Who Can Use the Official Agent Toolkit?

• Faculty and Staff

• Those who are creating agents on behalf of a unit, team, or college with approval.

Users intending on creating agents on behalf of their units will be expected to confirm that they have read this documentation on the request form.

Getting Started

To access the Official Agent editor, click on your user profile name at the bottom left of the UI and toggle the UI to Management.

Open

Under the Manage Agents header in the left toolbar you will see a menu option for Agents.

Here you should see the Agent Template you’ve been assigned to edit. In this example it is called “Official Agent Template”, this name my change depending on the request.

Open

Click on the name of the agent to begin editing.

• 1 Who Can Use the Official Agent Toolkit?
• 2 Getting Started
  • 2.1 Editor Interface
    • 2.1.1 Chat Playground
    • 2.1.2 User Experience
• 3 Instructions
  • 3.1 Additional tips
• 4 Capabilities
• 5 Knowledge Sources
  • 5.1 File Libraries
  • 5.2 Sites
  • 5.3 APIs
    • 5.3.1 Add Parameter
    • 5.3.2 Select Parameter Value
  • 5.4 Agent Handoff
    • 5.4.1 Handoff to another Agent
    • 5.4.2 Ask another Agent
• 6 Additional Settings
  • 6.1 Reduce Hallucinations
• 7 Publishing Your Agent
  • 7.1 Users Access
  • 7.2 Token Quotas
  • 7.3 Management Access
  • 7.4 Agent Status
  • 7.5 Finalizing and Publishing
  • 7.6 Managing Your Published Agent
    • 7.6.1 Activation Form

Editor Interface

The editor interface looks a lot like the Personal Agent’s interface with a few more options at your disposal.

Chat Playground

In the Chat Playground you can:

1. Change the Agent’s Name

   1. Give it a descriptive name that is meaningful.

2. Give it a description

   1. Optional - describe the agent, this is displayed to the user in the interface.

3. Switch the Model used

4. Set Instructions

   1. Addressed in Instructions below.

5. Capabilities

   1. Addressed in Capabilities below.

6. Add Knowledge Sources

   1. Addressed in Knowledge Sources below.

7. Test the Agent without saving or publishing.

   1. Here you can test out the changes you’ve made without saving, it will update automatically in the playground window.

Open

User Experience

In User Experience you can:

1. Edit Appearance

   1. You can upload a custom avatar to suit your agent.

2. Set the Welcome Message

   1. This is what the users see when they open the agent, this usually states the agent’s purpose and features in a friendly and conversational tone.

3. Add Starter Prompts

   1. You may add up to 4 starter prompts to an agent, these are prompts to the agent to get the users started. Typically these are frequently asked questions to the agent or a demonstration of what the agent is capable of so the user can get familiar with working with it.

Open

 

Instructions

To get started, UT Spark: Tips for Creating Personal Agent InstructionsPreview provides guidance on creating instructions for agents, this is true for Official Agents as well.

Additional tips

• Use the default instructions to build within that structure, it’s helpful to the agent author and the agent for the instructions to have a clear structure defined however it is not required.

• You can specify structured output by providing the agent examples:

• You can also provide it a template:

• Tell it what day it is! It can be helpful for you to add this line at the end of the instructions so the agent is aware of the date. This is an internal value, and it should already be in the default instructions.

• If your agent is configured to use an API, you can define it’s use here as well as in the Knowledge sections, however defining it here will be more like a global setting on “how to use the API calls”.

Capabilities

Open

 

• Allow Chat with File

  • This allows users to upload files into the message bar and chat with them.

• Allow Image Creation

  • This allows users to create images.

• Search the Internet

  • This allows the agent to search the internet for results. If you are intending this agent to deal with controlled or confidential data, you should keep this disabled.

Knowledge Sources

Here you can provide the agent with some information, and configure how that information is used or accessed by the agent.

File Libraries

If you have been given the ability to add a file library, you can click on File Libraries on the left side toolbar and click New File Library.

