AI assistant service

更新时间:
复制 MD 格式

The AI Assistant service lets you perform alert analysis, task diagnosis, code modification, O&M, and data governance through natural language — directly in the DataWorks console or through DingTalk, Feishu, and WeCom group chats. Built on the OpenClaw framework, it provides one-click network setup, knowledge base management, Skill management, scheduled tasks, and audit logging to help you build a dedicated enterprise-grade AI assistant.

Features

Unlike the built-in agent in the DataWorks console, the AI Assistant service extends platform capabilities to the Web client and IM clients (DingTalk, Feishu, and WeCom), so you can complete tasks without opening the console.

Core values

  • Multi-client access: Access the AI assistant from the DataWorks console (Web client) or IM clients (DingTalk, Feishu, and WeCom group chats) anytime, anywhere.

  • Instant response: Receive alerts and run AI-powered diagnoses directly in your IM group, reducing response times from minutes to seconds.

  • Intelligent diagnosis: Automatically analyzes root causes of task failures and the impact scope of alerts, and suggests fixes.

  • Interactive O&M: Use natural language to issue O&M commands, such as rerunning tasks or modifying resource groups. All write operations require confirmation before execution.

  • Knowledge enhancement: Build a knowledge base automatically through retrieval-augmented generation (RAG) and memory management, improving the AI assistant's understanding of your business context.

  • Scheduled inspections: Built-in templates for scheduled and triggered tasks cover SLA monitoring, asset governance, cost analysis, and more, delivering inspection reports on a regular basis.

  • Data security: All data remains within the DataWorks platform. AI inference runs on the platform's internal network, and sensitive data is desensitized before being sent to the AI.

Available regions and policies

Important

Join the AI Assistant service DingTalk group for more technical support.

  • Available to customers using DataWorks Basic Edition or higher.

  • Available regions: China (Zhangjiakou), China (Beijing), China (Ulanqab), China (Hangzhou), China (Shanghai), China (Shenzhen), China (Chengdu), China (Hong Kong), Singapore, Malaysia (Kuala Lumpur), Indonesia (Jakarta), Japan (Tokyo), and US (Virginia).

Core capabilities

The AI Assistant service covers the following core data O&M scenarios:

Scenario

Capabilities

Sample commands

Fault diagnosis (core strength)

Alert diagnosis (root cause identification and upstream analysis), instance diagnosis (single or batch), and workspace health diagnosis

Diagnose the cause of failure for task instance 12345678

Task scheduling

Query or modify task configurations, query, rerun, terminate, or mark instances as successful, freeze or unfreeze tasks, adjust priorities, view dependency topologies, and migrate resource groups

Rerun instance 12345678

Workflow management

Query or modify workflows, backfill data or run smoke testing, manually trigger, start, stop, or rerun workflows, and refresh dependencies

Backfill data for workflow wf-001 from 2026-01-01 to 2026-01-31

Monitoring and O&M

Configure baselines and query SLA status, manage monitoring rules, query alert messages, and query duty rosters and schedules

View the SLA status of baseline base_test

Data Quality

Manage data quality monitoring rules and alert rules, and automatically configure data quality monitoring

Query the DQC rule configuration for table ods_user_info

Data Governance

Tag management (create, associate, and disassociate tags), query governance issues and rules, asset analysis (storage, compute, synchronization, and scheduling), asset health score, and governance inspection

View the data governance score for the project

Project management

Query the workspace list and details, query the member list and role list, and query the current user information

View the member list for workspace xxx

Resource management

Query resource group list and details, and monitor resource group utilization

Query the load status of resource group rg-001

Note

The following capabilities are currently read-only in the AI Assistant service: member permission management and role management in project management, and network configuration in resource management. The AI Assistant service does not support creating workspaces, creating tasks or nodes, or deleting workspaces or tasks.

Supported IM platforms

IM platform

Required configuration

Configuration reference

DingTalk

Client ID, Client Secret

Integrate the AI Assistant service with DingTalk

Feishu/Lark

App ID, App Secret

Integrate the AI assistant with Lark

WeCom

Bot ID, Secret

Integrate AI assistant service with WeCom

Prerequisites

  • Make sure that you have activated the DataWorks Data Agent feature. If you have not activated it, see Activate Data Agent.

