Create and manage data agents

Data agents let you curate the Conversational Analytics experience for your users. With agents, you can provide Conversational Analytics with context and instructions to enable it to answer questions more effectively for specific use cases. Agents empower analysts to map business terms to specific fields, specify the best fields for filtering, and define custom calculations.

This page guides you through the following processes:

• Create and edit data agents
• Share data agents with other users
• Publish data agents to other applications
• Delete data agents

Learn how and when Gemini for Google Cloud uses your data.

Before you begin

Think of a data agent as another type of Looker content — like a dashboard, Look, or folder.

The use of a data agent is managed through a combination of content access, data access, and feature access. To perform the tasks that are described in the following table, you must be assigned a Looker role that has the required permissions for the models that your data agent will query and, in some cases, access to the agent itself.

Task	Required Looker permissions	Required level of data agent access
Create, edit, share, and delete agents	Added 25.18 admin_agents	No content access must be granted
Create, edit, share, and delete agents	Added 25.18 save_agents Users can create agents that use only the Explores for which they have been granted this permission on the underlying model. To edit, delete, or share a data agent that was created by another user, users must be granted a role that contains this permission on every model that is used by the agent.	Manage access; Edit (this access is granted automatically if the user creates the agent; otherwise, Manage access; Edit access must be granted by the agent's creator by sharing the agent)
Publish agents to other Google Cloud applications	Added 26.6 publish_agent_externally Users publish existing agents to other Google Cloud applications, such as Gemini Enterprise.	Manage access; Edit (this access is granted automatically if the user creates the agent; otherwise, Manage access; Edit access must be granted by the agent's creator by sharing the agent)
Chat with a data agent from the Agents tab in Conversational Analytics	access_data (on each model that contains the Explores that are used by the data agent) Added 25.18 chat_with_agent (on each model that contains the Explores that are used by the data agent)	View access
Chat with a Looker Explore from the Explores tab in Conversational Analytics	access_data (on each model that contains the Explores that are used by the data agent) Added 25.18 chat_with_explore

Looker also has the following default roles that contain subsets of these permissions for all models on the instance:

• Conversational Analytics Agent Manager: With this role, a user can create, edit, share, delete, and chat with agents that they have Manage access; Edit access for and chat with Explores
• Conversational Analytics User: With this role, a user can chat with an agent that they have View access for
• Admin: By default, this role (Looker Admin) has all permissions and content access across the instance.

A Looker admin can grant these roles and permissions on the Roles page in the Admin section of the Looker instance. For more information about Looker roles, see the Admin settings – Roles documentation page.

The creator of the data agent can manage individual users' access to the agent by sharing the agent.

Create and edit data agents

As an early-stage technology, Gemini for Google Cloud products can generate output that seems plausible but is factually incorrect. We recommend that you validate all output from Gemini for Google Cloud products before you use it. For more information, see Gemini for Google Cloud and responsible AI.

To create a new data agent, follow these steps:

1. Navigate to the Conversations page.
2. In the Agents tab, select + New agent. Or, in the left panel, select sparkManage agents, and then select + New agent.

3. On the New agent page, provide the following information about your data agent.

   • Agent name: Enter a name for the agent. The name should be unique and descriptive.
   • Agent description: Briefly describe what this agent can do and the data that it uses. Users will see this description when they select the agent to start a conversation or when you share the agent with them, so make sure that the description clearly explains the agent's purpose and how it can be helpful.
   • Data: Follow these steps to connect to up to five existing Looker Explores:
     1. In the Data field, click + Select Explores.
     2. In the Search Explores window, click the Explores that you want to include in the data agent. These Explores will appear in the Selected Explores panel of the window.
     3. To add the selected Explore to the data agent, click Save.

4. Instructions: Provide context to help Conversational Analytics understand how to interact with your data and provide accurate and relevant responses. For examples of the types of context that you can provide, see Writing agent instructions.

5. Optionally, to enable the Code Interpreter for all conversations with your agent, select Enable Advanced Analytics.

6. Optionally, you can test your agent to refine your instructions and its settings.

7. To save your new data agent, click Save.

After you save the data agent, you can share the agent with other users and start a conversation with the agent.

Write agent instructions

When you create a data agent, you can add free-form instructions that define your data agent's core behavior and provide it with foundational context to consider before processing a user's prompt.

Important: Although the Instructions field is optional and the structure is up to your discretion, we recommend that you provide well-structured instructions to enable the agent to give more accurate and relevant responses.

Here are some examples of the types of context that you can provide in the Instructions field:

• Key fields: The most important fields for analysis
• Excluded fields: Fields that the data agent should avoid
• Filtering and grouping: Fields that the agent should use to filter and group data
• Golden queries: Pairs of natural language questions and their corresponding Explore queries
• Persona: A role or expertise, character, or tone that you assign to the agent

Note: The types of authored context that you can provide a data agent in Looker differ slightly from those that you can provide an agent that you are creating using the Conversational Analytics API. Some of what you provide in the Instructions field in Looker — such as golden queries — can be given to an agent as structured context using the Conversational Analytics API.