Open

Click on + Add File Library when prompted.

In the File Library menu, you can:

1. Drop or upload files into the library using the widget.

2. Create or Change the file library name.

3. Display Citations

   1. This will show the user where the agent got it’s information. This can be helpful if the user wants to confirm accuracy.

4. Allow Citation Download

   1. This will allow the user to download the citations, which means the user would be able to download the referenced file from the library. If the documents should not be downloaded by the end users, keep this setting off.

For the Admin Roles, this will be determined by the UT Spark Team as needed.

Open

To add the document library click on Add Knowledge Source + in the Chat Playground.

Open

Use the dropdown menu to choose the appropriate library.

In the agent you will see that the source has been added, you can modify it as such:

1. Maximum Documents to Use

   1. You can tell it how many documents to use within the file library.

2. When to Use

   1. These can be a set of mini-instructions to the agent about how to use the files. For example, you can instruct the agent to “only use this library when users request specific information about XYZ”.

Additionally you can delete the knowledge source here with the icon at the top, in this case it is a File Library, so deleting the knowledge source for the agent will not delete the file library. It can still be referred to again if needed or by another agent with access.

Open

Sites

To add Sites to the agent click on Add Knowledge Source + and chose Website.

Open

You may use this feature with the Internet toggled off, however this does mean that the agent searches the website provided using the Internet. It will limit the scope of the searching to the defined sites, however when dealing with controlled or confidential data within an agent this feature should not be used.

When a Website is added to the agent, you will see it show up in the Chat Playground.

Here you can enter:

1. URL

2. Maximum Pages to use

   1. This will inform the agent how deep to link and search, the larger this value is the more inefficient the agent becomes, keeping it below 5 improves results. If the sites contain a lot of information on each level, keep this as small as possible. You may run into errors in testing if the information retrieved is extremely large.

3. When to use

   1. Like the File Library, these are mini-instructions to the agent on when and where to use the provided URL. For example: “Search the site when the user specifically asks about: foo, bar”

 

You can add multiple sites to the agent, as well as remove them with the delete icon at the top of the dialog.

Open

APIs

To add API calls to the agent click on Add Knowledge Source + and chose API Call or Mock Call to API. Mock calls are used for testing.

Open

When an API Call is added to the agent it can be seen in the Chat Playground.

Here you can:

1. Name the Call

2. Provide a Description - These are mini-instructions to the agent on when and how to use the call.

3. Set the Parameters Schema

4. Set the Endpoint

5. Choose the Method - Get and Post

6. Add Parameters - You may add multiple.

Add Parameter

Open

Select Parameter Value

You do not need to use the “Select” option. This is limited, however contains current user information that may be helpful in some cases.

Open

 

Open

 

Agent Handoff

If you are working on multiple agents, you can orchestrate agent handoff.

Open

To add Agent Handoffs to the agent click on Add Knowledge Source + and chose Handoff to another Agent or Ask another Agent.

Handoff to another Agent

This will hand the user to the other agent, they will exit the current agent and will conduct their conversation with the agent specified. This can be helpful if the user is asking about another topic handled elsewhere or if the agent is designed to route to other agents.

Use When to Use to define this.

Open

Ask another Agent

This will ask the agent specified, use this when you want the user to stay within this agent but want to leverage another agent for the response to a certain topic.

Use When to Use to define this.

Open

Additional Settings

Reduce Hallucinations

You may see this appear after choosing knowledge sources, you may toggle this and see how it responds. However we have seen that agents tend to perform far better without this setting enabled.

Publishing Your Agent

Users Access

Open

This section controls who can interact with your agent. There are two access modes available:

Public Endpoint

Critical Security Warning
DO NOT enable the "Allow anonymous users access" toggle under any circumstances!

Enabling public endpoint access would allow anyone on the open internet, including unauthorized users and bots, to utilize UT Spark services without authentication. This violates university policies and creates significant security and cost risks.

