Data Agent for Analytics

Use cases

Business users: Define data requirements in plain language and generate reports on sales trends, user behavior, and more.
Data analysts: Run exploratory analysis, offload tasks like automated feature engineering and data cleaning to the agent, and reuse generated Python code.
Managers: Gain quick insights into key business indicators for data-driven decisions.

Features

Automated insight discovery: Describe requirements in natural language. The agent interprets your data, discovers insights, and generates a report.
Faster time to value: Focus on business outcomes, not complex data workflows.
Seamless data connection: Natively supports ApsaraDB and DMS-managed data sources for instant analysis.

Integrations

Core component: Data Agent Skill
One-click integration: Install from a Git repository or run the following command to quickly deploy and load the agent.
```
npx skills add https://github.com/aliyun/data-agent-skill
```
Feature summary: Download and integrate this skill package into your agent service. It builds a secure bridge to your enterprise data sources through standard connection protocols, enhancing data access and analytical depth for custom data intelligence applications.

Limitations

Data Agent for Analytics is currently available only in the following regions: China (Hangzhou), China (Shanghai), China (Shenzhen), China (Chengdu), China (Beijing), China (Zhangjiakou), China (Hong Kong), and Singapore.

Interface layout

Initial interface:
- Data Center: Add and upload data for analysis.
- All Tasks: View the history of your analysis tasks.
- User information: View details of the current account.
- Chat area: Start a new analysis task and interact with the agent.
Workspace interface:
- Chat area: Continue your conversation with the agent during the analysis.
- Execution area: View the agent-written code and its corresponding results for each step.

Procedure

Log on to the Agent console.
Choose an edition. Compare options in Data Agent editions.
- Free Edition:
  
  On the Free version card, click Start Free Trial.
- Personal Edition:
  - On the Personal card, click Upgrade to Personal Edition.
  - Select a region, the number of seats, and a subscription duration.
    
    Note
    In the Subscription Duration section, select a monthly duration.
  - Click Buy Now and complete the purchase.
- Enterprise Edition:
  - On the Enterprise card, click Upgrade to Enterprise Edition.
  - Select a region, the number of Large Language Model (LLM) resource packages, and a subscription duration.
    
    Note
    In the Subscription Duration section, select a duration from 1 to 11 months.
  - Click Buy Now and complete the purchase.
Upload a data sample from a local file or use existing data.
- Click to upload a local file. Supported formats: CSV, XLSX, and XLS (up to 200 MB).
- Click to use existing data. Add data in the Data Center.
Describe your analysis requirements in the chat area and press Enter or click to start the analysis.
Review the execution plan generated by the agent.
- If the plan is satisfactory, click Start Task.
- If not, click Modify Tasks to provide additional details until the plan meets your needs.
Wait for execution. You can view the agent's code and results for each step.
- Code written by the agent:
- Code execution result:
The agent can generate a visual web report to present data insights.

Data Center

In the left-side navigation pane, click Data Center. Click Add data and select a connection method.

Local upload:

Drag a file to the upload area or click to select one. After upload, click Confirm.
The Free and Personal Editions support RDS Database, PolarDB database, and AnalyticDB Database. The Enterprise Edition supports 40+ data sources listed in Databases Supported by DMS.
1. Select a region and an instance.
2. Enter the Database user name and Database password.
3. Click Test link to continue.
4. Select the database and table.
5. Click Confirm.

Long-term memory

The agent automatically identifies and saves key information from your conversations as long-term memory. This memory is recalled in future sessions to better understand your business requirements, improve analysis results, and provide a more contextual experience.

Manage memory settings

This feature is enabled by default. To stop automatic memory storage, disable it.

Go to the Memory Management page.
Find the Generate memory switch and turn it off.

Memory list

On the Memory page, you can view and manage all stored memories:

Field	Description
Source	The conversation where the memory was generated. Click the link to view the original conversation.
Content	The key information the agent extracted from the conversation.
Heat	A score based on recall frequency. A higher heat value indicates the memory is more actively used in your analyses.
Status	Memories have two statuses: • `Remembered`: The memory is active and can be recalled in subsequent analyses. • `Forgotten`: The memory is inactive and will not be recalled.
Operation	Edit or delete memories in the Actions column: Edit: Click the edit icon to correct or update a memory. In the dialog box that appears, you can manually correct or complete the memory content and then save it. This ensures that the agent learns the most accurate information. Delete: Click the delete icon to remove a memory. The effect depends on the memory's status: For a `Remembered` memory: The status changes to `Forgotten`. The record remains in the list and can be reactivated. For a `Forgotten` memory: The memory is permanently removed and cannot be recovered.

Data privacy

The agent does not use your data — including uploaded files, queries, analysis results, and reports — for model training or iteration.
Your data is processed in an exclusive computing instance within your account. The instance and its data are destroyed after the task completes. Other accounts cannot access your data, processes, or history.
For local file analysis, uploaded files are stored in an isolated environment not visible to other accounts.
For database analysis, data stays in your database instance. No data is migrated during computation. All intermediate results are destroyed after analysis.