This article provides guidance on content configuration when creating and configuring the intelligent analysis Agent.

The configuration content is divided into four modules: data retrieval recommendation questions, analysis recommendation questions, PC, and Lark.

• Data retrieval recommendation questions: Supports setting data retrieval-related recommendation questions for users or Lark groups.
• Analysis recommendation questions: Supports setting data analysis-related recommendation questions.
• PC: Displays summary messages to users when they enter the Agent homepage on the PC.
• Lark: Displays a welcome card and proactive push messages to users when they enter the Agent bot on Lark.

1. Go to the Agent configuration page, select "Content configuration" on the left, and then enter the data retrieval recommendation questions configuration page.
   

2. After entering the editing page, you must first select a configuration method. Currently, "Unified user configuration" and "User-specific configuration" are supported.
   

   • "User-specific configuration" allows you to configure recommendation questions individually for each Agent user.
   • "Unified user configuration" means all users receive the same recommendation questions.

3. Click "Add recommendation question configuration" and configure the recommendation questions in the configuration pop-up window. You can drag to adjust the order. The order of the questions here determines the order in which recommendation questions are displayed on subsequent Agent usage pages.
   
   The relevant parameter configuration instructions are as follows:

   Parameters	Configuration instructions
   Configuration name	Set the configuration name for recommended questions.
   Recommended questions - Add fuzzy questions	Supports configuring multiple fuzzy questions. Fuzzy questions support configuring detailed sub-questions and associating sub-questions with datasets. tip The intelligent analysis agent also provides an intent clarification feature. When the user's question is unclear, the agent can ask clarifying questions based on intent clarification rules to help further specify the question. For details, see Intent Clarification Configuration. For scenarios involving data queries, it is recommended that the question include key components such as the query time and filter conditions. For details, see Best Practices: Recommendations and Examples for Intelligent Data Query Questions. Each sub-question must be bound to a dataset. The intelligent agent will retrieve data based on the dataset bound to the sub-question to provide answers. When a fuzzy question is configured with multiple sub-questions, clicking a question will display the query results for all sub-questions simultaneously. Supports enabling the generalization switch for fuzzy questions. Configure the decomposition approach for a question. After enabling question generalization, the decomposition approach will also apply to similar questions. For example: After configuring the decomposition approach for "Analyze the sales status of each region", enabling generalization allows the decomposition approach to be reused for "Analyze the sales status of each province/city".
   Recommended questions - Add non-fuzzy questions	Supports configuring multiple non-fuzzy questions and associating non-fuzzy questions with datasets.

On the Agent configuration page, select "Content configuration" on the left, and then enter the analysis recommendation questions configuration page. Click "Add question" and set the recommendation questions and select associated knowledge in the pop-up parameter configuration interface. After configuration, click "Save".

Go to the Agent configuration page, select "Content configuration" on the left, find "PC", and click Edit in the upper right corner to configure the copy displayed on the PC homepage. After entering the content, you can preview the display effect.

Users must first complete Lark Platform Agent Configuration and Testing before accessing the agent configuration page. On the left, select "Content configuration" to view the "Lark endpoint" page. Here, you can configure the Lark endpoint welcome card and the content for proactive push notifications.

Welcome card #

The Lark endpoint agent welcome card is shown to users the first time they use the agent. You can add an introduction to the agent here.
After configuring a card in the Lark card builder tool, users can copy the JSON code into the welcome card settings for quick configuration. For Lark card configuration, see Lark card overview.

Create a proactive push task #

The Analysis Agent supports proactively pushing in-depth research reports via Lark cards and allows enabling the MCP switch during the push. This section describes how to configure Lark proactive push cards.

Prerequisites #

To enable the MCP switch during a push, ensure that the MCP switch is enabled at the system or agent level. For configuration details, see MCP Service Management.

Procedure #

1. Select the relevant agent from the agent list, then click the Agent configuration button below the agent to access the agent configuration page.

2. In the side navigation, click Content configuration, then click the Lark endpoint tab. Below the welcome page, above the Lark endpoint proactive push task list, click the Create proactive push task button to enter the configuration interface.