Important: UT Spark is exclusively intended for use by Faculty, Staff, and Students of The University of Texas at Austin.
Public access is strictly prohibited.

Always keep this toggle OFF.

Private Audience

This is the recommended and required access mode for all agents. When you enable "Allow authenticated users access," you can control exactly who within the UT Austin community can use your agent.

Access Options:

• All UT Austin Faculty, Staff, and Students - Anyone with valid UT Austin credentials can access your agent

• Custom Defined Groups - Restrict access to specific groups, departments, or roles within UT Austin (e.g., "College of Engineering Students", "Research Lab Members")

Note:
Before making your agent accessible to all UT Austin faculty, staff, and students, please consult with the AI Studio team (Enterprise Technology - Learning Technology). Broad campus-wide deployment may have implications for support, security, and scalability. The AI Studio team can help you evaluate best practices and determine whether wide release is appropriate for your agent’s intended use.

How to configure:

1. Enable the "Allow authenticated users access" toggle

2. Define your audience by selecting appropriate user groups or leaving it open to all authenticated UT Austin users

Token Quotas

The Limit token usage setting allows you to manage and control the computational resources your agent consumes. This helps prevent unexpected costs and ensures fair resource allocation.

When enabled, you can configure:

Agent-Level Limits:

• Maximum total tokens the agent can generate across all users combined

• Warning message displayed when the agent approaches its quota

• Behavior when the agent reaches its maximum limit

Per-User Limits:

• Maximum tokens any individual user can consume

• Warning message shown to users approaching their personal quota

• Restrictions applied when a user reaches their limit

Open

Use cases:

• Control costs for high-usage agents

• Ensure fair distribution of resources among users

• Prevent abuse or runaway token consumption

Tip
Enable token quotas for production agents, especially those with broad access, to maintain predictable resource usage.

Management Access

This section defines who can edit and manage your agent after it's published.

By default, the agent creator has full management access. You can add additional managers by:

1. Clicking "+ User Role"

2. Selecting specific Admin User Roles (predefined groups with management permissions)

3. Adding individual users or groups who should be able to modify the agent

Important: Management access is separate from user access. Users with management access can:

• Modify agent configuration and instructions

• Change capabilities and knowledge sources

• Update user experience settings

• Manage access controls

• Delete the agent

Best practice: Only grant management access to trusted individuals who need to maintain or update the agent.

Agent Status

The Agent Status toggle is the master on/off switch for your agent. This control determines whether users can actually interact with your agent.

When ENABLED (toggle ON):

• Users defined in the Private Audience can access and chat with your agent

• The agent appears in the available agents list for authorized users

• All configured features and capabilities are active

When DISABLED (toggle OFF):

• The agent is effectively offline and inaccessible to all users

• Previously authorized users can no longer interact with it

• The agent remains saved with all its configurations intact

Use cases for disabling:

• Temporarily take an agent offline for maintenance or updates

• Pause agent availability during testing or troubleshooting

• Deactivate an agent that's no longer needed without deleting it

Note: You can toggle the agent status on and off at any time without losing your configuration.

Finalizing and Publishing

Once you've carefully configured all publishing settings:

1. Verify that Public Endpoint is disabled

2. Enable Private Audience access and define your authorized users

3. Configure token quotas if needed for resource management

4. Set management access for any additional administrators

5. Enable the Agent Status toggle to make your agent live

6. Click "Save Changes" to publish your agent

Congratulations! Your agent is now live and ready to assist your users!

Managing Your Published Agent

After publishing, you can return to this screen at any time to:

• Adjust access controls

• Modify token quotas

• Update management permissions

• Toggle the agent on or off

• Make changes to any configuration from previous steps

Once you have completed your agent and tested it, please fill out the form below and the UT Spark team will look over the agent, and enable it for the AD group that you specified on the original request. The UT Spark team can help with tuning the agent or any questions that might arise during the process reach out to utspark-support@austin.utexas.edu.

Activation Form