  • Prepare a serverless resource group to run the AI Assistant service instance. An AI Assistant service instance requires at least 2 CUs.

  • To use the IM client, you must have created a bot application on DingTalk or Feishu and obtained the required configuration information, such as AppKey/AppSecret (DingTalk) or App ID/App Secret (Feishu).

Create an instance

Important

Only Alibaba Cloud accounts and RAM users with the AliyunDataWorksFullAccess permission can create and manage AI Assistant service instances.

  1. Log on to the DataWorks console, select the target region, and in the left-side navigation pane, click AI Assistant Service.

  2. On the AI Assistant service page, click Create.

  3. In the Create AI Assistant Service Instance dialog, configure the following parameters:

    Parameter

    Description

    Instance Name

    Required. The name of the AI Assistant service instance, used to identify and distinguish different instances. The name must start with a letter or Chinese character, and can contain letters, Chinese characters, digits, underscores (_), and hyphens (-). The name can be up to 128 characters in length.

    Resource Group

    Required. Select the serverless resource group to run the AI Assistant service instance. An AI Assistant service instance requires at least 2 CUs.

    Note

    If the available CUs of the selected resource group are insufficient for the AI Assistant service instance, the resource group is not selectable.

    Execution Identity

    Required. Specify the identity that the AI Assistant service uses to access DataWorks. If the AI assistant is used only as your personal assistant, we recommend that you set this to Current Logon User. If the AI assistant is shared with multiple users, we recommend that you set this to Create New RAM Role, which allows you to specify the workspaces and workspace roles that the role can access.

    Instance Description

    Optional. The description of the instance. The description can be up to 256 characters in length.

  4. Click OK.

    Creating an instance takes approximately 1 to 3 minutes. After the instance is created, you can start using the AI assistant.

Manage instances

Basic operations

On the AI Assistant service page, you can view and manage the AI Assistant service instances that you have created.

  • Start/Stop an instance: An instance is started by default after it is created. After you modify the instance configuration, you can stop and then start the instance for the changes to take effect.

  • Upgrade an instance: When the image version of an instance is earlier than the latest version, you can click the Upgradable tag next to the image column in the instance list to upgrade the instance to the latest version and access new features.

  • Execution Role Management: Modify the access permissions of the specified access identity. You can edit the list of accessible workspaces and workspace roles.

  • Edit instance: Modify the instance name and description.

  • Delete instance: Click Delete in the Actions column of the target instance to delete the instance.

  • Retry: For instances that failed to be created or started, click Retry to re-execute the operation.

Important

Only the instance owner, Alibaba Cloud accounts, and users with FullAccess permissions can edit or delete instances.

Channel configuration

The AI Assistant service supports IM channel configuration, which allows you to chat with the AI assistant directly on DingTalk, Feishu/Lark, or WeCom. This lets you perform O&M tasks without opening the DataWorks console. If you do not configure an IM channel, you can still use the AI assistant through the Web client.

Click the configuration button for a channel and enter the required IM information in the dialog that appears:

  • DingTalk: Enter the Client ID and Client Secret.

  • Feishu/Lark: Enter the App ID and App Secret.

  • WeCom: Enter the Bot ID and Secret.

After you complete the configuration, click Test Network Connectivity to verify the network connectivity. If the network is not connected, click One-Click Network Setup to configure the network. The following two network solutions are supported:

  • PrivateZone + private connectivity (recommended): If the resource group is deployed in a VPC that does not have Internet access, use this method to allow resources in the VPC to access IM platform endpoints through a fully private network, ensuring that traffic never leaves the private network.

    This solution requires a vSwitch in the specified availability zone. If no vSwitch exists, create one as prompted on the page.

  • Internet access: Connect to the Internet through an Internet NAT Gateway and an EIP. This option is simpler to configure but less secure.

Note

Different network solutions incur different fees. For details, see Billing.

Skill management

A Skill is an extensible capability module that the AI assistant can invoke. Skill management allows you to extend the AI assistant's capabilities for data O&M and governance scenarios. The AI Assistant service supports both built-in system Skills and custom Skills. System Skills are provided and maintained by the platform, covering scenarios such as data development assistance and product Q&A. Custom Skills allow you to upload your own capability modules to meet specific business needs.

Click the target instance name to go to the instance details page, and click the Skill Management tab.

