ADBAgent Getting Started Guide

Three-layer data exploration

ADBAgent uses an on-demand, progressive strategy that standardizes data exploration into three layers, delivering full-pipeline data insights at minimal cost.

Core value: Simple queries get instant responses, while complex questions receive deep insights — making data analysis more accurate and efficient.

Industry-specific analysis Skills

In the advertising and attribution analytics domain, ADBAgent includes built-in industry-level Skills such as conversion rate analysis and campaign diagnostics. These Skills form a comprehensive diagnostic framework from macro trends to micro root causes to strategy recommendations.

Step 1: Create a cluster

1. Log on to the AnalyticDB for MySQL console. In the upper-right corner, click Buy Cluster.

2. On the buy page, configure the following key parameters. You can leave other parameters at their default values or configure them as needed. For more information about all parameters, see Create a cluster.

   • Product Type: Select Pay-As-You-Go.

   • Edition: Select Enterprise Edition (Multi-replica).

   • Deployment Mode: Select Single-zone Deployment.

   • Region and Primary Zone: ADBAgent is available only in the China (Hangzhou), China (Shanghai), and China (Beijing) regions.

   • Auto-enable PVL Service: Select the check box.

   • Reserved Resource Nodes: Set as needed. When set to 0, only external tables are supported. Internal tables cannot be created or read/written.

3. Click Buy Now and follow the on-screen instructions to complete the purchase.

   After payment, wait approximately 10 to 15 minutes. On the Clusters page, when the cluster status changes to Running, the cluster is ready.

Step 2: Create a database account

ADBAgent requires a database account to connect to AnalyticDB for MySQL clusters. Different account types have different permissions.

1. On the cluster list page, click the ID of the target cluster to go to the cluster details page.

2. In the left-side navigation pane, click Accounts.

3. On the Database Accounts tab, click Create Account. Create the following two account types:

   1. Privileged account: Used to manage permissions for standard accounts.

   2. Standard account: This account will be used for the ADBAgent database connection.

4. Configure permissions for the standard account.

   Click the ⋮ icon in the Actions column of the standard account row, and then click Edit Permissions. We recommend that you grant the SELECT permission.

Step 3: Initialize ADBAgent

1. In the left-side navigation pane of the cluster details page, click ADBAgent (AI Assistant).

2. Click Free Trial.

   The system automatically initializes on first access. This takes approximately 2 minutes.

Step 4: Configure a whitelist

Before you access ADBAgent, you must enable public network access for the ADBAgent service and configure a whitelist to restrict access to specific IP addresses.

1. On the ADBAgent (AI Assistant) page, in the upper-right corner of the Whitelist section, click Add Whitelist Group.

   

2. In the Add Whitelist Group dialog box, configure the following settings.

   • Group Name: Enter a custom name.

   • Whitelist: Enter the public IP address or IP range of the machine that needs to access ADBAgent. To find your public IP address:

     • Run the following command in the shell: curl ifconfig.co/json

     • Or visit this website in a browser: https://ipconfig.com/zh

3. Click OK.

Step 5 (Optional): Enable internet access

A public NAT gateway is required when you use any of the following ADBAgent capabilities:

• Channel integration: Connect to messaging tools such as DingTalk or Feishu.

• Intelligent diagnostics: Use the AnalyticDB for MySQL AI Diagnosis Skill for cluster health checks.

• Public data sources: Connect to data sources over the public network.

Perform the following steps:

1. Obtain the VPC and vSwitch information of the AnalyticDB for MySQL cluster.

   1. Log on to the AnalyticDB for MySQL console. In the upper-left corner of the console, select a region. In the left-side navigation pane, click Clusters. Find the cluster that you want to manage and click the cluster ID.

   2. In the left-side navigation pane, choose Cluster Management > Cluster Information.

   3. In the Network Information section, find the VPC ID and vSwitch ID.

2. Create a public NAT gateway and bind it to an EIP.

   1. Go to the NAT Gateway - Buy Public NAT Gateway page.

   2. Configure the public NAT gateway.

      • Region: Select the region where the AnalyticDB for MySQL cluster is located.

      • Network and Zone: Select the VPC and vSwitch obtained in Step 1.

