Set up tools for an agentic virtual agent

AI-generated summary

and naming conventions, with tools preferably using a "get" prefix and requiring clear descriptions. The platform supports custom data action configuration with detailed input variable setup options, either creating new variables or linking existing ones. Virtual agent profiles are managed through multiple sections including Agent, Tools, Knowledge, Configuration, and Guardrails, allowing detailed output object setup and precise tool behavior definition. The system enables integration with Architect bot flows and digital bot flows through Call Guide action, supports guide duplication, and includes a preview widget for testing. Best practices are enforced through recommended instruction handling and proper variable descriptions, while allowing users to save incomplete configurations during development.

Series: Create an agentic virtual agent

Previous suggested step: Define an agent’s role and guidelines

Feature coming soon: Agentic virtual agents

PrerequisitesClick to expand

Tools are data actions that provide your agentic virtual agents access to real-time data, enabling them to complete end-to-end actions.

Enter basic details and choose a data action

Click Menu > Orchestration > AI Studio > Agentic Virtual Agents.
(Optional) To search for a specific agent, start typing the name of the agent in the search box. As you type, the matching agents are shown.
To open an AI agent, click the agent’s tile. The Agent profile view opens.
Select the Tools tab.
Click Create. The Create Tool view opens.
Under Overview, add a meaningful name. For example, “validate user login code”. It is also recommended to add a get prefix when the agent gets something from the tool, for example, “get available hotels given distance”.
Add a description that clearly states the tool’s objective. For example, “Sends the user’s login credentials to an API endpoint. If the login credentials are valid, return a session token (aka auth token). Otherwise, return an error message”.
In the Source section, select a data action from the Choose a data action list. For more information about creating a custom data action, see Create a custom action for integrations.
Click Next to proceed to the Input tab.

Tool name examplesClick to expand

Do / Don’t	Example	Explanation
	get user checking details authorize single user session validate user login code pay bill to electricity provider get all authorised customer orders	Being clear in its purpose.
	get available hotels given distance	Including the word “get” when the tool gets you some data.
	this_is_my_tool tool_get_stuff get_user_details tool123 check_user	Being too vague about its purpose.

Tool description examplesClick to expand

Do / Don’t	Example	Explanation
	Sends the user’s login credentials to an API endpoint. If the login credentials are valid, returns a session token (aka auth token). Otherwise, it returns an error message.	A good description explains what the tool does, when to use it, what kind of parameters are used, and what it returns.
	This is a tool that returns all general knowledge about the user. This information must be remembered and can be used for any questions the user might ask.	The description does not provide any value. Also, a single tool cannot do everything. It must be specific.

Set up inputs

Add pre-instructions

Under Pre-instructions in the Input tab, click Add to add any conditions that must be met before this tool is used by the agent. For example, “Obtain only the user’s login code to login. Once they are authenticated, do not ask for authentication again”.
Add as many pre-instructions as needed for the intended tool usage.

Pre-instruction examplesClick to expand

Good pre-instructions answer these questions:

How should the agent use the tool that you are specifying?
What information should the agent ask the user for?
What should the agent do before it calls the tool?
Do you want the agent to confirm information from the user before calling this function?
What should the agent think about before it gets to this tool?

Do / Don’t	Example	Explanation
	You must obtain only the user’s login code to log in. Once the user is authenticated, do not ask for authentication again.	Indicate what the agent must do before it calls the tool.
	This function retrieves all accounts for a user. If the user has questions about accounts in general, or about opening new accounts, call search_knowledge_base.	Provide more information about what the tool does. If the model decides to call it, specify what it should consider beforehand. With this information, it may choose to call another tool instead.
	If the user requests to exit the conversation or log out, you must call this function to end the session. This will end the current session.	Provide more context about what happens when the tool is called so that the agent can give the user more context.
	When you call this tool, you act as a School Assistant agent.	Incorrect use of pre-instructions. Do not switch to another domain.
	You can call all other tools whenever you want.	Incorrect use of pre-instructions.

Set up variables

The input variables are pulled from the data action. To ensure that they are set up for the tool, follow these steps:

To set up the input variables for the tool, follow these steps:
1. Under Inputs in the Input tab, select one of the following options:
  - Create new: It allows you to create a new input variable to use with the tool.
  - Link existing: It allows you to link to an existing variable from another tool or start context from outside the agent.
2. If you selected Create new in the previous step, enter these details:
  - In the Name field, enter a variable name.
  - In the Description field, provide a clear description about the variable. For example, “8-digit login code that is used to authorize a user into the system.”
  - From the Additional details list, select List and then add a comma-separated list of values if you want the agent to accept only those values.
3. If you selected Link existing, do these steps:
  1. Click Select to view all available variables on the right pane.
  2. Select the required variable or create the relevant start context and click Select.
Click Next to proceed to the Output tab.

Set up the output object

Under Outcome Instructions in the Output tab, click Add to define how the agent must present the outputs to the participants or what action the agent must take.
Add as many outcome instructions as needed for the intended tool usage.
To set up the output object for the tool, follow these steps:
1. Under Outputs, enter the following details:
  - In the Name field, enter an output object name. For example, “PaymentRecord“.
  - In the Description field, provide a clear description about the output object. For example, “Record of a processed payment transaction.”
  - Select the output parameters to include in the output object.
Click Next to proceed to the Summary tab.

Tool outcome instruction examplesClick to expand

Do / Don’t	Example	Explanation
	You have retrieved the details of the customer’s orders. If the status of an order is completed, call the send_sms_to_customer tool.	Instruct the agent to move to another tool directly after completing one.
	Always confirm the payment amount and payment method with the customer before processing the payment.	Instruct the agent always to confirm the critical actions after the tool is called.
	The user authenticated successfully! You must not ask the user to provide login details again.	Showing that the tool has been called successfully.
	List the available providers in a clear format and ask if they want details about any specific provider.	Specify how to present the content to the user.
	If is_authenticated is true, then you have authenticated the user correctly.	It can do some very simple evaluation of the variables.
	Give the user the total amount by adding the first and second balance.	This instruction must be handled to a tool. Do not use the model for maths.
	Ask the user for their email address.	This instruction is ideal for a pre-instruction instead.
	Be polite to the user.	This instruction is ideal for a role or guideline.
	Tell the user that the operation is complete.	This instruction does not add any value. The user can see if it is a success or not.
	Only call this function if the user explicitly asks.	This instruction is ideal for a pre-instruction instead.

Review and save the tool

In the Summary tab, review all the parameters of the tool.
Click Edit any of the tiles if you want to edit the parameters.
Click Create to save the tool.

Navigating the agentic virtual agent profile

The profile contains the following sections and a preview widget:

Show meClick to expand

You can switch to any of the sections of the agentic virtual agent while creating a new agent, editing an existing agent, or viewing an agent. Any incomplete section is indicated by the icon preceding the section name. You can save an agent with an incomplete section but you cannot publish the agent until all sections are complete.

After you have completed all necessary sections of the agent setup, you can click Publish to publish the flow.

[NEXT] Was this article helpful?

Get user feedback about articles.

Set up tools for an agentic virtual agent View summary