Getting Started with Zia Agents (ServiceDesk Plus Cloud) – FAQs

Getting Started with Zia Agents (ServiceDesk Plus Cloud) – FAQs

1. What is an AI Agent? 

Till date, AI in ITSM Cloud has primarily been used to predict and generate outputs based on predefined prompts embedded within generative features such as Reply Assist and Resolution Assist. These capabilities were largely focused on assisting users through content generation.

AI Agents represent a shift towards AI-driven decision-making, where actions are governed by the instructions (prompts) defined for the agent. An AI Agent is an AI-powered autonomous system capable of making decisions, performing actions, and contributing to the achievement of business goals.

Unlike traditional generative AI, AI Agents extend beyond content creation by enabling AI to handle multi-step problems independently. They operate across various stages of the business lifecycle—from ideation to post-sales support—ensuring improved efficiency, precision, and intelligent execution.

Zia Agents introduce a new era of productivity through agentic AI, where autonomous agents can independently execute complex business tasks, rather than being limited to generating responses.

2. What are the steps involved in creating an effective agent?

Below are the processes that need to be defined before creating an effective agent:
Define the process that the AI agent is expected to handle, such as notifications, integrations beyond ServiceDesk Plus, intent identification, or workflow automation. Clearly outlining the scope ensures the agent remains focused and effective.
 
Once the process is defined, navigate to Agent Studio from ServiceDesk Plus Cloud under Setup → Artificial Intelligence → Agents, or access it using the URL agentstudio.zoho.com (URL may vary depending on the data center).
 
1. Knowledge Base → Ensure the agent has access to relevant sources before taking actions, such as policies, procedures, solution articles, or web content. These should be properly grounded and added within the Knowledge Base of Agent Studio. While optional, the Knowledge Base significantly enhances the agent’s reasoning and decision-making capabilities. It can include product information, workflows, step-by-step instructions, compliance guidelines, routing logic, and ownership rules. For example, when assigning ownership based on expertise, the agent can refer to KB rules and identify relevant keywords to determine the most suitable owner.
 
2. Tools → Tools act as the execution layer. Whenever the agent performs actions such as fetching or updating records, sending emails, or triggering workflows, it does so through APIs enabled as tools. Without tools, the agent can only respond; with tools, it can take action.
 
If automation is required within ServiceDesk Plus Cloud or third-party applications, ensure the necessary tools are available in Agent Studio. If not, custom tools can be created using YAML. Existing scopes can be reused or authorized, while new scopes may require creating custom connections. Custom tools must be tested successfully before they can be saved.
 
3. Agent Instructions → This is the core of the agent’s intelligence. Define what the agent should do, which tools it should use, and the rules it must follow. This includes process-specific logic as well as global rules such as prioritization or escalation. Well-defined instructions ensure consistent and accurate behavior.
 
4. Guardrails → Under Additional Settings, custom guardrails can be configured to control agent behavior. These define what the agent should do (Do’s) and what it must avoid (Don’ts). Up to 10 rules can be added in each category to ensure the agent operates within defined boundaries and maintains compliance.
Notes
Building simpler agents focused on specific tasks tends to deliver better results. Overloading a single agent with multiple tools and complex automations can reduce effectiveness and make instructions harder to manage.

3. What are the best practices for creating AI agents?

Write detailed instructions → Clearly define the agent’s expertise, expected behavior, responsibilities, and limitations. The more specific and structured the instructions are, the more reliable the agent’s output will be. Vague instructions often lead to inconsistent or unclear results.

Be specific about actions and outcomes → Clearly describe what the agent should do and what the expected outcome should be. Precise prompts and instructions help the agent deliver more accurate and useful responses compared to generic or ambiguous requests.

Load relevant knowledge → Provide only the necessary information that aligns with the agent’s role. Adding excessive or unrelated data can introduce noise and reduce the accuracy of the agent’s reasoning.

