MaxAgent
MaxAgent is the AI assistant for the MaxCompute platform. It provides end-to-end self-service and intelligent operations capabilities to lower the learning curve and improve O&M efficiency. With natural language, MaxAgent helps you handle job diagnostics, data queries, cost analysis, resource optimization, permission audits, and metadata exploration—without memorizing complex commands or searching documentation.
Disclaimer
Before using MaxAgent, please read the following notes carefully:
-
AI-generated content is for reference only: Content generated by this service is from an AI model. Alibaba Cloud cannot fully guarantee the security, reliability, compliance, completeness, or accuracy of the content. The generated content does not represent the position or views of Alibaba Cloud. Exercise your own judgment and verify all information before performing critical operations. Alibaba Cloud is not liable for any losses resulting from decisions or actions based on the generated content.
-
Use caution with configuration changes: For any suggestions involving configuration changes or high-risk operations (such as deletions or permission changes), always validate them in a test environment before executing them in a production environment.
-
Permission boundaries: MaxAgent runs operations as the currently logged-in user and never exceeds that account's permission scope. To use features like diagnostics, optimization, and data analysis, which may involve accessing project information, job logs, table metadata, and table data, you must have the corresponding permissions.
-
Data privacy: The diagnostic and analysis features only read relevant job logs and metadata. They do not modify or transmit your data.
-
Lawful and compliant use: You are responsible for all your actions when using this service. Ensure that any content you publish, upload, or provide through the service is lawful, compliant, and does not infringe upon the rights and interests of others.
-
Prompt quality affects results: The more specific your questions are (including key parameters like an instance ID, time range, or project name), the more accurate the diagnostic and analysis results will be.
Supported regions
China (Hangzhou), China (Shanghai), China (Beijing), China (Zhangjiakou), China (Ulanqab), China (Shenzhen), China (Chengdu), Hong Kong, Indonesia (Jakarta), Singapore, Malaysia (Kuala Lumpur), Germany (Frankfurt), US(Silicon Valley), and US (Virginia).
Target users
|
User role |
Use cases |
|
Data platform administrator |
Resource management, cost optimization, permission management, batch inspections |
|
O&M engineer |
Troubleshooting, automated O&M, LogView analysis |
|
Enterprise technical lead |
Cost visibility, higher resource utilization, higher team productivity |
|
Data development engineer |
Job diagnostics, performance optimization |
Billing
-
Credits cover the consumption of tokens and compute resources for MaxAgent instances. These charges do not include the costs of tasks submitted to the engine, such as the SQL execution costs generated by the intelligent data retrieval module.
-
You must enable the Standard Edition to use MaxAgent. You can then purchase Credits packages as needed.
Purchase options
MaxAgent is available in two forms: a Standard Edition and Credits packages.
-
MaxAgent Standard Edition: Enable for free and receive 300 free Credits each month.
-
Credits packages: Monthly packages are available in sizes of 100, 500, 1,000, 5,000, and 10,000 Credits.
For detailed pricing, see Credits billing.
Permissions
When you use MaxAgent features such as diagnostics, optimization, and data analysis, MaxAgent makes data requests using your identity. These requests may involve accessing project information, job logs, and table metadata. Ensure your account has the necessary control plane or data plane permissions.
-
Alibaba Cloud account: You can use MaxAgent directly.
-
RAM user:
-
When using AI Ops features such as resource viewing and job diagnostics, you must have the required permissions to access resource and job data.
-
When using AI Query or DataAgent to view metadata or query data, you must have the required permissions to access MaxCompute table data.
For more information, see Permissions overview.
-
Quick start
-
Log in to the MaxCompute console and select a region in the upper-left corner.
-
In the upper-right corner, click MaxAgent to open the dedicated MaxAgent page, which includes your conversation history and the main chat interface.
Core capabilities
MaxAgent currently offers OpsAgent capabilities that cover the following core scenarios. It uses built-in, specialized Skills to automatically match your intent and run the corresponding actions.
Knowledge Q&A
Get comprehensive answers about MaxCompute features, use cases, and best practices. You can query product documentation using natural language.
-
Example prompts
-
"How do I create an external table in MaxCompute?"
-
"What is Tunnel? What are its transfer limits?"
-
"Which data formats does MaxCompute support?"
-
"How do I configure network connections for MaxCompute?"
-
-
Usage tips
For more accurate answers, describe your specific scenario, such as your operating environment or the exact problem you are facing.
Job diagnostics and log analysis
Single job diagnostics
Analyze a single SQL job to determine the cause of failure or identify performance bottlenecks.
-
Supported diagnostic types
Type
Description
Failed job diagnostics
Categorizes errors for failed jobs and suggests fixes.
