The new DataWorks Data Agent, powered by Qwen Code, is now generally available on the public cloud. It gives you a dedicated full-page environment for complex, multi-step data tasks — covering Data Integration, data development, Data Governance, and Data Security. It plans, writes, runs, and self-corrects code automatically, with built-in DataWorks Skills out of the box.
Core capabilities
The new Data Agent is a next-generation agent for the DataWorks platform, built on Qwen Code. Unlike the original agent — a sidebar chat panel — the new Data Agent gives you a dedicated, full-page environment built for complex, multi-step data development and governance tasks that require deep, sustained interaction.
The new Data Agent provides the following core capabilities:
Deep task understanding and planning: Powered by Qwen Code, it accurately interprets complex data development and data governance requests, then automatically breaks them down into multi-step execution plans.
High-quality code generation and debugging: It generates, optimizes, and debugs code across multiple data development languages — including SQL and Python — so you can move from intent to executable code quickly.
Built-in DataWorks Skills: Official Skills for Data Integration, data development, Data Governance, and Data Security are included out of the box. No extra configuration needed to call core platform capabilities. A built-in DataWorks ChatBI Skill is also available. After you enable it, you can perform natural language data queries and in-depth analysis in conversations.
Conversational interaction: Interact with the agent through an intuitive Chat mode. Describe your data development and governance needs in natural language, and the agent automatically interprets your intent and executes tasks.
Access the Data Agent
You can open the new Data Agent in three ways:
From the navigation pane
Log on to the DataWorks console and click Data Agent in the left navigation pane.
From the All Products menu
On the Data Studio page, click All Products on the left. Under the Agent category, click Data Agent.
From the original Data Agent panel
In the original Data Agent Chat panel, click Experience the new DataWorks Data Agent to navigate to the new Data Agent.
Quick start
Step 1: Select or create an instance
After you open the new Data Agent, select a Running instance from the instance list and click it to enter the session interface. If no instances are available, create one by performing the following steps:
On the instance list page, click Create Instance.
In the Create DataAgent Instance dialog box, configure the following parameters:
Parameter
Description
Instance Name
The custom name of the DataAgent instance, used to identify this instance.
Workspace
The DataWorks workspace to associate with this instance.
Resource Group
The serverless resource group used to run this instance.
CU
The amount of compute resources required to run the instance. The available CU quota is automatically displayed after you select a resource group.
Click Create Instance to complete the creation. After the instance starts, its status changes to Running. You can then click it to start using it.
Step 2: Create a session
After you enter an instance, click New Task in the left sidebar to create a new session. You can also click an existing session in the history list on the left to continue a previous conversation.
Step 3: Switch the model and approval mode (optional)
Data Agent uses the qwen3.7-max model and the Default approval mode by default. You can use the drop-down menus in the lower-left corner of the input box to switch models or approval modes.
Data Agent supports the following models:
Model | Description |
qwen3.8-max | The latest Qwen flagship model with well-rounded performance, ultra-long context support, ideal for complex data development and governance tasks. |
qwen3.7-max (default) | Qwen flagship model that excels at complex reasoning and code generation, with balanced cost-effectiveness. |
qwen3.7-plus | Cost-effective Qwen model with fast response, suitable for batch tasks and everyday conversations. |
deepseek-v4-pro | DeepSeek reasoning model with deep reasoning capabilities, excels at logical analysis and code tasks. |
glm-5.2 | Zhipu GLM model with strong Chinese semantic understanding, suitable for general text processing tasks. |
Data Agent provides the following approval modes:
Mode | Behavior | Applicable scenario |
Plan (plan) | Analysis only. Does not modify files or execute commands. | Scenarios where you need to understand the current state before making decisions. |
Default (default) | Edit operations require user confirmation before execution. | Everyday use that balances efficiency and safety. |
Auto-edit (auto-edit) | Automatically approves edit operations without step-by-step confirmation. | Efficient development in trusted environments. |
Auto (auto) | The agent autonomously assesses operation risks and requests user confirmation only for high-risk operations. | Automated development scenarios that balance efficiency and safety. |
Full access (yolo) | Automatically executes all tool calls and command operations without any confirmation. | Fully automated scenarios where you have full confidence in the execution results. |
Step 4: Configure and use Skills (optional)
Data Agent supports extending AI capabilities through Skills. You can use the built-in official DataWorks Skills or upload custom Skills to encapsulate specific business processes or analysis methods as reusable Skills.
Upload custom Skills
In the Expand area of the left sidebar on the session page, click Skills to open the Skills management panel.
Click Install Skill in the upper-right corner. In the dialog box, configure the following parameters:
Parameter
Description
Name
The unique name of the Skill. We recommend that you use English letters and hyphens for naming (for example,
excel-analysis) for easy reference in conversations.Description
The description of the Skill, which helps users quickly understand its purpose and applicable scenarios.