3. Select the push type to configure different settings for the proactive push task. The following explains the configuration of push content for different push types:

   Configuration items	Specific name of configuration item	Note
   Push type	Select push type	In-depth research, insight report, or hybrid push.
   Push content	Push card title	Set the title displayed on the Lark message card to help recipients quickly identify the task topic. Custom push card title.
   	Push card description	A brief explanation of the push content, which can supplement background information or key tips.
   	Push content	Enter the specific requirements for this push task, for example, "Please analyze last week's sales performance in the East China region and provide optimization suggestions."
   	Push format	Select the delivery method for the content. Multiple selections are supported: • Clickable link: Push a clickable link. The recipient can click to be redirected to the details page. • Screenshot: Push the report or analysis results in image format. • PDF download link: Push a link to download a PDF file.
   	Push report	This field must be filled in when the push type is an insight report. Used to select a pre-configured report template. The system will automatically generate the insight report content based on the template.
   	Push summary	This field is optional when the push type is a mixed push. Enter an overall summary or key conclusion for this push content to highlight the core information in the message card.
   Push user	Permission basis	Required. Determines the basis for task viewing or execution permissions: • Creator: Send content with the task creator's permissions. • Recipient: Send content with the specified recipient's permissions (up to 5 recipients can be selected).
   	Target user	Required. Click "Search Lark user" to select individual users from the Lark address book who need to receive the push.
   	Push Lark group	Enter the Lark group name to send the push content to the specified group chat.
   	Shared push	This field is optional when the push type is a mixed push. Supports adding collaborators to share the push task with other users or teams, facilitating collaborative viewing and management.
   Push frequency	Push interval	Select the push cycle: daily, weekly, or monthly. Choose the specific time or date according to the selected cycle.
   	Push timing	Determines the execution time point of the task: • Immediate push: Execute and push immediately after saving. • Push after data is ready: Automatically pushes after waiting for data updates or calculation completion (suitable for scenarios that depend on T+1 data or require complex calculations).

Lark endpoint proactive push task list #

Successfully created proactive push tasks are displayed in a list. Here, you can view the task configuration details, enable or disable tasks, and perform various operations on the tasks. The following provides a detailed explanation of the list items:

{"pushConfig":{"enablePushImage":true,"imageType":"html","enablePushUrl":true,"enablePushSummary":true,"customSummaryPrompt":"总结成以下内容：观察期<YYYY-mm-DD>~<YYYY-mm-DD> VS 基准期<YYYY-mm-DD>~<YYYY-mm-DD> \n• 观察期【<指标名>】，对比基准期<上涨/下跌><指标变化值>，变化率 <指标变化率>\n"},"questionType":"deep_research","contentParams":{},"name":"今年是 2025 年。\r\n基准周期日期 和 观察周期日期：前天 vs 昨天\r\n指定维度：xxx","agentConfigs":{"enabledMCPServers":[{"mcpServerName":"归因工具","tools":[{"name":"dimension_attribution","selected":true},{"name":"formula_attribution","selected":false},{"name":"公式拆解"},{"name":"trend_detection","selected":true}]}]}}

The syntax and parameter description for the proactive push summary JSON script are as follows. For your convenience in viewing the JSON script, line breaks and indentation are used here. When actually filling in the push configuration, the script content must be merged into a single line.

{
  "pushConfig": {
    "enablePushImage": true, // Whether the card needs to include a report screenshot in html/markdown format
    "imageType": "html", // Report screenshot format, set to markdown or html
    "enablePushUrl": true, // Whether the card needs to include a "Full Report" link, which can directly jump to the report display page
    "enablePushSummary": true, // Whether the card needs to include a section summarizing the main content of the card
    "customSummaryPrompt": "" // Used to prompt the large model to generate summary content on the card
  }, // Required
  "agentConfigs": {
    "enabledMCPServers": [
      {
        "mcpServerName": "<拓展 mcp 名字>",
        "tools": [
          { "name": "<mcp 工具名称>", "selected": true } // Whether a tool is selected
        ]
      }
    ]
  },
  "questionType": "deep_research", // Required deep_research
  "name": "<深度研究问题>", // Required
  "contentParams": {} // Must include
}

After completing the configuration, you can send a preview. Click "Send Preview" to view the result of the push configuration.