Support Agent is an AI-powered assistant designed to understand user queries and retrieve relevant knowledge base information. Before deployment, you must first create and configure your agent in the InOne panel. This process involves defining its personality and connecting relevant knowledge base sources. These foundational elements work together to ensure the agent responds accurately, contextually, and in line with your brand’s tone.
This guide aims to answer these questions:
Requirements
To successfully create and deploy a Support Agent, you must ensure the following foundational components are in place:
User Roles: Only users with Administrator or Editor roles can create, edit, and delete a Support Agent in InOne.
Read User Roles for further information.
MindBehind Company and InOne Account (Mandatory): Using Support Agents requires both access to InOne and provisioned credentials for the MindBehind platform. The Insider One team will grant access to both environments. This connection is essential, as chatbot deployment takes place within the MindBehind platform.
Personality Configuration: Each Support Agent must be assigned a personality that defines:
Communication style and tone
Level of formality
Preferred response length and format
Overall purpose or intent
This ensures consistent, brand-aligned communication across interactions.
Knowledge Base: A structured knowledge base is required for the agent to provide accurate responses. This is generated by crawling a specified website, where the agent learns relevant content. Admins can customize crawl settings to target the specific information they need. Without a knowledge base, the agent cannot generate informed answers.
Once these requirements are met, follow the steps below to set up your Support Agent:
Step 1: Create your Knowledge Base
The Knowledge Base (KB) is the primary information source for Support Agents. It enables agents to deliver accurate, real-time responses by referencing content directly extracted from your website.
Support Agents do not use static or hardcoded answers. Instead, they dynamically generate replies based on the KB’s content, enabling flexibility and relevance across a wide range of user queries.
To learn how to set up your Knowledge Base, refer to Create a Knowledge Base for Agents and continue with Step 2 below.
Step 2: Create your Support Agent
The Support Agent is designed to help you quickly configure a conversational AI assistant tailored to your specific needs. After creating your Knowledge Base, you should set up your Support Agent.
Support Agent can be configured in any language supported by the language model, but it performs best in English.
Follow the steps below to create your Support Agent:
Navigate to InOne > Agent One > AI Agents.
Click the Create button located at the top right.
Select Support Agent as the agent type and enter a name for your Support Agent.
This name is used for internal reference only and won’t affect how the agent behaves or communicates. It’s simply to help you identify and organize your agents within the platform.
In the Personality page, you can define the agent’s personality. This determines how the agent interacts with users and presents itself in conversations.
In the Agent Identity section, you define who your agent represents and how it introduces itself to users. This identity shapes the agent’s tone and behavior during customer interactions.
Refer to the Best Practices for Support Agent guide to create effective identities.
In the Communication Style, you can define how your agent should communicate with users. You can choose your agent’s tone of voice from the following options:
Friendly
Neutral
Direct
Formal
Humorous
You can also specify the message length preference for the agent’s responses:
Concise: Short and to the point
Standard: A balanced level of detail
Detailed: More comprehensive and elaborate replies
In the Custom Instructions, you can provide specific guidelines to tailor your Support Agent’s behavior further.
You can use it to define:
How to respond to sensitive topics
Words or phrases to avoid
Preferred terminology or phrasing
What to do when a user’s message is unclear or ambiguous
Refer to the Best Practices for Support Agent guide to learn how to create custom instructions for your Support Agent.
These settings provide you with complete control over the agent’s tone and behavior, ensuring it aligns with your brand voice and support guidelines.
To save your changes, click Save and Continue.
Test AI Agent Panel
You’ll find the Test AI Agent panel on the right side of the screen while editing your agent. It allows you to interact with the draft version, preview it, and see how it responds in real-time.
Triggering a test will automatically save any unsaved changes to the draft.
What can you test in the Test AI Agent panel?
In the Test AI Agent panel, you can test:
Tone and personality
Knowledge base responses
Actions and output behavior
Test preview focuses on the Support Agent only. It doesn’t reflect full MindBehind chatbot flows such as fallbacks, routing, or handovers. For those, you must use the MindBehind panel.
After each input you submit, the agent’s response appears along with valuable details to help you understand the reasoning behind it:
Sources: Shows which knowledge base source contributed to the response. To view the specific URL used during response generation, hover over a source.
Actions: If the response triggered an action, you’ll see:
A debug ID
The parameters that were used
The values passed to the action
This transparency helps you pinpoint whether a response came from the correct source and whether actions are functioning as intended. The Test AI Agent panel is your go-to tool for validating the accuracy, functionality, and quality of your Support Agent before publishing it live.
Step 3: Link the Knowledge Base
Linking a Knowledge Base allows the Support Agent to generate accurate and context-aware responses based on your website content. It serves as the primary source of information for the agent’s answers.
