引言：刚开始使用MaxCompute时常会遇到这样的场景：想了解一个 MaxCompute 功能，常常要在几篇文档之间查找。遇到一条失败或运行缓慢的 SQL，还要先找到作业信息，再去查看日志。
MaxAgent 是 MaxCompute 平台的 AI 智能助手。你可以用自然语言询问产品知识，或结合具体作业和资源信息开展诊断分析。
从这一篇开始，我们会围绕日常任务介绍 MaxAgent 的使用方法。
第一课先完成一次简单体验：进入产品、确认开通和权限、提出一个问题，再核验回答中的一条关键结论。
本文适合首次使用 MaxAgent 的数据开发、运维、平台管理员和技术负责人。示例只做产品知识问答，不创建表、不执行 SQL，也不修改配置，无需准备业务数据。
为什么要使用 MaxAgent
MaxCompute 的日常工作横跨产品文档、作业日志、资源观测、成本与权限等多个入口。很多任务本身并不复杂，时间却花在寻找入口、补充上下文和来回核对上。MaxAgent 将这些场景放进自然语言对话，让使用者先描述目标，再逐步补齐项目、作业或时间范围。
降低产品学习门槛
不熟悉某个概念或功能时，可以直接用日常语言提问，不必先掌握全部命令和页面位置。MaxAgent 可以回答 MaxCompute 产品功能、使用场景和最佳实践等问题，帮助新用户更快找到继续学习和操作的方向。
让业务问题直接进入数据分析
DataAgent 是 MaxAgent 的数据分析角色，在控制台中对应“查数”。它可以在当前账号有权访问的数据范围内，用自然语言查找项目、Schema、表和分区，执行有边界的查询，分析指标、诊断异常、检查数据质量，并将结果整理为数据表、图表、仪表板或分析报告。使用者不必先知道完整表名或熟练编写 SQL，但仍需明确时间范围、指标口径和分析目标，并复核查询证据与结论。
缩短诊断与分析链路
面对 SQL 失败、运行缓慢或数据倾斜等问题，可以结合具体作业信息发起诊断。平台管理员也可以围绕成本、Quota、存储、权限和元数据开展分析。
从分析结果走向管理闭环
MaxAgent 不仅能分析问题，也已支持项目管理、Quota 管理、权限管理等操作。具备相应权限的用户可以用自然语言发起部分平台管理任务，减少在多个页面和命令之间切换的成本。
让周期性工作持续运行
对于异常作业巡检等重复任务，支持配置定时任务，并通过钉钉机器人接收结果。这样，部分工作可以从临时发问转为周期运行，再根据报告继续下钻问题。
权限沿用、敏感操作受控、过程可回看
MaxAgent 以请求者身份访问和管理资源，不会绕过现有权限。
- DataAgent 只能读取当前账号有权访问的数据，用于分析查询，对表或表数据进行操作除了验证当前账户权限还需二次确认。
- 对于项目、Quota、权限等管理操作，只有具备相应权限的用户才能发起，并需按实际页面完成确认或审批；执行前应检查目标对象、变更内容和影响范围。
- 历史对话可以回看并保留；管理操作、定时任务执行记录和 Credits 明细则为复盘过程提供其他线索，重要结论和变更结果仍需人工核验。
MaxAgent 的价值不在于替用户做所有决定，而在于把知识查询、DataAgent 数据分析、作业诊断、受控管理和周期巡检连接起来。例如，可以先通过 周期巡检发现作业异常，再结合相应能力下钻诊断；对于平台问题，还可以发起受控管理操作并核验执行结果。这样能减少查找、整理与跨页面操作的重复工作，让用户把精力放在判断、确认和处置上。不同能力的开放范围并不完全相同，本文先从适合新用户起步的产品知识问答开始。
完成标准版开通
登录 MaxCompute 控制台，在左侧导航栏找到 MaxAgent，进入独立页面。
使用 MaxAgent 前，需要先开通标准版。如果页面提示尚未开通，请根据页面引导完成标准版开通；已经开通的则可以直接开始对话。
Credits 是 MaxAgent 的用量额度。标准版 0 元开通，每月提供 300 Credits，可以先用免费额度完成体验，再根据需要购买 Credits 包。
确认账号权限
开通服务与访问项目资源是两件事。MaxAgent 涉及到读取项目基本信息、作业日志或表元数据时，会以请求者身份发起请求。请求者身份需要具备查看对应项目、日志或表信息的权限。
如果请求者身份是 RAM 用户，需要联系主账号或管理员确认相应授权，按实际任务配置所需权限。注意，能进入 MaxAgent，也不代表能查看所有项目。
