AI Agent¶

Class 5 AI Agent

Overview¶

Traditional human call center agents are effective in handling inquiries but are constrained by availability, response time, and operational costs.

In contrast, ConnexCS AI Call Center Agents deliver instant, 24/7 support with zero downtime. They automate repetitive interactions, and ensure highly scalable, latency-free communication, reducing reliance on human intervention while maintaining service accuracy and efficiency.

With AI-driven automation, businesses can enhance customer experience while significantly lowering operational overhead.

Use Cases¶

• AI Receptionist:
  • Front Desk Automation
  • Appointment Scheduling
  • Customer Query Handling
  • After hours support
• Conversational Data Collection
• Scheduling call back

Benefits¶

• 24/7 Availability
• Reduced Operational Costs
• Faster Query Resolution
• Improved Appointment Management
• Automatic collection and updation of records

Steps to create AI Agent¶

1. Login to your account.
2. Navigate to Class 5 AI Agent.
3. Click on the blue + icon.
4. A window will open up where you can create your customizable AI Agent for your customers.
5. Fill in the following fields to setup your AI Agent:

Task¶

A Task refers to a specific goal or objective that the agent is designed to accomplish.

It usually involves processing inputs, making decisions, and taking actions to achieve the desired outcome.

flowchart TB
A([📝 Create Initial Task]) --> B([⚙️ Go to Advanced Settings])
B --> C([📌 Set as First Task])
C --> D([🔧 Configure Task Settings])
D --> E([➕ Add Next AI Agent Task])
E --> F{🧠 Is Condition Met?}

F -- ✅ Yes --> G([➡️ Proceed to Next Task])
F -- ❌ No --> H([⏸️ Stop])

G --> I([🔁 Repeat for Next Task])
I --> E

%% Styling
classDef step fill:#E0F7FA,stroke:#00ACC1,stroke-width:2px,color:#004D40;
classDef decision fill:#FFF3E0,stroke:#FB8C00,stroke-width:2px,color:#E65100;
class A,B,C,D,E,G,I,Z step;
class F,H decision;

Steps to Create a Task¶

1. Log in to your account.
2. Navigate to Class 5 AI Agent.
3. Click on Task.
4. Click on the blue + sign (Add next node). It will update the first task.

5. Fill in the following fields to setup the task for your AI Agent:

   1. Basic:

      • Name: Enter the name of your task.

        

   2. Model:

      • Model: Select a model for the selected platform. A model is a specific version or configuration of an AI system to perform tasks like text generation, real-time interactions, etc.

        Info

        Different platforms offer specific models suited to their capabilities.

      • First Message: The initial message spoken by the agent. For example, when the agent begins the conversation.

      • Prompt: The prompt field allows you to enter a set of instructions to guide the AI Agent to perform certain tasks. It provides the foundation for the AI to comprehend the context, interpret the user's purpose, and generate relevant results.

        

   3. Transcriber

      • Transcriber: Select a single or multiple transcribers for converting audio to text. You can select the transcriber depending on the selected platform.

        

   4. Advanced

      • Built-in Functions:
        • Hangup: Use this flag to terminate the call. You can include a prompt such as, "When call ended, hangup.".
        • Set Variable: Set variables locally. Set variables locally. For example, in a prompt, you could ask, "What's your name?" and save the response to the variable name.
        • Transfer: Allows transferring calls to another agent.

      • Tool Call Allow List: Select from a list of functions available in Apps in IDE.

      • Guard Rails: Choose from multiple Guard Rails options to define constraints for your AI Agent.