Use guardrails for non-negotiable rules → Define strict boundaries for actions the agent must always follow or avoid. Critical restrictions should be enforced through guardrails rather than relying only on instructions.

Test before deployment → Validate the agent by running realistic scenarios in the testing environment. Refine instructions and configurations iteratively until the agent consistently produces expected results.

Back up your data → If the agent has permissions to create, update, or delete data, ensure proper backups are in place to prevent unintended data loss or errors.

6. Is there any limit on agents?

      There is currently no limit on the number of Zia agents that can be created. However, it is recommended to use them efficiently based on your requirements.

7.  What are the number and size limits on Knowledge Base?

      Each agent can have up to 100 Knowledge Base documents. The total storage limit per organization is 1 GB, with a maximum file size of 50 MB per document.

8. Are agents supported in Non-IT instances?

      Yes, agents are supported in both IT and Non-IT instances.

9. What are the supported formats for Knowledge Base?

      Supported formats include TXT, MD, PDF, DOCX, and web scraping.

10. Is there any limit on tools that can be added?

      You can add up to 100 tools per agent.

11. What is the maximum length of agent instructions?

      Agent instructions can be up to 10,000 characters.

12. How can I add more tools in Agent Studio?

 You can create additional tools by navigating to Agent Studio → Tools → Custom Tools, where you can define and configure your own tools.

13. Is it possible to add more tools to an existing custom tool group?

Currently, once a custom tool group is tested and saved, it is not possible to add additional tools to that group. However, you can create the required tool as a separate tool group and then associate both the existing group and the newly created group with the agent.

14. Is there a specific format that the YAML file should follow?

      Yes, the YAML file must follow the OpenAI-compatible schema format, as it defines how the agent understands and interacts with tools. This includes clearly structured definitions for endpoints, parameters, authentication, and responses.

The YAML should include details such as:

  • API metadata (name, description, version)

  • Endpoints and methods (GET, POST, etc.)

  • Input parameters with types and descriptions

  • Request and response schemas

  • Authentication configuration, typically using OAuth

In addition, the YAML must have the required OAuth scope information embedded, ensuring the agent has the correct permissions to execute actions securely.

It is important to ensure that:

  • The structure is well-formatted and valid YAML

  • All fields are clearly defined and unambiguous

  • Only necessary scopes and permissions are included (principle of least privilege)

  • The configuration is tested successfully before saving, as invalid or incomplete YAML definitions will not be accepted

15. Should I create custom tools every time for testing agents that require both SDP and other actions?

      No, it is not necessary to create custom tools every time. You should only create custom tools when the required action is not already available within the existing predefined tools.

If the functionality you need is already covered by built-in tools, it is recommended to reuse them instead of creating new ones. This helps maintain simplicity and avoids unnecessary duplication.

Additionally, agents can utilize multiple tools within the same workflow. You can partially use different tools based on the requirement, allowing the agent to combine actions from ServiceDesk Plus and other systems effectively during execution.

16. What are the supported AI providers in Agent Studio?

      Agent Studio supports multiple AI providers to offer flexibility based on organizational needs. Currently, the supported providers include Zia-hosted LLM, OpenAI, and BYOK (Bring Your Own Key) models.
This allows organizations to either use Zoho’s native AI capabilities, integrate with OpenAI, or connect their own preferred LLM provider using BYOK for greater control and customization.

17. Should I select the agent and provide a prompt to make it work inside the SDPC Ask Zia chatbot?

      No, it is not required to manually select an agent when initiating a conversation. During your first interaction after deployment, Zia automatically identifies the intent of your prompt and invokes the most relevant agent.
Once an agent is invoked, it remains active for the subsequent follow-up interactions within the same conversation.
If you want to switch to a different agent, you can either manually select it or use the “@” command, which will display all the available and deployed agents for selection.

