Understanding and configuring chat agent

Note: This feature is currently available as an early access feature for the enterprise plan in Creator 6.

In a nutshell

Chat agent lets you create an AI powered assistant to your Zoho Creator application. Powered by the AI provider configured in Zia, it enables end users to interact with application data using natural language to retrieve information, gain insights, and perform actions through a chat interface. You can create a chat agent by describing its purpose in plain language, after which it automatically generates its initial configuration, including instructions, access level, and relevant modules. This configuration can be further customized to suit your business requirements.

Availability

Available as an early access feature on the enterprise plan in Creator 6.
Only super admins and admins can enable chat agent from Operations > Zia.
Super admins, admins, and developers can create and configure chat agents.
Users and portal users with the required permission can access chat agent in live mode.

1. What is chat agent?

Chat agent enables you to create an AI assistant in the Zoho Creator application that lets end users interact with the application's data entirely in a natural language. Chat agent understands what the user needs, fetches the relevant data, and responds through a conversational interface in live mode. Each chat agent is scoped to a specific purpose. You define what it can access, what it can do, and how it should respond through its instructions, access level, and connected modules. You can configure as many chat agents as needed within the same application, each with a different scope and access level to serve different audiences or use cases.

AI agents can be linked to chat agent during its configuration to handle operations that require multi-step or cross-application workflows. Chat agent automatically detects the intent from the user's prompt and invokes the relevant linked AI Agent, without requiring the user to explicitly trigger it. Learn more about AI Agents.

In live mode, end users can perform all data operations in plain language through the chat interface. They can query records, retrieve insights, create or update entries, and ask follow-up questions within the same conversation to refine results or dig deeper. A confirmation step is shown before any data is modified.

Chat agent can assist end users with the following:

Query data: Search for records, retrieve insights, and look up information from connected modules.
Modify records: Create, update, or manage records in plain language. A confirmation step appears before any changes are applied.
Trigger AI Agents: Invoke linked AI Agents to handle complex, multi-step operations that go beyond querying or updating records. Chat agent automatically detects the intent from the user's prompt and triggers the relevant AI Agent without any manual input from the user. Learn more about AI Agents.

Note:

End user actions in chat agent are governed by the configured scope. Users can retrieve information from modules with Read access and can create or update records in modules with Read & Write access.
In addition to the agent's scope and access level, users must also have the permissions to access the relevant forms and reports within the application.

1.1. How does chat agent work?

Chat agent takes a plain-language input from the user, interprets the intent behind it, and maps it to the data and operations available in your application. It uses the LLM configured in Zia to understand the request and generate a response grounded in your application's records.

The agent's behavior is shaped by its configuration. The instructions define its scope and tone, the connected modules determine what data it can reach, and the access level controls whether it can only retrieve data or also modify records.

For retrieval requests, the agent returns information directly. For write operations, it presents the user with the proposed change for confirmation before applying it. Requests that fall outside the configured scope are rejected entirely.

1.2. Chat agent interface in live mode

Note: This section is relevant only to end users interacting with chat agent in live mode.

The Chat agent interface consists of the following elements when accessed from live mode:

New Chat: Starts a new conversation with the agent. Use this option when you want to begin a different topic or when the context of the current conversation is no longer relevant. Previous conversations remain available in the conversation history.
Input area: The text box where you enter prompts and interact with the agent. It includes the following options:

Attach Files (): Upload a PDF, image or text file to provide additional context for your request. Use this to retrieve information from the file, update records with values from the file content, or attach the file directly to a record. The agent can access the file content only if file uploads are permitted by the configured guardrails.

Note: You can attach up to 5 files per prompt, with a maximum size of 10 MB per file.

Send (): Sends your message to the agent.

Quick actions: A set of prompt suggestions displayed below the input area. Click any button to send the corresponding query without typing. AI auto-generates these based on your application to help you get started quickly.
Conversation history: Your previous conversations with the agent are accessible from the hamburger icon (). Select any conversation to continue it from where you left off. Use More options (...) next to a conversation to Rename or Delete it.
Chat messages: The main conversation area where messages between you and the agent are displayed. User messages appear on the right, while agent responses appear on the left alongside the agent's avatar.

2. Business use cases

Case 1: Creating an employee self-service assistant for HR

An HR administrator at a mid-sized company manages a Zoho Creator HR application used by employees to apply for leave, track request statuses, and access company policies. While employees can view their leave balances and submit requests directly in the application, they often need help understanding specific policy rules such as leave eligibility, carry-forward limits, and approval conditions. These clarifications are typically routed to the HR team.