Slow job diagnostics
Analyzes execution bottlenecks for slow SQL jobs and provides optimization advice.
Data skew analysis
Identifies data skew nodes in a job, analyzes the root cause, and recommends solutions.
UDF performance analysis
Analyzes the execution performance of a user-defined function (UDF) to identify bottlenecks.
-
Example prompts
-
"Help me diagnose this job: [instance_id]"
-
"Why is this SQL job running so slow? LogView URL: [url]"
-
"Is there any data skew in this job [instance_id]?"
-
-
Usage tips
Providing an instance ID or LogView URL yields more accurate diagnostic results.
Job inspection
View overall job health at the platform or project level to track failure rate trends and top resource consumers.
Example prompts
-
"What is the job failure rate for project [project_name] in the last 24 hours?"
-
"Which jobs are consuming the most resources?"
-
"Show me today's job execution status."
Resource and cost optimization
Bill and cost analysis
View cost summaries, trends, and the distribution of billable items based on metering data.
Example prompts
-
"What are my MaxCompute costs for this month?"
-
"What is the computing cost trend for the last 7 days?"
-
"Which billable item accounts for the highest percentage of my costs?"
Quota analysis and planning
Analyze Quota resource utilization and scheduling, assess whether capacity is sufficient, and receive optimization recommendations.
Example prompts
-
"Are the resources for Quota [quota_name] sufficient? Are there any risks?"
-
"Help me analyze the utilization of [quota_name]."
-
"Is the scheduling configuration for Quota [quota_name] optimal? Any suggestions for improvement?"
Storage analysis and optimization
Analyze storage distribution at the table and partition level, identify top storage consumers, and receive recommendations for storage optimization.
Example prompts
-
"Which tables are using the most storage?"
-
"What is the storage usage for project [project_name]? Is there room for optimization?"
-
"Is there any cold data that can be moved to tiered storage?"
Optimization recommendations
Materialized view optimization
Analyze the usage and benefits of materialized views and AutoMV, and receive recommendations for potential materialized views.
Example prompts
-
"Are there any recommended MVs I should review?"
-
"What are the current hit rate and benefits of AutoMV?"
-
"Which queries are good candidates for creating a materialized view?"
Table clustering optimization
Get recommendations on which tables to cluster and evaluate the benefits of existing clustered tables.
Example prompts
-
"Which tables are suitable for cluster optimization?"
-
"Help me analyze the clustering benefits for [table_name]."
-
"What is the effectiveness of my currently clustered tables?"
Permission and security audit
Audit user, role, and ACL permission configurations to identify excessive permissions, ACL anomalies, or cross-project trust risks.
Example prompts
-
"Help me audit the permission configuration for project [project_name]."
-
"List all users who have the admin or super_administrator role for any project, along with the corresponding projects."
-
"Are there any suspicious cross-project trust configurations?"
Metadata exploration
Browse metadata for projects, tables, columns, partitions, and functions to understand your data structures.
Example prompts
-
"What tables are in project [project_name]?"
-
"What is the column schema of table [table_name]?"
-
"Which tables are related to [business name or description]?"
-
"Show me the partition list for [table_name]."
Prompting tips
Describe what you want to know or what problem you need to solve. Avoid being too vague.
-
Not recommended: "My job has an issue." Recommended: "The job with instance ID [instance_id] in project xxxx failed. Please analyze the cause."
-
Recommendations:
-
Provide key context: Including details like an instance ID, project name, Quota name, or time range helps MaxAgent provide a precise response.
-
Focus on one question at a time: Avoid asking multiple unrelated questions in a single message. A step-by-step approach works best.
-
Prompt templates:
|
Scenario |
Prompt template |
|
Job diagnostics |
"Analyze the cause of failure for the job with instance_id=[xxx]" |
|
Cost query |
"Analyze the computing cost trend for [project_name] over the last 30 days" |
|
Quota analysis |
"Analyze the utilization of [quota_name] and provide optimization suggestions" |
|
Permission audit |
"Audit project [project_name] for users with excessive permissions" |
|
Storage analysis |
"List the top 10 tables by storage usage in [project_name]" |
|
Product Q&A |
"How do I configure tiered storage in MaxCompute? What are the key considerations?" |
Conversation history
MaxAgent saves your conversation history, allowing you to review past diagnostic results and analysis reports. On the dedicated MaxAgent page, you can click any past session in the left-side panel to view it.
Scheduled tasks
Scheduled tasks are suitable for recurring automated O&M and data analysis scenarios. After a scheduled task is configured, MaxAgent automatically runs inspections, generates reports, and pushes the reports to a target group or individual through a bot based on the specified schedule. This reduces the cost of manual inspections and shortens response times.
