How to Create and Manage AI 2.0 Agent in BoldDesk
BoldDesk AI 2.0 is an advanced AI-powered virtual assistant that autonomously handles customer interactions across multiple channels and tickets quickly and accurately, reducing response times while maintaining high-quality support.
With the ability to support multiple AI agents tailored to different use cases, BoldDesk offers a scalable solution for modern support teams. These agents can automate responses based on user actions, assist human agents with real-time suggestions, and improve overall efficiency and engagement.
This guide walks you through the complete process of creating and configuring an AI 2.0 Agent in BoldDesk. It covers key steps such as agent creation, training, customization, channel mapping, managing multiple agents, testing, and setting permissions, helping you streamline support operations and deliver faster, more consistent customer experiences.
Creating an AI 2.0 Agent
To create an AI 2.0 Agent, follow the steps below;
- Click on AI module.
- Click on AI Agents
- Click on Create AI Agent.
- Use the search icon or brand filters to quickly find AI Agents. Selecting edit or view conversations will take you to the AI Agent configuration page, where you can manage instructions, libraries, tools, and other settings.
- The AI Agent creation page is your main workspace for designing, configuring, and deploying agents. To manage an existing agent, click More options to edit, deploy, or view conversations. You can also use the edit icon to update the agent’s name, avatar, or description, and the refresh icon (next to it) to update the agent’s content.
-
You will be redirected to the AI 2.0 Agent creation page.
-
In the dialog box, provide the following information. AI Agent name, description, widget settings and profile image are shared across all versions.
Field Description Avatar Represents the visual icon or image for the AI agent/bot. This helps users visually identify the bot in chats or interactions. The edit (pencil) icon suggests you can upload or change the avatar. This is optional. Name The unique name of the AI bot or assistant. This is what users and agents will see when interacting with it (for instance, “Support Bot”, “Help Assistant”). The asterisk (*) indicates it is a required field. Description A short explanation of what the AI bot does. This helps administrators and users understand the bot’s purpose (for example, “Handles customer queries about billing and orders”). This field is also required. Model The AI model that powers the bot. This determines how the bot understands and responds to queries (for example, GPT-based models or other supported AI engines). Selecting the appropriate model impacts performance, accuracy, and capabilities. You have the following AI models to choose from;
Supported AI Models Provider Model Names Azure OpenAI GPT-5.2, GPT-5.4, GPT-5.4 Mini and GPT-5.4 Nano. -
After completing all the required fields, click Create to finish. You will be redirected to the configuration page.
AI 2.0 Agent Configuration
This is a parent section grouping all setup options required to configure the AI agent’s behavior and capabilities. After creating the agent, configure its behavior and capabilities.
Instructions
This is a critical section that defines how the AI behaves, including tone, rules, and responsibilities (for instance, how to handle bookings and refunds) that directly influence the AI’s responses.
You can quickly generate instructions by leveraging AI assistance.
- Click on Generate using AI to generate the instruction from a brief description you provide to quickly fill the instructions field.
- Click Save after making changes.
Libraries
Add knowledge sources such as files, websites, or knowledge base content. The AI uses these as references to give accurate, context-aware answers.
- Click Link Library to attach sources.
- Use unlink to remove them.
Explore Libraries.
AI Tools
This is used to connect the AI agent to external systems or services (via APIs or servers). Link MCP Servers and AI Actions to build powerful AI Agents. You can create both MCP servers and API Actions under AI tools in the AI Agent creation page. Only the created tools will be available for selection for each AI Agent. Learn how to Configure External Connections for AI.
MCP Tools
This enables real-time actions like fetching booking data or processing refunds. Click Link MCP Server to add MCP server tools. Explore How to Create and Configure an MCP Server in BoldDesk.
API Actions
AI Actions are tasks or operations carried out by an AI system to achieve a specific goal. Click Link API Action to add API Actions. Explore How to Create and Publish API Connections in BoldDesk.
Settings
A setting refers to a configurable option in the AI Agent creation page that controls how the AI behaves, responds, and interacts with users. You can optimize performance by adjusting parameters, module selection, and other settings.
Configurations
You can customize the following under configurations.
Module Configuration
Here you can set the following settings if you choose Gemini AI model. Switching to a different AI model will give varying configurations.
| Field | Description |
|---|---|
| Model | Choose a reliable and balanced AI model that supports strong reasoning and conversational abilities. This ensures the agent can handle booking queries, policy explanations, and refund processes accurately. |
| Max Output Tokens | Set to a moderate range (300–800) so the AI can provide clear, complete explanations (for instance, refund steps or booking details) without being too lengthy. You can reset it by clicking reset to default. |
Trace Logging
This section logs the selected AI execution details, such as LLM inputs and tool interactions. It enables full logging of AI activities for debugging and monitoring. It is categorized further into LLM Tracing and Tool Tracing.
LLM Tracing
-
Log LLM Inputs
Captures all messages sent to the AI (customer, agent, system context). -
Log LLM Outputs
Records AI-generated responses and any tool/action calls.
Tool Tracing
-
Log Tool Schemas
Logs definitions and parameters of tools/actions available to AI. -
Log Tool Responses
Captures results returned by tools (e.g., ticket updates, API outputs).