      • Flags: Select from various flags:

        • Async Guard Rails: Currently not in use.
        • Local Voice Activity Detection: Optimizes VoIP performance by identifying speech in real time at the user’s device.
        • Disable ScriptForge: Disables ScriptForge completely. Enabling this flag will deactivate the Tool Call Support functionality.
        • Listen First: Enabling this flag ensures the LLM prioritizes listening to the user before responding.
        • Vad Start Interrupt: A VAD Start Interrupt is triggered when the system detects the onset of voice activity, indicating that speech has begun.
        • Complete Word on Interrupt: If the AI is interrupted, it will complete the current word before stopping execution.
        • Async First Greeting: When enabled, only the first message of the main prompt will be executed before proceeding directly to the first task. When disabled, the system will follow the default behavior and execute the entire main prompt sequence.
        • Include Primary Chat History: Includes the full conversation history from the primary session, allowing the AI to access prior interactions for better context and continuity.
        • Include Primary Prompt: Appends the initial prompt from the primary task to the current session, ensuring the AI has visibility into the original instruction or objective.
        • Sync Primary Chat History: Enables real-time synchronization with the primary chat session, so updates and responses are reflected across both the original and current contexts.
        • Summarize History on Exit: Generates a concise summary of the conversation when the task concludes.
        • Bill Task Separately: Marks this task to be billed independently from the rest of the session

      • Office Background Volume: This feature enables the playback of background office sounds.

      • Temperature: Refers to a parameter that controls the creativity of the model's output.

      • Low temperature (close to 0) makes the model's responses more predictable by selecting the most probable words.

      • High temperature (closer to 1 or above) increases randomness, leading to more diverse and creative but potentially less coherent responses.

      • Vars^TOML: Select the variables you want pass into the ScriptForge script.

   5. Click Save.

      

   6. To use this created task select it under Advanced First Task.

   Note

   1. If Include Primary Chat History is checked, Include Primary Prompt will be ticked automatically.
   2. If Sync Primary Chat History is checked, Include Primary Chat History and Include Primary Prompt will be ticked automatically.

AI Agent Task Conditions¶

In the AI Agent Task configuration, the Condition field defines the rule that determines when a specific task should be executed.

It allows you to control the flow of conversation by linking one task to another based on user input, recognized entities, or system logic.

• Field Description

  • Condition: The logic that triggers an AI Agent Task. If the condition evaluates as true, the defined task will run.

  • AI Agent Task: Its the current task.

  • Next AI Agent Task: This task will be executed if the condition is met.

• Types of Conditions

  • Default: Executes when no other specific condition is met.

  • Example: If the caller has not given a recognizable input, the AI Agent moves forward with the default task.

• Custom Conditions: Allows branching based on input or context.

  • Common use cases:

  • User Input: If the caller says “yes” or “no.”

  • Entity Recognition: If the system has captured an email address, phone number, or name.

  • System Variables: If a variable (e.g., customerID) exists or matches a value.

• Benefits of Using Conditions

  • Enables dynamic branching in AI conversations.

  • Improves user experience by adapting responses to input.

  • Provides greater control over workflow design.

You can pass variables using JSON in the prompt section to configure or initialize them.

Make sure you select SetVariable under Advanced Built-in Functions Set Variable

AI Platforms, Models, Voices, TTS, Transcribers¶

Chat¶

The Chat interface will provide responses based on the prompts you submit. Each reply will be tailored specifically to the content and context of your input, ensuring relevant and accurate information is delivered efficiently.

To improve the relevance and accuracy of the AI's replies, you can use the Add Context feature.

Before starting, choose a model mode—Standard for fast results (default) or Expert for deeper reasoning with longer response times

How to access the Chat feature?¶

1. Login to your account.
2. Navigate to Class 5 AI Agent.
3. Click on Chat.

4. Click on Start Chat.

   

5. Before initiating interaction with the AI chatbot, you must configure a few initial settings:

1. Add Context (optional field): Allows you to provide additional information to the AI before or during a conversation so it can give more accurate, relevant, and helpful responses. You can either add Text Input or upload text files using File Upload option.

   

   .
   Click Add to Chat.

2. Select the mode of the model:

   1. Standard (default): A fast model focused on quick results.
   2. Expert: A reasoning model, takes longer to respond.
3. Toggle for fullscreen.

