🤖Creating and configuring AI | AI

The “AI” section is intended for creating and editing artificial intelligence. This section displays all the AIs you have created, their types, and the tasks in which they are used.

Choose how to get the information: a video tutorial (up to timestamp 07:55) or the text with images below.

Video on DzenVideo on YouTube.

VK Video.

General view of the section

In the Selarti personal account go to the AI.

Main elements

The central part of the screen displays a list of all created AIs.

  • Name — the name of your AI;

  • Type — the selected type of communication (“Incoming communication” or “Incoming and outgoing communication”);

  • Used in — lists the tasks in which the AI is used;

  • Actions — buttons to copy and delete the AI.

Creating an AI

Go to the AI, click on “Create AI” in the top right.

A window for creating an AI will open, in which you need to fill in the fields:

  • Name — specify the AI name so you can identify it;

  • Select GPT version — from the dropdown list choose the version that fits your goals. We recommend GPT-4o. More about their differences in the article;

  • Service operation format — choose one of two types of communication with the client:

    • “Incoming communication” for replying only to incoming messages;

    • “Incoming and outgoing communication” for sending outgoing messages and replying to incoming ones.

Additional settings for “Incoming and outgoing communication”

When this item is selected, an additional field will appear — “Start message”, the initial phrase with which the AI will begin the dialogue with the client. For a productive dialogue we recommend starting it with a greeting and a brief explanation of why the manager is writing to the client.

For VK, Telegram, WhatsApp channels we recommend creating variation in the start messages so the account is not blocked for mass messaging. For this use spintax and the construct {Message 1|Message 2|Message 3}. This way the AI will randomly choose the message to start the dialogue.

Example starting phrases:

{Good afternoon{!|)|👋|😊} I found you in the Marketing for Business group. {Tell me, are you interested in discussing sales automation using AI|Tell me, are you interested in sales automation using AI}?|Greetings{!|)|👋|😊} I found you in the Marketing for Business group.{Tell me, are you interested in discussing sales automation using AI|Tell me, are you interested in sales automation using AI}?|Hello{!|)|👋|😊} I found you in the Marketing for Business group. {Tell me, are you interested in discussing sales automation using AI|Tell me, are you interested in sales automation using AI}}

For Iframe and Page Messenger channels variation at the start of the dialogue is not yet provided. However, you can make the dialogue begin with three messages using the construct {Message 1&&Message 2&&Message 3}. These messages will be sent sequentially.

  • Response delay in seconds — the time the AI will “think” after the client's message. If the client sends additional messages during this pause, the AI will process them in one combined reply within a single context, instead of replying to each message separately. This setting also allows virtual managers to seem more human (it’s hard to believe you’re talking to a human if you receive a whole paragraph in response within a couple of seconds);

  • Creativity — adjustable with a slider, affects the AI’s ability to imagine and invent answers. Setting creativity too high can lead to so-called “AI hallucinations,” incoherent text, text in multiple languages simultaneously, or other variants. We recommend setting creativity to 0.5 if you don't know which to choose. More about the parameter in the article;

  • Diversity threshold — adjustable with a slider, affects the diversity of AI responses. The higher the diversity threshold, the more unique the virtual managers’ responses, but be careful — too high a value can lead to “AI hallucinations.” If you don’t know what to set, choose 0.5;

  • Instruction — also called a prompt, this is the most important stage of creating an AI. Write the instruction as clearly and fully as possible, in the language the managers will use to communicate with clients. It is important to give the AI a clear technical task, define its role, task, experience, even its name. Before writing, read the article about writing instructions for AI.

For some AI models there are no “Creativity” and “Diversity threshold” settings; there are alternative additional model settings similar in functionality. More in the article about models.

Using the “Click to upload knowledge base” button make your model even smarter. You can upload up to 100 files with useful information that the AI can use in dialogues. For example, you can upload: sales scripts, phone conversations converted to text, dialogs between managers and clients, presentations, knowledge bases. Supported formats: TXT, DOC, DOCX, PDF, HTML, MD.

The minimum size of an uploaded file is 1 kb, the maximum is 50 MB.

After filling in the fields click “Create AI”.

Tag settings

Tags are needed so that the AI can automatically label dialogues and you can easily find them in the personal account.

Click on the created AI, and in the opened settings window on “Professional settings”, to configure tags.

Choose the required tag from the dropdown list, describe in which cases the AI will use it and click “Add function”. More about configuring and using tags in the article.

Testing AI

Test the AI to make sure it works correctly. To do this open the created AI, click on “AI Testing”.

A dialog with the AI will open on the right, and in the top right corner of the dialog there are three buttons — to share, clear and restart the dialog and to close the testing mode.

Write your question in the input field “Enter your message” and click “Send”.

If you are not satisfied with the bot's answer, you can edit the instruction and click “Save”, and then continue testing.

Editing AI

To make changes to an existing AI click on it, make the necessary changes in the opened editing panel and click “Save”.

Last updated