Context Management & Reasoning
Context Management & Reasoning in a BoldDesk AI 2.0 configuration defines how the AI retrieves past context and controls its reasoning depth and response style. You can configure the following under context management.
| Configuration | Description |
|---|---|
| History message limit | Limits how many previous conversation messages are retained |
| Number of Chunks | Controls how many retrieved knowledge chunks are passed to the model |
| Reasoning effort | Determines how much internal reasoning the AI performs (options: none, low, medium, high, extra high) |
| Reasoning verbosity | Controls how detailed the reasoning/explanations are (auto, concise, detailed) |
Response & Execution Controls
These controls define how the AI behaves when responding, handling errors, and limiting execution depth in BoldDesk. It is used to control the following settings.
| Setting | Description |
|---|---|
| Grounded response only | Limits AI responses to only available knowledge; uses fallback if none found. You can toggle switch it on or off. |
| Custom error message | Message displayed when an AI/system error occurs |
| Fallback instruction | Message returned when no relevant answer is found |
| Max iterations | Controls how many processing attempts the AI can make before stopping |
Tool Configuration
These settings control external data access, allowing the AI in BoldDesk to use URLs and live web data to provide more accurate and up-to-date responses.
| Tool Setting | Description |
|---|---|
| URL Context Tool | Enables the AI to extract and use information from provided URLs as context |
| Web Search | Allows the AI to fetch real-time information from the internet to improve responses |
Response Format
This setting controls how AI responses are formatted, depending on whether the output is for human readability (Text) or system processing (JSON / Structured).
| Format Option | Description |
|---|---|
| Text (Default) | Returns natural language responses suitable for conversations |
| JSON | Outputs responses in structured JSON format for system/API integrations |
| Structured output | Provides predefined, schema-based responses for consistent and formatted data |
Selecting the edit option for the AI 2.0 Agent in BoldDesk will redirect you to the AI Agent configuration page. Here, you can set up the AI 2.0 Agent Configuration as discussed above.
Managing AI 2.0 Agent
Once you have created and configured the AI 2.0 Agent, you can publish, deploy, and view conversations for each AI Agent created.
Publishing AI 2.0 Agent
Once you have customized your AI Agent, you can proceed to the playground on the right pane to test your AI agent before deploying it. Publishing constitutes the final phase in making your AI Agent operational.
After creating and properly configuring your AI Agent, it is advisable to test it before publishing. Utilize the playground to test the AI 2.0 agents. Explore How to Test an AI Agent in BoldDesk AI 2.0.
Once you have published the AI Agent, you can configure the Agent Hub and Deploy options. Published AI Agents are listed under versions. Explore AI Agent Versioning in BoldDesk AI 2.0. You can unpublish a published AI Agent by clicking on unpublish. Only AI 2.0 Agents that have been published can be deployed or mapped to channels or widgets.
Deploy Channels
This allows you to map the AI Agent to specified channels. You can deploy created AI Agents 2.0 in BoldDesk. This option only becomes available after publishing the AI Agents.
Explore How to Assign AI Agent 2.0 to Omnichannel in BoldDesk.
Conversations
The Conversations section displays a list of AI Agent interactions, allowing you to search, review, and revisit past exchanges. Selecting a conversation shows the full context, including the ticket details and the AI-generated response, helping you track queries and responses for troubleshooting or reference. To view conversations from the BoldDesk portal, click on more options on the particular AI Agent and choose view conversations.
You will be redirected to the AI Agent conversation page from where you can view the conversations history.
Permission
To be able to create an AI 2.0 Agent in the AI Agent Creation page, an agent must be assigned the AI Agent Builder Full Access. Learn more on How to Configure AI Agent Builder Access Permission.
Available Access Levels
| Access Level | Description | Default Behavior |
|---|---|---|
| Full Access | User can create, view, edit, and delete AI agents. Provides complete management control. | Agents must be assigned manually. |
| Restricted | User has no access; cannot view, create, edit, or delete AI agents 2.0 in the AI Agent creation page. | Assigned to existing agents by default |
This model ensures that AI agents are protected by default, and only users with explicitly granted permissions can manage them.
Troubleshooting
“AI Agents” or “Create AI Agent” is not visible
Confirm the user is assigned AI Agent Creation – Full Access.
AI agent answers are irrelevant or inconsistent
- Review Instructions for missing constraints or unclear responsibilities.
- Confirm required Libraries are linked if the AI agent depends on knowledge sources.
- Review Context Management & Reasoning settings (History Message Limit, Number of Chunks, Reasoning effort and Reasoning verbosity).
AI agent cannot perform external actions (for example, lookups/refunds)
- Confirm the required MCP Servers are linked.
- Confirm the correct MCP server tools are configured and not unlinked.
Deployed AI agent is not responding on a channel
- Confirm the AI agent is published.
- Confirm the AI agent is mapped to the channel under Deploy > Map Channels.
- If the AI agent is intended for discovery via Agent Hub, confirm Share Publicly is enabled and the correct visibility scope (Global/Groups) is selected.
Frequently Asked Questions
-
What permission is required to create an AI 2.0 Agent?
A user must have AI Agent Creation – Full Access. -
What does Restricted access do?
Restricted prevents users from viewing, creating, editing, or deleting AI agents. -
Where do I change the AI agent’s name or avatar?
Use the edit icon above the AI agent name in the AI Agent creation page. -
What is the difference between Libraries and MCP Servers?
- Libraries provide knowledge sources for answering questions.
- MCP Servers provide tool connections to external systems for taking actions or fetching real-time data.
-
Do I need to publish before deploying to channels?
Yes. The AI agent must be published before it is operational and deployed. -
Can I remove an AI agent from production use?
You can unpublish a published AI agent, and you can also adjust deployment mappings and Agent Hub sharing.