本文先询问通用产品知识，不需要读取某个项目的日志或表数据。
完成第一次提问
第一次体验，可以选择一个不依赖生产数据的问题。例如，准备使用外部表前，先了解它的用途和限制。在会话输入框中发送：
请介绍 MaxCompute 外部表的适用场景、创建前提和常见限制。
请区分通用说明与不同数据源的特殊要求，并附上相关官方文档。
如果某项限制需要结合数据源或格式判断，请明确说明。
这个问法明确了想了解的内容、需要的依据和操作边界。预期回答应解释外部表适合什么场景、使用前需要确认什么，以及哪些限制取决于具体数据源或格式。
如果回答仍然过于宽泛，可以在同一会话中补充上下文：
我的数据存放在 OSS，暂时只希望了解读取方案。
请收窄到 OSS 外部表，说明需要确认的授权和数据格式。
“回答是否正确？”，核验一条事实
收到回答后，选择一条会影响后续决策的结论，回到官方文档核对。比如可以检查：“使用外部表，会不会把源数据复制后存入 MaxCompute？”
MaxCompute 官方文档《外部表介绍》说明，外部表的数据存储于外部系统，使用外部表时不会将数据复制后存储在 MaxCompute。
回看历史对话
在 MaxAgent 独立页面左侧，可以找到历史会话并点击回顾，重要回答和核验依据可以复制文本或截图保存。
看懂 Credits：共享额度、个人限额和用量
Credits 是 MaxAgent 的用量单位，主要覆盖模型 Token 和 MaxAgent 实例拉起的计算资源消费。输入的问题、上下文和生成的回答都会形成消耗；提交到 MaxCompute 引擎执行任务的费用不包含在 Credits 中。本文只做知识问答；使用 DataAgent 执行 SQL 查询或后续场景实际提交 SQL 时，还需另行考虑 MaxCompute 计算资源消耗。
一个租户，也就是一个主账号及其下的相关身份，共享租户可用的 Credits。管理员可以设置租户总限额，并为 RAM User、RAM Role 等身份配置默认个人限额或单独限额。个人限额只是共享额度中的使用上限，并不代表为该身份预留了一份独立 Credits；租户可用余额、租户总限额或个人限额中的任一约束不满足，都会阻断新的 MaxAgent 请求。
Credits限额管理，可进入 MaxCompute 控制台的“智能助手 > MaxAgent管理>限额管理”进行操作。
需要查看额度与用量，可进入 MaxCompute 控制台的“智能助手 > Credits明细”，查看 Credits 总额度、已用额度和剩余额度，以及使用明细。
遇到问题，先判断卡在哪一层
- 如果页面提示尚未开通，先确认标准版开通状态。
- 如果提示无权限，确认当前登录身份和所需授权；涉及具体项目时，还要检查是否具备对应资源权限，不通过扩大授权绕过问题。
- 如果提示 Credits 不足，先区分三种情况，不要通过切换无关账号绕过配额控制。
- 租户共享余额已经用完时，由有权限的管理员按当前站点规则购买或补充 Credits。
- 租户仍有可用额度、但总限额已经达到上限时，检查是否需要调整租户总限额。
- 只有当前身份达到个人限额时，再调整默认个人限额或该身份的单独限额。
- 若问题缺少项目信息、作业信息或时间范围，则先补充上下文；这类输入不完整，与权限不足、额度不足是不同的问题。
下一步：把一个问题问完整
现在打开 MaxAgent，复制示例完成提问，用官方文档核验一条结论，再回看历史会话和额度页面。完成这次练习后，可以选择一个与你工作相关、无需改动资源的问题继续体验。
maxcompute.console.aliyun.com/
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Introduction: When MaxCompute was first used, it was often the case that to learn about a Max Compute function, it was often to search between several documents. When you encounter a failed or slow SQL, you also need to find operational information and then look at the log.