Scheduled tasks are suitable for the following scenarios:
-
Routine inspections: Automatically inspect project and job health to proactively identify issues.
-
Recurring reports: Automatically generate daily Quota utilization reports and cost analysis reports to reduce manual reporting work.
-
Information delivery: Automatically push key reports through a bot to shorten response times.
Built-in templates
MaxAgent provides the following built-in task templates. A template can be selected to quickly create a scheduled task.
|
Task template |
Recommended frequency |
Intended users |
Scenario |
|
Daily inspection for abnormal jobs |
Daily |
Data platform administrators and O&M engineers |
Automatically inspects job health at the MaxCompute project or Quota level every day, identifies failed jobs, abnormally slow jobs, and jobs with abnormal resource usage, and generates a structured inspection report. This helps promptly identify and handle production anomalies and reduce their impact. |
|
Daily compute Quota resource utilization report |
Daily |
Data platform administrators and enterprise technical owners |
Automatically analyzes the resource utilization, scheduling, and queueing status of each Quota every day. It performs a comprehensive assessment based on dimensions such as the peak and off-peak characteristics of CU (compute unit) utilization, queueing duration, and scheduling route matching, and provides capacity planning recommendations, including time-specific configuration recommendations for elastic reserved CUs. This helps refine resource management and optimize costs. |
|
Weekly storage cost optimization |
Weekly or monthly |
Data platform administrators and enterprise technical owners |
Automatically analyzes project-level storage usage distribution, cost trends, and optimization opportunities every week. It examines table- or partition-level storage distribution to identify waste such as cold data, duplicate data, and expired data, recommends lifecycle management and tiered storage strategies, and quantifies expected optimization benefits. This helps enterprises continuously reduce storage costs. |
|
Monthly compute cost analysis report |
Weekly or monthly |
Enterprise technical owners and data platform administrators |
Automatically summarizes compute resource consumption and fees every month. It provides a comprehensive analysis based on dimensions such as fee overview, billing item distribution, top consumer attribution, and fee trends, identifies cost waste hotspots, and provides refined cost reduction recommendations to support budget management and cost governance. |
|
Weekly permission security audit |
Weekly |
Data platform administrators and security O&M personnel |
Automatically audits tenant- and project-level MaxCompute permission configurations every week, checks for security risks such as excessive permissions, ACL anomalies, cross-project trust risks, and long-unused privileged accounts, and generates a risk list and remediation recommendations. This helps meet data security and compliance requirements. |
|
Scheduled NL2SQL (natural language to SQL) data queries |
- |
Data development engineers and business analysts |
Describe a recurring data query in natural language. MaxAgent automatically converts the natural language into SQL, runs the SQL in a specified project, and generates business data reports on schedule. This template is suitable for scenarios such as routine business metric monitoring, automatic generation of weekly data reports, and periodic data exports. |
Create a scheduled task
-
Log on to the MaxCompute console and select a region in the upper-left corner.
-
In the left-side navigation pane, choose . On the left side of the dedicated MaxAgent page, click Scheduled Task.
-
On the Scheduled Task page, click Create Scheduled Task on the right and configure the following parameters in the dialog box:
-
Task Name: Specify a custom task name, such as Daily abnormal job inspection for project XXXX.
-
Scheduled Time: Select an execution frequency (hourly, daily, every Monday, or the first day of every month) and an execution time.
-
Instruction: Configure the content to run. Write a prompt that describes the inspection, analysis, or query task.
-
Notification Channel: Select a configured bot. For information about how to configure a bot, see the Bot configuration section. After the task is complete, the result report is automatically pushed to the target group or individual.
-
-
Click OK to create the scheduled task.
Alternatively, select an applicable template on the Template Gallery tab, edit the corresponding content on the edit page, and click OK to quickly create a scheduled task.
View execution records and results
After a scheduled task is created, it automatically runs at the scheduled time. The execution records and detailed results of each run can be viewed.
Manage scheduled tasks
The scheduled task list displays information such as task name, scheduled time, task owner, enablement status, and task description, and supports the following operations:
|
Operation |
Description |
|
Execution records |
View historical execution records and results. |
|
Edit |
Modify configurations such as the task name, scheduled time, task content, and notification channel. |
|
Run manually |
Immediately trigger a task run without waiting for the scheduled time. |
|
Disable/Enable |
Pause or resume scheduled task execution. |
|
Delete |
Delete the scheduled task. |
Execution mechanism and security considerations
-
Execution identity: A scheduled task runs as the task owner and can access only the projects and data for which the owner has permissions.