To link a Knowledge Base, you must first create one, as explained in Step 1.
Each Support Agent can be linked with up to 4 Knowledge Bases (KBs). Similarly, a maximum of 4 KBs can be created per account.
Before linking a Knowledge Base to a Support Agent, make sure its status is set to “Active.” Knowledge Bases in “Processing” or “Error” states can’t be assigned and won’t appear in the selection list. Always verify that the crawl is complete and the Knowledge Base is finalized before connecting it to an agent.
Select the Knowledge Base you want to link, click Save and Continue.
Step 4: Create Actions
Actions are predefined operations your AI Agent can trigger in response to specific user intents. They allow your Support Agent to either exit the conversation or perform external tasks via APIs.
Each agent can include up to 17 actions in total, and must have at least one Exit Action.
In the Actions tab, you'll see a list of all actions you've previously created. If you haven't created any actions yet, the list starts empty. To add a new action,
Click the Create Action button.
Select your Action type, enter your Action name, and click Next.
API Call Action
If you select the API Call Action, you’re enabling your agent to perform an external operation, like checking an order status, submitting a request, or retrieving data from another system. For example: A customer says, “I want to check the status of my order.” → The AI Agent recognizes the intent and triggers an API call to fetch the order status automatically.
After selecting the API Call Action, you should fill in the required fields:
In the Action Prompt section, you need to define the triggering intent or question that will initiate the action. For example: "This action should be triggered when a user asks to retrieve information related to the status of their recent order, such as 'What's the status of my recent order?' or 'Can you check my delivery update?'"
In the Parameters, you can collect necessary data from the user (same as the Exit Action structure). For example:
Name: order_id
Type: String
Description: "The order number shared by the customer"
In the API Endpoint section, you should define the full URL of the endpoint to call. For example: https://api.example.com/orders/status
Method: GET, POST, DELETE, PATCH, PUT
In the Headers tab, you can:
Define key-value pairs (e.g., Authorization: Bearer token)
You can use static values or dynamic parameters collected from the user.
Use the Add Parameter button to dynamically bind parameters to header values.
In the Body tab, you can define the JSON body to send with the request.
Both static content and parameterized values are supported.
Enter Key / Enter Value: Located under the Body section, this setting allows you to define additional key-value pairs that will be included in the API request body.
These fields are required if the API expects them. Values can be defined statically or dynamically using collected parameters.
In the API Call Action type, both the Action Prompt and Endpoint URL fields are mandatory and must be filled in to proceed.
Exit Action
If you select the Exit Action, the agent will end the conversation or transfer it to another system—such as a live agent or an external flow—because it cannot or shouldn’t handle the request further. For example, if a customer requests to speak with a human, the AI Agent should exit and transfer the conversation to a MindBehind Flow for live support.
After naming the Exit Action, fill in the required sections:
In the Action Prompt, you can describe the type of user message or intent that should trigger this action. This helps ensure the agent responds accurately when similar phrases or requests are detected. For example: "This action should be triggered when a customer asks to speak with a human or requests live support."
In the Exit Action type, Action Prompt field is mandatory and must be filled in to proceed.
In the Parameters, you can define the key pieces of information the agent should collect before executing the Exit Action.
Each parameter includes the following:
Name: A specific name for the parameter (e.g., userlanguage) (This will be stored as a bot parameter in MindBehind Flow. During runtime, the system automatically adds the agentone_ prefix (e.g., userlanguage becomes agentone_userlanguage).
Type: The format of the data to be collected and the options include: String, Number, Integer, Boolean, or Array
Description: A brief explanation of what the parameter is used for.
Optional: Enable this setting if the parameter is not required for the action to proceed.
Here is an example of a human handover scenario with the Exit Action:
Field | Value |
|---|---|
Name | userlanguage |
Type | String |
Description | "User’s preferred language for live support" |
Flow Output | agentone_userlanguage |
Step 5: Test and Update Support Agent
When you're editing, a clear indicator on the interface will show whether you're working in Live or Draft mode. .png)
The Draft version of the Support Agent feature lets you test updates to your Support Agent—such as changes to its configuration, knowledge base, or actions—without affecting the live version. This provides a safe space for you to experiment before sharing anything publicly.
Live vs. Draft
Currently, the version history or version snapshots are unavailable. Only two states exist for each agent:
Live version: This is the version currently in use. MindBehind chatbots only operate with the live version, so any changes here will directly affect your chatbot.
Draft version: A working version where updates are saved and tested before being published.
Once changes are saved to the draft, you cannot roll back to a previous draft. However, if the specific agent has already been launched and you're editing it, the Restore Active Version button will appear. Clicking this button resets the draft to match the current live setup.