18. Can I trigger agents via Custom Function?
      Yes, agents can be triggered using a Custom Function by leveraging the Agent API. This allows agents to be integrated into workflows and triggered based on specific events or conditions. Using this approach, you can incorporate agent capabilities into automation scenarios, making them part of your existing workflow logic.
Additionally, support for directly using agent actions within workflows and triggers is currently being worked on and is expected to be available as a built-in capability in the future.

19. How is an AI agent different from normal automations? When should I use it?

      AI agents and traditional automations serve different purposes. Regular automations in ServiceDesk Plus Cloud are rule-based and deterministic, meaning they follow predefined conditions and execute fixed actions. They are highly reliable for structured, repetitive tasks that do not require interpretation or decision-making.
In contrast, AI agents are context-aware and dynamic. They can understand intent, interpret unstructured inputs, make decisions based on context, and choose appropriate actions or tools accordingly. This makes them suitable for scenarios that involve variability, ambiguity, or require intelligent reasoning.
However, agentic actions should be used only when necessary. If a requirement can be effectively handled using standard automations, business rules, or custom functions within ServiceDesk Plus Cloud, it is recommended to rely on those instead. Traditional automations are more predictable, easier to maintain, and do not incur additional operational costs.

AI agents, on the other hand, consume tokens for each interaction and action, which can make them a more expensive option if overused. Therefore, they should be applied thoughtfully in use cases where intelligence and adaptability are truly needed.

As a best practice, evaluate whether the requirement can be achieved using existing automation capabilities first. If not, then consider leveraging AI agents. For further guidance, you can also reach out to ServiceDesk Plus Cloud support to validate the most optimal approach.

20. Any tips and tricks on writing effective AI agent instructions?

Define a clear role → Treat the agent like a new employee. Clearly state its purpose, responsibilities, and expected outcomes.

Be specific → Avoid vague instructions. Clearly define actions, decision logic, and expected results.

Guide tool usage → Explicitly mention which tool to use in which scenario to avoid incorrect actions.

Structure instructions → Organize into goal, tasks, rules, and constraints for better clarity.

Use examples → Provide sample scenarios or outputs to improve accuracy.

Handle uncertainty → Instruct the agent on what to do when it is unsure (e.g., ask for clarification).

Keep it simple → Avoid overloading a single agent with too many rules or tools.

Use guardrails → Enforce strict do’s and don’ts outside of instructions.

Test and iterate → Continuously refine instructions based on real usage.

21. What are Guardrails?

Guardrails are configurable rules that control how an AI agent behaves. While creating an agent, you can enable Custom Guardrails under Additional Settings to define clear boundaries for its actions and responses.

They allow you to specify:

Do’s → Actions and behaviors the agent is encouraged or allowed to follow.
Don’ts → Actions and behaviors the agent must strictly avoid.

You can define up to 10 rules in each category, helping ensure the agent operates consistently, safely, and within your defined guidelines.

22. What are Tools?

A Zia Agent can understand requests and reason through them, but it cannot perform actions in your business systems unless it is provided with tools.

Tools act as the action layer of the agent. Whenever an agent performs tasks such as fetching records, updating data, or sending emails, it does so by invoking APIs that are enabled as tools.

Without tools, the agent is limited to responding to queries. With tools, it can execute real operations and automate workflows across systems.


23. What if the application I’m trying to connect is not available in Agent Studio?

If the application you want to integrate is not available by default, you can still connect it by creating a custom tool group.

As long as the application has published APIs, you can define the integration using a YAML configuration. Once created, the custom tool must be tested successfully before it can be saved and used by the agent.

24. What is Agent API?

Agent APIs allow you to invoke and interact with your agents outside Agent Studio.

They enable external applications, workflows, or custom functions to trigger agents, pass inputs, and receive responses—making it possible to integrate agent capabilities across different systems.

25. What does Multi-Agent capability do in Zia Agent Studio?

The Multi-Agent capability allows you to create multiple agents for different purposes and have them work together seamlessly.