The Skill Management tab is visible only when the instance image version is v3.0.0 or later. If you do not see the tab, upgrade the instance image version first.

System Skills

Built-in system Skills are provided and maintained by the DataWorks platform. The following system Skills are currently available:

  • Data development Skill: Supports data development assistance capabilities such as code analysis and SQL generation.

  • Product Q&A Skill: Supports Q&A about DataWorks product usage.

System Skills cannot be deleted. You can enable or disable them by using the toggle switch.

Custom Skills

  1. Import a Skill

    1. On the Skill Management tab, click Import Skill.

    2. In the Import Skill dialog, configure the following parameters:

      Parameter

      Type

      Required

      Description

      Name

      Text box

      Yes

      The name of the Skill, used for identification. Maximum 64 characters.

      Description

      Text area

      No

      The description of the Skill, which helps you understand its purpose. Maximum 256 characters.

      Skill File

      File upload

      Yes

      The compressed Skill file. Only .zip format is supported. Drag-and-drop upload is supported.

    3. After you complete the configuration, click Confirm to import the Skill.

  2. Enable or disable a Skill

    In the Skill list, use the toggle switch on each Skill card to enable or disable the Skill in real time. After a Skill is disabled, the AI assistant no longer invokes its capabilities.

  3. Upgrade a Skill

    When a new version is available for a system Skill, the Upgradable tag is displayed on the Skill card.

    1. Click the Upgrade button.

    2. The system automatically disables and re-enables the Skill. The upgraded Skill takes effect immediately.

  4. Delete a custom Skill

    In the Skill list, find the custom Skill that you want to delete, and click the Delete button to confirm.

System Skills cannot be deleted. Only custom Skills can be deleted.

Get more Skills

On the Skill Management tab, click Alibaba Cloud Skill Portal to go to the Alibaba Cloud Skill Portal page, where you can browse and obtain more available Skill modules.

Memory management

The long-term memory feature helps you persist critical business context for an AI assistant instance. Through memory management, you can save important knowledge such as project asset information, data standards, and business processes as files in the AI assistant instance. This allows the AI assistant to retain an understanding of the context across sessions, providing more accurate and personalized O&M recommendations.

On the AI Assistant service list page, click the target instance name to go to the instance details page, and click the Memory Management tab.

  • Browse directories: Browse memory files in a tree directory structure. You can navigate into subdirectories and return to the parent directory.

  • View files: Click a file to view its content. For large memory files, the content is truncated and a truncation message is displayed.

  • Import memory: Upload files to a specified path. You can export memory from a local OpenClaw instance or another AI Assistant service instance for reuse.

  • Export: Supports single file download or batch export as a package.

  • Delete: Supports deleting files individually or in batches.

Knowledge base management

The knowledge base feature is based on RAG (retrieval-augmented generation) technology, enabling the AI assistant to automatically understand your DataWorks project context. The knowledge base automatically extracts table metadata, task code, lineage, and other information from your DataWorks workspaces to build a structured knowledge system, providing the AI assistant with more accurate data governance and O&M recommendations.

On the AI Assistant service list page, click the target instance name to go to the instance details page, and click the Knowledge Base tab to go to the knowledge base management page.

Generate a knowledge base

When you use the knowledge base feature for the first time, you must generate the knowledge base first.

  1. On the empty Knowledge Base tab, click Start Generating Knowledge Base.

  2. In the mode selection dialog, select a generation mode:

Mode

Description

Estimated time

Quick mode

Parses metadata schemas, task code, and lineage, and performs lightweight LLM enhancement (table display_name, business description, and data domain).

Approximately 10 to 20 minutes

Deep mode (recommended)

Includes all capabilities of quick mode, plus data sampling and profiling statistics (distinct values, null values, and value distribution to assist PII identification), and deep LLM semantic enhancement (table and column descriptions, derived metric identification, and synonym and glossary aggregation).

Hours (varies based on asset volume)

  1. After you confirm your selection, the system starts generating the knowledge base. The page enters the generating state and displays real-time progress.

  2. After the generation is complete, the page automatically switches to the knowledge base browsing view.

Deep mode considerations: Deep mode triggers SQL sampling queries (MaxCompute, Spark, and others), which may consume compute resources. Tables that you do not have permission to access are automatically skipped. We recommend that you use deep mode during off-peak hours.