For tips and best practices for writing agent instructions, see the Best practices for configuring Conversational Analytics in Looker.

Define a Looker golden query

To define each Looker golden query for a given Explore, provide values for both of the following fields:

• natural_language_questions: The natural language question that a user might ask
• looker_query: The Looker golden query that corresponds to the natural language question

Tips for defining Looker golden queries:

Include different types of questions and queries that have a variety of filters and filter values.

Although there is no limit to the number of golden queries that you can include to looker_golden_queries, we recommend including no more than 30-50 question-and-query pairs.

For the natural_language_questions field, consider the questions a user might ask about that Explore, and write those questions in natural language. You can include more than one question in this field's value. You can obtain the value for the looker_query field from the Explore's query metadata.

Golden queries support the following fields:

• model (string): The LookML model that was used to generate the query. This is a required field.
• explore (string): The Explore that was used to generate the query. This is a required field.
• fields[] (string): The fields to retrieve from the Explore, including dimensions and measures. This is an optional field.
• filters[] (string): The filters to apply to the Explore. This is an optional field.
• sorts[] (string): The sorting to apply to the Explore. This is an optional field.
• limit (string): The data row limit to apply to the Explore. This is an optional field.

Note: Conversational Analytics doesn't support golden queries that contain pivots or custom fields.

You can retrieve the Explore's query metadata directly from the Explore by following these steps:

1. In the Explore, select the Explore actions menu, and then select Get LookML.
2. Select the Dashboard tab.
3. Copy the query details from the LookML. For example, the following image shows the LookML for an Explore called Order Items:

Copy the selected metadata for use in your Looker golden query:

  model: thelook
  explore: order_items
  fields: [order_items.order_id, orders.status]
  sorts: [orders.status, order_items.order_id]
  limit: 500

Test an agent

When you're creating or editing an agent, the agent details page includes the Preview your agent pane. You can test agent settings and instructions by starting a conversation with the agent.

You must click Update for a change to be reflected in the preview. If the save status is Not saved, any updates to settings won't be reflected in the preview.

Edit an existing data agent

To edit an existing data agent, follow these steps:

1. On the Conversations page, select sparkManage agents.
2. On the Manage agents page, select the data agent that you want to edit.
3. Update the details about the agent as needed. You can modify the details that you specified when you created the agent, including the Agent name, Agent description, Data, and Instructions fields. You can also opt to enable the Code Interpreter for your agent.
4. To save your changes, click Update.

Share data agents

Sharing lets other users chat with your agent and its Explores. You can share a data agent with other users by granting content access to the agent. Only a user with the appropriate permissions and content access can share an agent. Once an agent is created, it may take a few minutes for it to become shareable.

Note: Granting a user content access to an agent does not also grant that user access to the Explore that the agent uses. When you share an agent, make sure that the users also have the permissions that they need to use Conversational Analytics with the model that the Explore is based on. Your Looker admin may need to grant these permissions.

To share a data agent, follow these steps:

1. On the Conversations page, select sparkManage agents in the left panel.
2. Open the menu for the chosen agent by clicking its More options icon, and then click Share.
3. Once you have added an individual or groups to the Who can access this agent section, and chosen what level of permissions they should have, click Add to place them in the shared list.
4. If you want new users or groups to receive a notification email, select the Email the people you have just added checkbox.
5. After all changes have been made, click Save.

You can also share an agent that you've just created or one that you're editing by clicking Share on the agent settings page and following the aforementioned steps.

Revoke access to a data agent

To revoke access to an agent, follow these steps:

1. On the Conversations page, click sparkManage agents in the left panel.
2. Open the menu for the chosen agent by clicking its More options icon, and then click Share.
3. Click the X next to the user or groups that should have their access removed.
4. After all changes have been made, click Save.

If the removed users have an ongoing conversation, they will still have access for a minute or two while the changes propagate.

If a user tries to ask more questions once access to an agent is removed, then that user will see the message The agent in this conversation may not be shared with you, or may have been deleted. You can view any past conversations with the agent, but can't ask new questions.

Publish a data agent

Preview

This product or feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms. Pre-GA products and features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

You can publish your data agent in Gemini Enterprise, making your agent's predefined context and analytics available to a broader audience, including users who may not be familiar with the Looker interface. Gemini Enterprise serves as a central hub where users can manage data agents that were created across Google Cloud applications, which helps users to find the best agent for their analytical needs.

The procedure to create Conversational Analytics data agents in Looker and then publish them for chatting with in Gemini Enterprise involves tasks that are performed in the Looker instance, on the Gemini Enterprise page in the Google Cloud console, and in the Gemini Enterprise instance.

