moment Polling workflow is supported only.
• Account at https://platform.openai.com/docs/overview
• System configuration -> Live Help Configuration -> Chat configuration -> Misc Make sure you unheck Reopen chat functionality enabled and Allow user to reopen closed chats?. This is required because if chat is reopened we can be in the middle of previous task.

How to make UI snappy by delegating Rest API calls to background workers?

• Install https://github.com/LiveHelperChat/lhc-php-resque
• Modify Bot triggers and where you see Rest APi call response check Send Rest API Call in the background if we are not already in it.

Install

After clone or download put it under

extension/chatgpt - don't forget to lowercase a folder

Activate extension in main settings file lhc_web/settings/settings.ini.php extension section chatgpt by Adding lines:

'extensions' =>  array (  'chatgpt'  ),

Install database

Execute queries from

doc/install.sql

Or execute command

php cron.php -s site_admin -e chatgpt -c cron/update_structure

Back office

Navigate to back office and click clear cache

Under left Modules you will find ChatGPT click it.

Setup as a bot

• Click ChatGPT Bot integration settings
• Now just paste Project API Key and Assistant ID from https://platform.openai.com and click Create/Update Rest API/Bot

Now you can just assign newly created bot to your department or modify bot to the way you want.

How to avoid bot calling functions if visitor is not logged in?

In a ScheduleRun Rest API call you can have something like {is_empty__args.chat.chat_variables_array.is_logged} which checks that chat variable is set.

{
    "assistant_id": "asst_UAJYImd9WyXlRGNOa0bcTAgD",
     "parallel_tool_calls": false
     {is_empty__args.chat.chat_variables_array.is_logged}
        ,"tools":[]
        , "additional_instructions":"Visitor is not logged in and functions calls are not enable to him. You can answer questions only from documentation. Ask him to login to get personal account information. You can get personal information once visitor is logged in."
     {/is_empty}
     {not_empty__msg_url},"additional_messages" : [{"role" : "user", "content" :  {{msg_url}} }]{/not_empty}
}

Setup pas reply predictor for the agents

• Click ChatGPT Setting for answers suggesting
• Project API Key and Assistant ID can be different than Bot
• Fer reply predictions to work you have to activate bot from Setup as a bot step. We use some of the Rest API calls.

How to have only manual reply predictions tab in the chat interface?

Have only Enable reply prediction tab in chat UI checked.

How to have only manual reply predictions tab in the chat interface?

Have Automatically suggest answers based on visitor messages

Setup video tutorial

Video tutorial and use case - https://youtu.be/X9W99obVj8Q

ChatGPT Completion workflow model support

Please refer to this manual https://doc.livehelperchat.com/docs/bot/chatgpt-integration

Screenshots

Multiple tool calls

Update phone number.

Phone number can be evening and morning as an example.

Chat GPT function definition

{
  "name": "post-update-phone-number",
  "description": "Updates the phone number for a player.",
  "strict": true,
  "parameters": {
    "type": "object",
    "required": [
      "phone",
      "type"
    ],
    "properties": {
      "type": {
        "type": "string",
        "description": "The type of phone number (e.g., 'morning','evening').",
        "enum": [
          "evening_phone",
          "morning_phone"
        ]
      },
      "phone": {
        "type": "string",
        "description": "The new phone number to be updated."
      }
    },
    "additionalProperties": false
  }
}

Accessing function call arguments in the bot triggers.

{args.chat.chat_variables_array.chatgpt_action_args.phone}
{args.chat.chat_variables_array.chatgpt_action_args.type}

Accessing function call arguments in the Rest API

{
    "phone":  {{args.chat.chat_variables_array.chatgpt_action_args.phone}},
    "type":{{args.chat.chat_variables_array.chatgpt_action_args.type}}
}

Live Helper Chat

Retrieve withdrawal list for the visitor

Chat GPT function definition

{
  "name": "get_withdrawals",
  "description": "Retrieves a list of withdrawals for a visitor.",
  "strict": true,
  "parameters": {
    "type": "object",
    "required": [],
    "properties": {},
    "additionalProperties": false
  }
}

Live Helper Chat side. You can also as visitor can ask to return only last 3 items.