To reduce this dependency, the administrator creates a chat agent with Read access to the Leave Requests and Employee modules. Private Data Access is enabled on the Guardrails tab to ensure sensitive employee information is not exposed to the AI model.

Employees can now interact with the chat agent to check their leave balance, track the status of their requests, and get clear answers to policy-related questions in plain language, without navigating the application or contacting HR directly.

Sample prompt: "What is my current leave balance and the status of my latest leave request?"

Case 2: Enabling real-time inventory lookup for warehouse teams

A supply chain manager at a distribution company manages a Zoho Creator inventory management application used by warehouse staff across multiple locations. Staff need to check stock levels, locate items, and update inventory counts from the warehouse floor, where navigating desktop reports is impractical.

The manager creates a chat agent with Read & Write access to the Inventory and Warehouse modules, and configures quick-action buttons for the most common tasks. Staff can check stock levels by item name or SKU(Stock Keeping Unit), locate items across warehouse locations, and update inventory counts through the chat interface. Any update triggers a confirmation step before the change is applied.

Sample prompt: "Help warehouse staff check stock levels by item name or SKU, locate items across warehouse locations, and update inventory counts."

Case 3. Automating purchase order creation in a procurement application

A procurement coordinator at a manufacturing company manages a Zoho Creator procurement application. When stock levels drop, department heads need to raise purchase requests, a process that spans the inventory application, the vendor management application, and the finance application.

The procurement admin creates a chat agent with access to the Purchase Orders module and links an AI Agent that holds the end-to-end PO workflow as its toolset. When a department head types a request, chat agent invokes the AI Agent, which checks stock levels in the inventory application, retrieves vendor details from the vendor management application, verifies the available budget in the finance application, creates the purchase order, and notifies the vendor. All of this happens from a single prompt and across applications that chat agent alone cannot reach.

Sample prompt: "Raise a purchase order for 500 units of Product X from our preferred vendor and notify them."

4. Steps to configure chat agent

To make chat agent available to end users, create and configure it from your application's edit mode.

4.1. Prerequisites

Enable build agent from early access features

Navigate to Operations > Applications > Early Access Features.
Search for Build Agent and turn on the toggle to enable it for your account.

Configure Zia

To use chat agent, set up Zia to manage AI features across your Creator account. Zia is configured with Zoho GenAI by default, so no additional setup is required. IIf you prefer a different provider, you can integrate Zia with any of the supported LLM providers (OpenAI and Anthropic). Setting up Zia is a one-time process. Learn how to configure Zia.

Note: Zoho GenAI LLM is configured by default and available for free and processes all prompts within Zoho. External LLM providers process data in their own systems. Avoid sharing sensitive or regulated information when using external providers with Zia.

Once Zia is configured, enable the Access toggle for Chat Agent under the Enable Features section to start using it in your applications.

4.2. Create a chat agent

Navigate to the edit mode of your preferred application and click the + icon in the header strip.
Select Chat Agent. Create Chat Agent screen will be displayed.
Type a plain-language description of what the agent should do in the text area and click Create Agent. Learn how to write effective descriptions for chat agent.

Note: The agent is ready to use at this point. Click Done to skip customization and deploy it with the default configuration, or continue through the tabs below to refine its behavior.The agent will be created and automatically configured, and you will be directed to the chat agent builder for further customization. This includes four configuration tabs: Basic Details, Connectivity, Personalization, and Guardrails. These customization steps are optional.

Review and adjust the following in the Basic Details tab, then click Save to proceed to the Connectivity tab:

Instructions: Auto-filled based on your description. Refine to ensure the agent responds accurately within the intended scope. Clear, specific instructions improve response quality.
Access Level: Auto-set based on your description. Change to Read to allow data retrieval only, or Read & Write to allow record creation and modification.
Modules: Auto-selected based on your description. Modules refer to the forms in your Creator application. Add or remove modules to control the scope of data the agent can retrieve and act on.

Select the AI Agents in your organization if you want to associate with the current chat agent in the Connectivity tab, if required. Click Save to proceed to the Personalization tab.
Linking AI Agents enables a multi-agent setup, expanding the scope of Chat Agent across specialized functions without requiring a single agent to cover every use case. Based on the user's prompt, chat agent automatically invokes the relevant linked AI agent to handle the request. Learn more about AI Agents.
Configure the following in the Personalization tab, then click Save to proceed to the Guardrails tab:

Chat Agent Name: Auto-filled based on your description. Update to set the display name shown to end users in live mode.
Link Name: Auto-filled based on the agent name. Update to change the URL-friendly identifier for the Chat Agent.
Avatar: Click Upload and select a custom image from your system to represent the agent in the chat interface. You can also remove the uploaded image by clicking the Remove button.
Welcome Message: Auto-filled based on your description. Update to set the opening message displayed when a user starts a new conversation.
Response Tone: Defaulted based on your description. Choose the communication style for the agent's responses, such as Helpful or Supportive, to match your application context and audience.