-
Result delivery: After a task is complete, the result is formatted based on the report template and can be pushed to a target group or individual through a configured bot.
-
Credits consumption: The Credits consumption of scheduled tasks follows the standard metering rules, and consumption details are recorded after each run. On the page of the console, the cumulative consumption of scheduled tasks can be viewed. If the Credits balance is insufficient, scheduled tasks fail to run.
Credits Quota Management
MaxAgent Credits are shared by all sub-accounts, RAM users, and RAM roles under the tenant. Administrators (the primary account or RAM users with MaxAgent management permissions) can set Credits consumption limits for different identities through quota management to prevent excessive consumption by a single account and achieve cost control at the department or member level.
Credits quotas are calculated on a calendar month basis, with used quotas reset at 00:00 on the 1st of each month. Quota management includes three configuration items: total quota, individual default quota, and individual quota details.
|
Quota item |
Scope |
Description |
|
Total quota |
All identities under the tenant. |
Enable Stop When Credits Are Exhausted. When the Credits package quota is exhausted, new MaxAgent session requests are no longer accepted. If disabled, the limit includes the total postpaid limit plus the package quota. |
|
Individual default quota |
All members without individually configured quotas. |
Sets the default monthly Credits limit for members without individually configured quotas. Enter 0 to disable usage, or leave blank for no limit. |
|
Individual quota details |
Individually configured RAM users / RAM roles |
Individually configures the monthly Credits consumption limit for specified identities, with higher priority than the individual default quota. Enter 0 to disable usage, or leave blank for no limit. |
-
Individual quotas take precedence over the individual default quota. If an identity has been individually configured in the individual quota details, the individually configured limit applies; otherwise, the individual default quota applies. If the individual quota is blank, the identity is not subject to individual quota restrictions and is only subject to the total quota.
-
Configuration entry: Log on to the MaxCompute console. In the left-side navigation pane, choose .
Resource viewing and usage
Log on to the MaxCompute console. In the left-side navigation pane, choose . Here, you can view details of all your active MaxAgent order instances and a breakdown of your Credits usage.
MaxAgent order instance information
MaxAgent order instance details include:
-
Order instance ID and specifications
-
Total, used, and remaining Credits
-
Instance start and expiration times
Credits usage details
Credits usage details support searching by time, instance ID, and session requester, and also support aggregated display by instance ID or session requester, making it easy to view and analyze Credits consumption.
-
In non-aggregated mode, the details list includes usage time, order instance ID, Agent type, Requester (including User Name and UID), Request ID (corresponding to one request), and Credits usage.
-
In aggregated mode, the list includes aggregation dimension, record count, and total Credits, which is suitable for analyzing Credits consumption by each requester or order instance ID within a specified time period.
Bot configuration
MaxAgent integrates with DingTalk bots (with support for Lark and WeCom planned for the future) to enable "chat as a service" interaction, allowing you to use MaxAgent without logging in to the console and manage tasks from your mobile device.
If an IP whitelist is configured for the bot, the egress IP of MaxAgent must also be configured; otherwise, the bot will not function properly. The egress IP of MaxAgent is as follows:
|
Region |
IP |
|
cn-beijing |
39.96.91.188 |
|
cn-shanghai |
47.102.46.190 |
|
cn-hangzhou |
120.26.251.232 |
|
cn-chengdu |
8.137.169.192 |
|
cn-shenzhen |
39.108.66.251 |
|
cn-zhangjiakou |
47.92.119.78 |
|
cn-wulanchabu |
8.130.23.158 |
FAQ
-
Q: Will MaxAgent automatically modify my configurations or data?
A: No. MaxAgent currently provides diagnostic, analysis, and optimization suggestions. It does not automatically perform any configuration changes or data operations. All changes must be explicitly confirmed and manually executed by you.
-
Q: What happens if I run out of Credits?
A: When your Credits are depleted, you can no longer submit new messages. The system will then prompt you to purchase a Credits package to replenish your balance.
-
Q: How can I get more accurate answers from MaxAgent?
A: For best results, include specific context in your prompts, such as an instance ID, project name, Quota name, or time range. Ask one specific question at a time rather than multiple unrelated questions. If you receive an inaccurate answer, try rephrasing the question with more information.
-
Q: Can a RAM user use MaxAgent?
A: Yes. However, an Alibaba Cloud account must grant the RAM user the necessary RAM permissions. The user can then only access resources and data within their permitted scope.
-
Q: Which MaxCompute job types does MaxAgent support for diagnostics?
A: Currently, MaxAgent supports diagnostics and LogView interpretation for SQL jobs. Support for Spark and MapReduce (MR) jobs is planned for future releases.