AI Prompts

On this page

AI Prompts section available from version 3.34.0. This documentation is work in progress and some features will be available in next releases.
From version 3.35.0 it’s also possible to use Assistants within the ChatGPT bot (in this case you need to enable on License AI and Assistant feature)

📋 What’s About

AI Prompts is the section on which you can configure OpenAI Assistants, designed to understand and answer to human queries, by starting from business files uploaded in the relative section.

Assistants are OpenAI models that can use tools, in our case reading uploaded files, to find answers for customers.
Explore how it works on Official documentation: Assistants overview | How Assistants work

OpenAI Assistants are in Beta version

Requirements

⚙️ Configuration

Assistants

You can define OpenAI Assistants both on XCALLY side and on OpenAI console. In both cases, assistants are registered on OpenAI (not on Motion database).

To create a new Assistant, you can click on Add assistant and insert:

Assistant Name
Cloud Provider (OpenAI type)
Model (gpt-4-turbo, gpt-4o, gpt-3.5-turbo, gpt-4o-mini)
Instructions, these are the bot training “instructions” to explain what the assistant has to say, questions it can manage, phrase it has to say when it doesn't know how to help the customer and so on.

If you want to manage the re-routing from to bot to a human agent, it is necessary to explicitly include this key phrase literally in the instructions:
If you don't know how to help user, tell the customer "I am redirecting you to a human operator".
If customer asks to be redirected to human operator, tell the customer "I am redirecting you to a human operator".

If you want that assistant uses files, you need to make reference in instructions to title file. E.g. you can write: If the customer asks about refund issues, search answer in the file “Refund instructions”

And click on Save

The created Assistant is saved also in Open AI platform under Assistants section

Files

So if you want to use business documents, on Files section, you can upload your files (for example pdf instructions, user manuals and so on).
Read on official OpenAI documentation, the supported files Open AI | Supported Files

You can here view an example of how to create a file:

So you can click on Files → Add file

And then select OpenAI Cloud Provider and upload Files from your computer → and click on Confirm

So in this case, by editing a created assistant, you can see files section to associate files to the assistant. By clicking on Files section you can add the relative files, among those uploaded.

In this way bot has an associated file.

Starting from version 3.40.0 it’s also possible to download an added file, by clicking the download button, but consider that files uploaded prior to this version are not downloadable

🤖 How to configure Assistants in OpenAI Bots

On OpenAI bots section, it is possible to configure the assistant so that the bot can retrieve useful customer information from uploaded files.
For example if you create an Assistant with files related to product return and the customer asks for this type of information, the bot provides an explanation, starting with the uploaded documentation.

So by creating a new OpenAI bot, you have to fill in the form with:

OpenAI Cloud Provider account
ChatGPT Assistant: from this section you can choose among the created assistants, associated to the cloud provider key
Bot Name
Exit Phrase
Welcome message
Option to analyze chat to know if pass the chat to an agent
Forward message
Error message
ChatGPT Attachment message

As you can see, when you use Assistant you can’t enter chatGPT model, prompt (so instructions), max conversation lenght in tokens, temperature because these options are already inserted in OpenAI Assistant configuration (as you can see below)
Follow official documentation Prompt engineering about how to properly create a prompt
From the section OpenAI Usage you can track costs by token usage

When you use Assistant, it is recommended to insert a queue after bot management so that customer service is forwarded to a human operator

So when for example a chat starts, the bot answers to customer with information that it can retrieve from files:

And agent will be able to see the interaction, in the relative channel interactions list.