To create a data agent in Looker and make it available to users in Gemini Enterprise, users with the roles in brackets must follow these steps, which are described in more detail in the following sections:

1. [Data agent editors]: Publish the agent in Looker.
2. [Discovery Engine Admin]: Enable the agent in the Google Cloud console.
3. [Discover Engine User]: Chat with the agent in the Gemini Enterprise instance.

Note: The Looker – Gemini Enterprise integration is in preview with limited support. We encourage you to share your feedback to help us improve. To report bugs or issues, send an email to [e-mail] with the following details:

• A clear description of the problem and the expected behavior
• Steps to reproduce the issue
• Any additional relevant details

Before you begin

Before you can publish an agent, the following requirements must be met:

• A Looker admin must enable the Publish to Gemini Enterprise setting on the Gemini in Looker page in the Admin panel.
• A Looker admin must grant you a role that contains the publish_agent_externally permission. You must also be able to edit a data agent.

Publish a data agent to Gemini Enterprise

To complete the steps that are described in this section, you must have a Looker role that contains the publish_agent_externally permission and the appropriate permissions and content access to edit data agents.

To publish an agent, follow these steps:

1. Edit the agent.
2. Select Publish settings to open the Agent publish settings window.
3. In the Agent publish settings window, enable the Gemini Enterprise setting and select Update. Enabling this setting publishes the agent to the Gemini Enterprise instance that is displayed on the Gemini in Looker page in the Admin panel.
4. Once you've published your agent, you can select Share to grant other Looker users access to your data agent, or select Start conversation to open a new conversation with the data agent.

Add the agent in Gemini Enterprise

To complete the steps that are described in this section, you must have the Discovery Engine Admin Identity and Access Management role for the Google Cloud project that contains the Gemini Enterprise engine.

The Discovery Engine Admin must complete the following steps in the Google Cloud console:

• Import the agent into the Gemini Enterprise instance.
• Grant users access to the data agent.

To import a data agent into a Gemini Enterprise instance, follow these steps in the Google Cloud console:

1. Navigate to the Google Cloud project that contains the Gemini Enterprise engine.
2. Open the Gemini Enterprise page.
3. Select Agents in the main navigation.
4. Select + Add agent. This action opens the Add an agent window.
5. In the Add an agent window, under Import from agent registry, find the agent and select + Enable.

To grant users access to the data agent, follow the instructions to share an agent. Users must be granted the Discover Engine User role to access and chat with the agent.

Chat with the agent in Gemini Enterprise

To complete the steps that are described in this section, you must have the Discover Engine User Identity and Access Management role for the Google Cloud project that contains the Gemini Enterprise engine.

To chat with a data agent in a Gemini Enterprise instance, follow these steps:

1. Open your Gemini Enterprise instance.
2. Select Agents in the main navigation.
3. On the Agents page, select that agent that you would like to chat with. When you select an agent, a conversation window opens where you can ask questions about your data.

Delete a data agent

Caution: Any users with whom you have shared this data agent won't be able to ask new questions after you delete the agent; however, existing conversations with the data agent are retained.

If you delete an agent, you will also delete and unpublish the agent in Gemini Enterprise.

To delete a data agent, follow these steps:

1.On the Conversations page, click sparkManage agents in the left panel. 1. Open the menu for the chosen agent by clicking its More options icon, and then click Delete. 1. In the Delete agent? window, click Move to trash to delete the data agent.

Agents that are moved to the trash will be permanently deleted after 30 days. You can permanently delete a data agent manually, or you can restore a data agent from the trash before it is permanently deleted. If you take no action, the agent will be deleted permanently after 30 days automatically.

Permanently delete a data agent

To permanently delete a data agent, follow these steps:

1. Navigate to the Conversations page.
2. In the left navigation panel, expand the Trash section.
3. To open the menu for the chosen agent, select its icon, and then click Delete Permanently.
4. In the Are you sure? window, click Delete forever.

Restore a data agent from the trash

To restore a data agent from the trash, follow these steps:

1. Navigate to Conversations page.
2. In the left navigation panel, expand the Trash section.
3. To open the menu for the chosen agent, select its icon, and then click Restore.

Related resources

• Conversational Analytics in Looker overview: The landing page for Conversational Analytics with a list of key features links to all Conversational Analytics documentation.

• Best practices for configuring Conversational Analytics in Looker: Strategies and best practices to help Looker administrators and LookML developers successfully configure optimize Conversational Analytics.

• Enable advanced analytics with the Code Interpreter: The Code Interpreter within Conversational Analytics translates your natural language questions into Python code and executes that code. Compared to standard SQL-based queries, the Code Interpreter's use of Python enables more complex analysis and visualizations.