How to use Functions for the Large Language Models (LLMs)?¶

To grant access to functions for an AI agent, follow these steps:

1. Create an App from IDE:

   1. Login to your account
   2. Navigate to IDE Application.
   3. Create a new App. Click here to follow the steps.

2. Add a Script Forge: Under the newly created app, add a Script Forge. Click here to know the steps.

3. Define the Function: Within the Script Forge, add the function you want the AI agent to access.

   Example Function

   1 2 3 4 5 6 7 8 9

   /** * Transfer * @param {Object} param - Param Object * @param {string} param.name - The name of the person who you wish to transfer the call to. */ function transfer() { console.log(`We are going to transfer the call`) return { action: 'transfer', destination: '4423456526787365' } }

4. Assign the App to the AI Agent:

   1. Navigate to your AI agent that you have previously created.
   2. Under the Basic in the App field app, choose the created App.

5. Allow Function Access:

   1. Navigate to Advanced tab (in AI Agent), then select the created function in the Tool Call Allow List and grant access from the list.

1
2
3
4
5
6
7
8

/**
* Transfer
* @param {Object} param - Param Object
* @param {string} param.name - The name of the person who you wish to transfer the call to.
*/
function transfer() {
return { action: 'transfer', destination: '160' }
}

Following these steps ensures that the AI agent has the necessary permissions to interact with the designated function.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22

/**
* update the database with user details. 
* @param {Object} param - Param Object
* @param {String} param.last - last name.
* @param {String} param.first - first name.
* @param {String} param.phone - phone number.
* @param {String} param.address - address.
* @param {String} param.from - from
*/
async function updateDatabase(data) { // Declares an asynchronous function named `updateDatabase`, which takes `data` (an object) as a parameter.
    const data2 = await cxUserspace.read('mydatabase12345', data.from) // Reads a record from the datastore `mydatabase12345` using `data.from` as a key
    const area = 'databaseadam'; // Datastore name
    const userId = data.phone; // Unique key for the user record
    const userData = { // Constructs an object (userData) containing user details
        first_name: data.first,
        last_name: data.last,
        phone_code: data.phone,
        address1: data.address,
        middle_initial: data2.value // The middle_initial field is extracted from data2.value, which was fetched from the database earlier
    };
    await cxUserspace.create(area, userId, userData); // Saves the user record in the 'databaseadam' datastore. Uses `userId` (phone number) as the unique key and stores `userData` as the associated record
}

Call Transfer Techniques¶

Calls can be routed to another agent through two different transfer options.

1. Using the built-in Transfer function:

   1. Select Transfer from the Advanced Built-in Functions tab.
   2. Navigate to the Model tab and insert the given prompt in the Prompt section. If requested you please call the transfer function with input phone number as the parameter destination.

   Security Warning

   Implementing this way will give the LLM the capability to transfer to ANY number. If a carefully executed instructions are passed to the LLM via a caller, it MAY be possible to encourage the LLM to dial a DIFFERENT number.

   We would recommend against this method outside testing. Furthermore, it's highly advisable that the routing is additionally protected, to all additional layers of security.

   See below for a recommended strategy to implement transfers.

2. Using Transfer function with Script Forge:

   1. Login to you account.
   2. Navigate to IDE ScriptForge.
   3. The script is as follows:

      1 2 3 4 5 6 7 8

      /** * Transfer * @param {Object} param - Param Object * @param {string} param.name - The name of the person who you wish to transfer the call to. */ function transfer() { return { action: 'transfer', destination: 'input phone number' } }

   4. Navigate to Class 5 AI Agent.
   5. In the Basic tab select the created transfer application from the dropdown in the App field.
   6. Navigate to the Advanced tab and select the created transfer function in the Tool Call Allow List.

How to Enable Reseller Billing for AI Agent Services?¶

Our AI Agent comes equipped with robust reseller capabilities. For instance, if you’ve created and assigned a package to a customer, you can associate their company with the AI Agent. Once this connection is in place, charges for using the AI Agent are automatically deducted. With the package already linked to your customer, managing reselling becomes effortless and efficient.

Integrating Knowledge Base into Your AI Agent¶

1. Create your AI Agent.
2. A few more adjustments are needed:
   1. Select Search Knowledge Base from the Built-in Functions option
   2. Select your Knowlegde Base from the Knowledge Base drop-down.
3. Click Save.

Instructions 
You can call the knowledge base only when you are required to and search for info that you require only 
Only do one 1 lookup per human interaction.
Keep your replies short