Instead of invoking each agent individually, the system can orchestrate them intelligently, especially when handling complex or multi-step workflow automations.

This makes it easier to manage complicated processes, where different agents handle different parts of the workflow, ensuring smooth and efficient execution.

26. I am getting an internal server error while testing the agent. What should I do?

In such cases, a recommended approach is to create or refine a custom tool using a YAML configuration.

You can either build the YAML from scratch or download the API schema (if available), modify it according to your business requirements, and upload it under the Schema section while creating the tool. Ensure the YAML is properly structured, includes correct endpoints, parameters, and authentication details, and is successfully tested before saving.

It is also important to understand common API error codes:

200 (Success) → The request was successful and the tool/API is working as expected.

400 (Bad Request) → The issue is usually with the request itself, such as incorrect parameters, missing required fields, or invalid input format. Review your YAML configuration and request payload.

500 (Internal Server Error) → This indicates a server-side issue, often due to incorrect API configuration, invalid endpoints, authentication issues, or problems within the external system.

Most testing errors occur due to misconfigured YAML, incorrect inputs, or authentication mismatches, so validating these areas typically resolves the issue.

27. What’s next in AI agents inside ServiceDesk Plus Cloud?

Upcoming enhancements include pre-built agents for quick setup and agent actions within workflows and triggers.

Zia Agents will also be available in contextual Ask Zia (Ziv) and Microsoft Teams for better accessibility.

There are plans to support more tools and integrations, along with making Ask Zia an agent orchestrator to manage multiple agents.

Additionally, user authentication support is being introduced for more secure interactions.

Notes

Supported Editions --> Enterprise

Supported DC's --> US, IN and EU (CA and AU in progress)
AlertModel configuration and selection are disabled by default when creating your first agent. To enable these options, go to Setup → Plans & Billing and activate the ZKS toggle. Once enabled, you can select Zoho’s hosted model and other available models.



      New to M365 Manager Plus?
      New to M365 Manager Plus?
        New to RecoveryManager Plus?
        New to RecoveryManager Plus?
          New to Exchange Reporter Plus?
          New to Exchange Reporter Plus?
            New to SharePoint Manager Plus?
            New to SharePoint Manager Plus?
                See all integrations
                New to ADManager Plus?
                See all integrations
                New to ADManager Plus?

                  New to ADSelfService Plus?

                  Start your free trial
                  Start your free trial

                    • Related Articles

                    • ServiceDesk Plus Cloud License Related FAQ's

                      1. How can I change the payment method for our ServiceDesk Plus Cloud account from offline to online? To change the payment method from offline to online, click on the username in the top right corner and scroll down to locate the Manage option, as ...

                    • ServiceDesk Plus Cloud Artificial Intelligence Related FAQ's

                      1. What is Zia? Does Zoho have its own LLM? Zia (Zoho's Intelligent Assistant) is Zoho’s in-house AI assistant integrated across various Zoho applications, including ServiceDesk Plus Cloud. It helps users with features like chatbot, solution ...

                    • Current AI Capabilities in SDP Cloud

                      What is LLM? LLM stands for Large Language Model, a powerful AI technology that processes and generates human-like text based on context. The natively hosted LLM has been specifically designed to enhance SDP Cloud's AI capabilities, offering secure, ...

                    • Zoho Directory - ServiceDesk Plus Cloud (FAQ's)

                      1. What is Zoho Directory? Zoho Directory is an identity and access management (IAM) platform that enables organizations to securely manage user identities and their access to various applications, both within Zoho and third-party services. It ...

                    • Support Options for ServiceDesk Plus, MSP & Asset Explorer Cloud

                      ServiceDesk Plus, MSP, and Asset Explorer Cloud customers can reach our support team directly through the Support icon located at the bottom-right corner of the application. We offer three modes of assistance: Chat, Call, and Email. Email Support ...

                      Related Products