Skill body
The core content of the Skill, written in Markdown format. You can define applicable scenarios, workflow steps, and output formats in the body. Data Agent executes tasks based on the instructions in the body.
Zip package
Upload a resource file package in
.zipformat. The maximum size of a single file is 50 MB. You can place aSKILL.mdfile in the root directory of the zip package. If the file exists, it overrides the Skill body specified above. Other files serve as resource attachments that can be referenced inSKILL.mdby using relative paths. Limits:SKILL.md≤ 200 KB, entire zip package ≤ 50 MB.After you complete the configuration, click OK to finish uploading the Skill. Successfully uploaded Skills appear on the Custom tab of the Skills list.
Use Skills in a conversation
Below the input box, click the 
button and select Skills from the context menu. Then, select the specific Skill that you want to apply to the current conversation. Data Agent executes the corresponding tasks based on the workflow and instructions defined in the Skill.
For more information about Skills management, see Data Agent Settings.
Step 5: Enter and execute instructions
In the input box at the bottom, describe your data development or governance requirements in natural language and press Enter to send. Data Agent automatically plans and executes tasks based on built-in DataWorks Skills.
Examples of common instructions:
"Create a data synchronization task from MySQL to MaxCompute."
"The department field for employee ID EMP001 in the employee information table my_project.ods_emp_info_d is empty. Help me identify the cause and provide a fix."
"Show the lineage of the dws_user_active_di table."
"Generate a SQL script for user activity analysis."
Human-in-the-Loop confirmation
During task execution, Data Agent pauses at critical checkpoints to request your confirmation, ensuring operations are safe and controllable. Data Agent will pause and wait for your confirmation in the following scenarios:
Confirmation type | Trigger scenario | Description |
Plan approval | After Agent breaks down a complex task | Displays the execution plan for your confirmation before proceeding |
Destructive action confirmation | Before delete, overwrite, or truncate operations | Such as DROP TABLE, TRUNCATE, or overwriting partitions |
Release/deploy confirmation | Before submitting code or tasks to production | Such as publishing scheduled tasks, deploying nodes, or submitting SQL to a production project |
Clarification | When Agent lacks critical information | Such as ambiguous table names, uncertain field types, or unspecified time ranges |
High-cost action confirmation | Before full table scans, large-scale JOINs, etc. | Displays estimated cost for you to decide whether to proceed |
Cross-boundary action confirmation | Before cross-project, cross-tenant, or cross-environment operations | Prevents unintentional breach of environment or permission boundaries |
Permission change confirmation | Before authorization, sharing, or ACL change operations | Such as GRANT table permissions or sharing datasets |
Long-running task checkpoint | At key checkpoints during long-running tasks | Reports progress so you can choose to continue or abort |
ChatBI integration
Data Agent comes with built-in DataWorks Skills. Among these Skills, DataWorks ChatBI provides natural language data query and in-depth analysis capabilities. After you enable this Skill, you can perform data analysis directly in Data Agent conversations and receive analysis reports that contain visual charts. For the complete list of built-in Skills and their descriptions, see Manage extensions.
Enable the DataWorks ChatBI Skill
In the left sidebar of the session page, click Skills to open the Skills management panel.
Switch to the DataWorks tab to view the list of official built-in Skills.
Find the DataWorks ChatBI Skill card and turn on the switch to enable the Skill.
Official built-in Skills are enabled or disabled at the instance level. For information about how to upload and use custom Skills, see "Step 4: Configure and use Skills (optional)" in Quick start.
Automatic routing to the ChatBI Skill
After the DataWorks ChatBI Skill is enabled, Data Agent automatically routes questions that involve data analysis or data query scenarios to this Skill. ChatBI then completes the analysis based on the target datasets and returns the results, without requiring you to switch tools.
Explicitly specify the ChatBI Skill
To ensure that your questions are precisely handled by the ChatBI analysis capabilities, you can also explicitly specify the Skill name when you ask a question. Enter /dataworks-chatbi-skill in the input box, and then enter your data question. Data Agent directly invokes the DataWorks ChatBI Skill to process the question.
Billing
For billing details of the new Data Agent, see Data Agent fees.
References
To learn about the overall architecture and core value of Data Agent, see Data Agent overview.
To learn how to use the original sidebar agent, see Use the data development agent.
To learn how to use Data Agent, see Use Data Agent.
To learn about the configuration management in the Data Agent Settings Center, see Data Agent Settings Center.
To learn how to manage built-in Skills, expert suites, and MCP servers, see Manage extensions.
To learn about the features and usage of ChatBI, see Overview.