Manage the knowledge base

After the knowledge base is generated, the following operations are supported:

  • Browse files: Browse the knowledge base files in a tree directory structure.

  • Edit files: Modify the content of files in the knowledge base. Only custom uploaded files can be edited.

  • Import files: Upload custom knowledge files to supplement the AI assistant's business knowledge. Supported file formats include .md, .txt, .pdf, .docx, .xlsx, .html, and .csv. The file size cannot exceed 10 MB.

  • Export files: Download a single file directly or export multiple files as a package.

  • Regenerate: After metadata changes in your DataWorks workspaces, you can regenerate the knowledge base to obtain the latest knowledge context.

  • Clear knowledge base: Clear all generated knowledge base files.

Instance monitoring

On the AI Assistant service list page, click the target instance name to go to the instance details page, and click the Monitoring tab to view the resource usage of the instance:

  • CPU usage: Used CPU / total specification.

  • Memory usage: Used memory / total specification.

  • Disk usage: Used disk space / total specification.

You can select a time range to view historical trends. The data for the last hour is displayed by default.

Instance logs

On the AI Assistant service list page, click the target instance name to go to the instance details page, and click the Logs tab to query the runtime logs of the instance:

  • Filter logs by subsystem.

  • Search logs by keyword.

  • Filter logs by time range.

Use the AI Assistant service

Use conversations

Use the AI Assistant service through the Web client

After an AI Assistant service instance is created, click Chat in the instance list to start a conversation with the AI assistant directly in the DataWorks console.

Session management

The Web client supports two types of session management: regular sessions and scheduled tasks. The session sidebar on the left provides the following features:

Regular sessions:

  • Create a session: Click + New Task to create a new conversation session.

  • Switch sessions: Click a session name in the session list to switch to the corresponding session.

Scheduled tasks:

  • Scheduled and triggered tasks: DataWorks provides multiple built-in templates for scheduled tasks. You can select an existing template to quickly start a new task.

Conversation interaction

In the conversation interface, you can interact with the AI assistant using natural language. The AI assistant automatically calls DataWorks APIs to perform O&M operations. All write operations require your confirmation before execution.

Use the AI Assistant service through IM clients

  • After an AI Assistant service instance is created, you can interact with the AI Assistant service by using natural language in a group where the bot is configured or in a private chat with the bot.

  • Proactive inquiry: Mention the bot in a group chat or start a private chat with the bot, and enter your question or command. The AI Assistant service automatically understands your intent and performs the corresponding operation.

    Scenario

    Sample commands

    Task failure diagnosis

    Diagnose the cause of failure for task instance 12345678

    Alert analysis

    Analyze currently unresolved alerts

    Baseline SLA query

    View the SLA status of baseline base_test

    Workflow backfill data

    Backfill data for workflow wf-001 from 2026-01-01 to 2026-01-31

    Rerun failed tasks

    Rerun the failed scheduled instances in workspace abc today

  • Automatic alert diagnosis: When a DataWorks monitoring rule triggers an alert, the AI Assistant service automatically receives the alert, runs an AI-powered diagnosis, and pushes the diagnosis results to the IM group.

    Tell the AI Assistant service: Enable automatic alert monitoring and push to the current session.

    The AI Assistant service automatically configures alert listeners and sends DataWorks-triggered alert information to the current session as message cards. You can then use natural language commands in the chat (such as "view failure logs", "rerun the current node", or "roll back to the previous version") to directly invoke underlying APIs for O&M actions without logging on to the console, reducing the time to resolve issues.

  • Enable governance inspection: You can enable automatic inspections. The AI Assistant service automatically performs asset governance inspections and pushes inspection results to the IM group on a regular basis.

    Tell the AI Assistant service: Enable automatic inspection and push to the current session.

    Governance inspections push inspection results to the IM client at 09:30 every day by default. You can modify the push time and frequency through conversation. You can also submit more specific governance questions to the AI Assistant service, such as "Enable governance inspection and push every 3 hours the number of tables whose owner is the primary account" to meet personalized requirements.

Important

Write operation confirmation: Before all write operations (such as rerun, freeze, delete, and modify) are executed, the AI Assistant service clearly displays the operation type, target objects, and affected quantity. The operation is executed only after you explicitly confirm it.