Step 5.1: Test your Draft
To store changes in your Support Agent, click the Save as Draft button. Alternatively, you can type your test message in the Test AI Agent panel to test and preview it and to instantly see how the agent behaves with the current draft configuration.
You can test changes:
Individually, right after editing
Or together, after completing all updates
This flexible testing workflow allows you to fine-tune the agent incrementally or all at once, depending on your team’s process.
If you're not satisfied, click the Restore Active Version button to discard the draft. This button becomes visible only if the specific agent has already been launched and you're editing it.
This two-version system helps you safely test and improve your Support Agent, without disrupting real user conversations until you’re ready.
Once a Support Agent is published, it becomes part of the Live version. Any changes made to the Live version will immediately affect the behavior of your chatbot in production.
Step 6: Deploy Support Agent
Before deploying a Support Agent, make sure it’s fully configured and has an Active status in your InOne panel. Only active agents can be used in chatbot flows.
Deployment involves two main steps:
Embed the Support Agent into a chatbot using MindBehind (Using the Support Agent Module).
Once your Support Agent is launched, use the credentials shared by the Insider One team to log in to the MindBehind platform. There, you can insert the agent into a chatbot flow and publish it across supported communication channels.
This setup allows your Support Agent to start handling real conversations in live environments.
Step 6.1: Launch the Support Agent
Once you’ve completed all configuration and testing steps for your Support Agent, your final stage is to visit the Launch page.
On this page, before clicking the Launch button, you can:
Compare the current draft version of your agent with the version that is actively live. This helps you review the exact differences before going live.
Identify missing elements from the agent creation process. Any required or recommended fields that have not been completed will be highlighted for your attention.
Ensure readiness for deployment through a review of both functional changes and structural completeness.
When you click the Launch button, all changes will go live instantly. This means the Support Agent will be updated in every MindBehind chatbot where it’s being used. Note that active conversations will not be affected by the update, but any new conversations started after the launch will reflect the changes.
Step 6.2: Use the Support Agent Module
This step should be handled into the MindBehind platform. Make sure you have access to the platform.
The Support Agent (LLM) module allows you to integrate a Large Language Model (LLM)-powered support assistant into your conversational flow. It leverages your predefined Support Agent configurations, including personality, knowledge base, and actions, to deliver intelligent and contextual responses to user inputs.
Log in to the MindBehind Flow application and select your assistant, who will use a support agent.
From the left menu, drag and drop the Use Support AI Agent (LLM) module onto your canvas (map).
Click the added module and select your InOne Account, which is linked to your company.
After selecting the account, pick a previously created Support Agent from the list.
Once you select your Support Agent, you must configure the Exit Action connections. These are the actions predefined in the agent that represent tasks the agent cannot handle on its own. For example, if a request requires human intervention, define an exit action such as “Human Handover” and connect it to the proper flow path.
Ensure that every Exit Action has a valid connection on the canvas.
You must define at least one fallback connection for the Support Agent module. Without a fallback, the agent will be unable to recover from unrecognized inputs, and all user messages might be routed to the fallback unintentionally.
The minimum fallback count must be set to 1.
Support Agents generate responses based on the last input received. For accurate behavior, always insert an Input Module right before the Support Agent module.
If a user previously clicks a button (e.g., "Connect to AI Agent"), that selection will be treated as the last input, and the agent might attempt to respond to that instead of a meaningful query. For example, if the last input is "Connect to AI Agent", the agent may interpret that as a message and generate a reply.
Support Agent modules always operate on the most up-to-date version of your agent. Any changes to the configuration, such as updates to the knowledge base, actions, or personality, are instantly applied in live environments.
Congratulations! Your Support Agent is now up and running.
If you encounter any issues or unexpected behavior, the following troubleshooting tips can help you resolve them quickly.
Troubleshooting for the Support Agent Module in MindBehind
No fallback count defined
When the fallback count is missing in the Support Agent module, the system has no limit for handling unrecognized inputs. As a result, every message from the user is immediately routed to the fallback path.
Solution: Set a fallback count of at least 1 in the module settings to define how many failed attempts should be allowed before triggering fallback.
User input not properly captured
If a user’s selection, such as a button click, is treated as the most recent input, the agent might respond in an unintended or irrelevant way.
Solution: Add a clear Input Module just before the Support Agent module to capture and define the user’s intent.
Exit actions are not connected
If an exit action configured in the agent module isn’t linked to the next step in the flow, the conversation might end unexpectedly or be routed to fallback.
Solution: Review the canvas and make sure each exit action is connected to the appropriate module or step.
The test flow doesn’t reflect the latest agent changes
When updates are made to the Support Agent (e.g., knowledge base, actions, or tone), existing chatbot flows might still reflect outdated behavior.
Solution: After making any changes to the agent, re-test the chatbot flow to ensure everything works as expected.