      • EIP: Select Select EIP if you already have an EIP, or Purchase EIP to create one.

   3. Click Buy Now.

3. Configure an SNAT entry.

   1. Go to the Public NAT Gateway console. In the left-side navigation pane, choose NAT Gateway > Internet NAT Gateway, and then click the NAT gateway you just created.

   2. Click the SNAT tab, and then click Create SNAT Entry.

   3. Select EIP, and then click OK.

Step 1: Access ADBAgent

1. On the ADBAgent (AI Assistant) page, click Credentials in the Actions column of the target service to view the Default Account and Default Password.

2. Click Access. Log on to ADBAgent with the default account and password.

Step 2: Configure a data source

On the ADBAgent page, in the left-side navigation pane, click Data Connection to go to the data source configuration page. ADBAgent supports multiple data sources for analysis.

Step 3: Run an analysis task

In the left-side navigation pane, click New Task. Describe your analysis requirements in the input box and send the message. ADBAgent automatically interprets your intent, matches Skills, performs layered data exploration, executes SQL, and presents the results in a structured format.

• Specify a scenario and Skills (Optional)

  Click the + button in the input box to select an analysis scenario, or type /SkillName to invoke a specific analysis Skill. The system then runs the task within the corresponding data scope and analysis logic. For information about configuring analysis scenarios and Skills, see the following sections.

  

• Multi-turn conversations

  Complex data analysis scenarios support multi-turn conversations:

  • Follow-up analysis: Ask follow-up questions based on the results to uncover deeper data insights.

  • Error correction: Point out issues when results do not match expectations. The system automatically adjusts its analysis logic.

  • Additional requirements: Add new analysis dimensions or filters at any time during the analysis.

Step 4: Save analysis workflows as Skills

You can save analysis workflows as reusable Skills. During an analysis task, describe the action in natural language (for example, "Summarize this task as a Skill and add it to my Skills"). The complete analysis logic is saved as a Skill that you can invoke directly in future tasks.

Step 5: Save as an analysis scenario

For frequently used business scenarios, you can save them as reusable analysis scenarios to reliably reproduce your analysis workflows.

1. In the left-side navigation pane, click Business Scenarios to go to the Business Scenarios (Plugin) page.

2. Click Create Scenario Plugin and enter a scenario name and description.

3. Associate the target data tables (you can select tables directly from a database).

4. Bind the required Skills.

5. After saving, select this scenario when creating a new task to restrict the analysis scope and Skills, ensuring consistency in analysis logic and metrics.

Prerequisites

Before OpenClaw can call ADBAgent, you must add the public IP address of the machine running OpenClaw to the ADBAgent whitelist. For more information, see Step 4: Configure a whitelist.

Channels integration

If your OpenClaw is already connected to messaging tools such as DingTalk or Feishu, you can interact with ADBAgent through OpenClaw directly from those tools.

Integration steps

1. Obtain the Skill

   ADBAgent integrates with OpenClaw through a Skill. Contact the development team to obtain the Skill package. After downloading, extract it and place it in the OpenClaw Skills directory.

   Reference path: ~/.openclaw/workspace/skills/

2. Configure access information

   Enter the ADBAgent access information in the config.json file in the ADBAgent Skill directory. Contact the development team for the access information.

3. Verify Skill recognition

   Check whether OpenClaw has automatically detected the ADBAgent Skill by using one of the following methods:

   • OpenClaw Dashboard: Skills > Refresh > Workspace Skills.

   • Shell command: openclaw skills.

   If the Skill is not detected, try restarting OpenClaw: openclaw gateway restart.

4. Start using

   Send natural language commands directly in OpenClaw. For example:

   • "Use ADBAgent to analyze xxx"

   • "Use ADBAgent to analyze xxx, using the xxx database table and the xxx Skill."

Output modes

The ADBAgent Skill uses Summary mode by default, showing only key information from the analysis process. To view detailed intermediate steps and SQL execution results, use Verbose mode by specifying it in natural language. For example:

• "Use ADBAgent in verbose mode to analyze xxx"

• "Use ADBAgent, show me the detailed process, and analyze xxx"