Scheduled tasks

The AI Assistant service includes multiple built-in task templates to help you quickly configure scheduled inspection tasks. In the session sidebar of the Web client conversation interface, click Scheduled Tasks to open the template panel, and select a template to create a scheduled task session.

Note

To configure scheduled tasks on the IM client, you can directly use natural language to instruct the AI assistant.

Scheduled inspection task templates

The following scheduled task templates run at 09:00 every day by default, automatically generate inspection reports, and push them to the session:

Task name

Description

SLA baseline completion report

Diagnoses and summarizes the overall baseline completion status for the project today, and generates a report

Workspace completion report

Diagnoses and analyzes the task completion status for the workspace today, and generates a report

Inspect unhealthy assets

Summarizes governance issues under your account, analyzes priorities, and generates a report

Inspect data quality

Summarizes high-priority tables without data quality monitoring configured, and provides analysis and recommendations

Daily inspection: Root cause analysis for failed tasks

Summarizes failed instances from the previous day and performs in-depth diagnosis on the top 5 frequently failing tasks

Daily inspection: Resource group utilization

Identifies resource group utilization, load, and queuing bottlenecks, and provides recommendations for scaling or load balancing

Daily inspection: Slow task ranking

Compares historical run times and analyzes the top 10 tasks with abnormally increased run and wait times

Daily inspection: Monitoring rule coverage gaps

Identifies monitoring rule coverage gaps for critical baselines and high-priority tasks, and provides recommendations for remediation

Daily inspection: Backfill data execution status

Summarizes the status of backfill data tasks from the past 7 days and analyzes failure causes

Daily inspection: Tables without lifecycle configuration

Summarizes tables without lifecycle configuration by storage volume and provides configuration recommendations for the top 20

Daily inspection: Ownerless tables and deprecated assets

Scans tables that have not been accessed for a long time and tables without an owner, and generates a candidate list for undeployment

Daily inspection: Code change governance compliance

Checks governance rule results for node changes in the last 24 hours and summarizes by owner

Daily inspection: Storage and compute consumption ranking

Lists the top 20 storage and compute consumers, and analyzes growth trends with optimization recommendations

Daily inspection: Asset cataloging coverage

Summarizes tables not included in asset catalogs and their percentage, and provides cataloging recommendations

Custom scheduled tasks

DataWorks also supports custom scheduled tasks. On the Web client, click Scheduled Tasks, and click Custom Create in the upper-right corner. Follow the AI assistant's prompts to specify the run interval and task content.

Billing

The AI Assistant service incurs the following fees:

Billable item

Billing method

Description

Data Agent seat

Billed per seat

A Data Agent seat is a prerequisite for using the AI Assistant service. For more information, see Data Agent billing.

Resource group fees

Pay-as-you-go (by duration)

The CU fees for the resource group consumed by the AI Assistant service instance. If the resource group requires network connectivity, additional network fees are incurred:

PrivateZone + private connectivity: PrivateZone (PrivateZone billing) and endpoints (Endpoint billing).

Internet access: NAT (NAT Gateway billing).

AI inference

Billed by token usage

The fees for AI model invocations for intelligent diagnosis and analysis by the AI Assistant service. For more information, see AI inference billing.

Differences between the AI Assistant service and DataWorks Agent

Comparison item

DataWorks Agent

AI Assistant service

Access point

Within the DataWorks console (Data Studio page)

Web client (AI Assistant service page in the DataWorks console) and IM clients (DingTalk, Feishu, and WeCom group chats)

Primary scenarios

All scenarios, including data development, Data Integration, and Data Governance

O&M scenarios (alert diagnosis, task management, and baseline monitoring) and Data Governance scenarios

Interaction method

Data Agent dialog in the browser

Dedicated Web client conversation interface or @AI assistant in IM group chats

Alert integration

Not supported

Supports receiving alerts and automatic AI diagnosis

Knowledge base

Not supported

Supports building workspace-level knowledge base (RAG)

Scheduled tasks

Not supported

Supports scheduled inspections and triggered tasks

Custom Skills

Not supported

Supports importing and managing custom Skills

AI Assistant service DingTalk group

Click the application link to join the "Alibaba Cloud Big Data AI Platform" organization first, and then scan the QR code below to join the AI Assistant service DingTalk group for dedicated product technical support.

image