Configure the following in the Guardrails tab, then click Save to finish the configuration.

File Upload: Enabled by default. Disabling this option prevents the agent from accessing data in files attached to records. Use this when the agent should not process documents or spreadsheets uploaded through forms.
Private Data Access: Enabled by default. Disabling this prevents the AI model from accessing fields containing personally identifiable information (PII) and electronic protected health information (ePHI). PII and ePHI fields are identified based on the privacy settings configured in the field properties. Recommended for applications that handle sensitive personal or health data.

Click Done. The chat agent is added to your application and appears in the navigation menu in edit mode.

5. Managing chat agent

To manage a chat agent:

Navigate to the chat agent in the navigation menu of your application's edit mode.
Hover over the chat agent and click Open Chat Builder.
Update the required settings across the Basic Details, Connectivity, Personalization, and Guardrails tabs and click Save.
Click the vertical ellipsis icon on the top right corner of the chat builder and select the required action:

Rename: Update the agent's Display Name and Link Name in the Rename Chat Agent popup and click Rename.
Delete: Remove the chat agent permanently from the application.

6. Configure an effective chat agent

The quality of a Chat Agent's responses depends on how clearly you define its purpose and scope. The following guidelines help you build an agent that is accurate, focused, and useful for end users.

6.1. Tips for best results

Write a specific description: When creating the agent, describe exactly what it should do and who it serves. A specific description produces more accurate auto-generated instructions than a generic one.
Scope modules carefully: Add only the modules the agent needs. Including unrelated modules increases the chance of unexpected responses and exposes data that end users should not access.
Set the right access level: Use Read for query-only agents and Read & Write only when users genuinely need to create or modify records. Read & Write always shows a confirmation before applying changes.
Refine the instructions: The auto-generated instructions are a starting point. Review them on the Basic Details tab and add constraints, scope limits, or specific behaviors.
Enable guardrails for sensitive data: If your application handles personal or health data, enable Private Data Access on the Guardrails to prevent sensitive information from being shared with the AI model.
Extend Chat Agent with AI Agents: For operations that require dynamic decision-making or sequential task execution, link AI Agents through the Connectivity tab instead of trying to handle everything within Chat Agent's native capabilities.

6.2. Example agent descriptions

Act as a front desk assistant for hotel staff. Help them check room availability by date, view guest booking details, and update reservation statuses. Respond only to queries related to reservations and guest management.
Help nurses log patient vitals, update medication records, and flag entries that fall outside normal ranges for doctor review. Limit responses to patient care data only. Do not share or display any patient identifiable information in responses.
Serve as a support desk assistant for IT staff. Help them log incoming service requests, assign tickets to the relevant team based on issue type, update ticket statuses, and track resolution timelines across all open requests..

7. Points to note

Operational behavior and capabilities

When a user requests a change, chat agent displays a confirmation screen with the affected record details before applying the change.
Chat agent does not support reverting changes. Deleted records cannot be restored. For record creations or modifications, changes can only be reversed by prompting the agent to update or delete the record again.
Chat agent operates only within the application from which it is accessed. It cannot directly access other Creator applications or external systems. To perform operations across applications, link other AI Agents through the Connectivity tab during configuration.
Chat agent cannot access records from modules not configured in its Basic Details tab.
If integrations are already configured and authorized in the application, chat agent can use them to perform the requested operations.

AI usage and data

Zoho GenAI LLM is available for free and it processes all prompts within Zoho, while external LLM providers process data in their own systems. Avoid sharing sensitive or regulated information in Build Agent when using external providers with Zia.
Chat agent does not consume any AI calls or API calls from Zoho Creator. However, the LLM provider configured with Zia may still enforce API rate limits. These limits are restrictions that the LLM provider imposes on the number of times a user or client can access their services within a specified period of time and vary based on their usage tier and the model consumed. Rate limits for LLM providers can be referred by their respective official documentation.

Along with your prompts, the system may add supplementary prompts to enhance the accuracy of chat agent responses. These system-generated prompts are also counted toward AI usage by the LLM provider configured in Zia.

Understanding and configuring chat agent | Zoho Creator Help