MaxAgent is an AI smart assistant to the MaxCompute platform. You can ask about product knowledge in natural languages or conduct diagnostic analyses in conjunction with specific operational and resource information.
From this point on, we'll introduce MaxAgent's method of use around his daily mission.
The first lesson begins with a simple experience: entering the product, confirming the opening and permission, asking a question and then checking one of the key conclusions of the answer.
This document is suitable for the first time using MaxAgent's data development, traffic, platform administrators and technologists. The example is used only for product knowledge questions and answers, without creating tables, without implementing SQL, without modifying configurations and without having to prepare business data.
Why use MaxAgent?
MaxCompute's daily work cuts across several entry points, including product files, operations logs, resource observations, costs and permission. Many tasks are not complex per se, but time is spent on finding access, supplementing context and checking back and forth. MaxAgent puts these scenes in a natural language dialogue, allowing users to describe the target and gradually complete the project, operation or time frame.
Lower product learning thresholds
When not familiar with a concept or function, questions can be asked directly in the daily language, without first having to have full command and page location. MaxAgent can answer questions about MaxCompute ' s functionality, use scenes and best practices, helping new users to find a faster way to continue learning and doing business.
Direct access to data analysis for operational issues
DataAgent is MaxAgent's data analysis role, corresponding to "checking " in the console. It can search projects, Schema, tables and partitions in natural languages and perform border-based queries, analyse indicators, diagnose anomalies, check data quality and organize the results into data sheets, charts, dashboards or analyses within the data to which the current account is entitled. Users need not first know the full name or know how to prepare SQL, but still need to identify time frames, calibration of indicators and analytical targets, and review the evidence and findings of queries.
Shortening diagnostic and analytical links
In the face of SQL failures, slow running or data tilt, diagnostics can be initiated in conjunction with specific operational information. Platform administrators can also conduct analyses around costs, Quota, storage, permission and metadata.
From analysis to management closed loop
MaxAgent not only analyses problems, but also supports operations such as project management, Quota management and permission management. Users with the corresponding permission can initiate some platform management tasks in natural languages, reducing the cost of switching between multiple pages and commands.
Keep cyclical jobs running
For repeat tasks such as abnormal operations inspection, support the configuration of timed tasks and receive results through nailing robots. In this way, part of the work could move from ad hoc questions to periodic operations and continue the drilling problem on the basis of reports.
permission continues, sensitive operations are controlled, process reversible.
MaxAgent visits and manages resources as a requester and does not bypass the existing permission.
- DataAgent can only read the data that the current account number is entitled to access, which can be used to analyse queries, and to perform a second confirmation of the table or table data in addition to verifying the current account permission.
- For management operations such as projects, Quota, permission, only users with the corresponding permission can initiate it, and confirm or approve it on the actual page; the target, change content and impact range should be checked before execution.
- Historical dialogue can be viewed and preserved; management operations, timed task execution records and Credits rules provide additional clues to the process, and key findings and changes still need manual validation.
The value of MaxAgent is not to make all decisions for users, but to link knowledge queries, DataAgent data analysis, operational diagnostics, controlled management and periodic inspections. For example, it is possible to detect operational anomalies through periodic inspections, which can be diagnosed with corresponding capabilities, and to initiate controlled management operations and validate performance results for platform problems. This reduces the duplication of search, collation and cross-page operations and allows users to focus on judgement, confirmation and disposal. The scope of openness for different capabilities is not identical, starting with product knowledge questions and answers that are suitable for new users.
Standard release completed.
Login to MaxCompute console, find MaxAgent in the left navigation bar, enter the stand-alone page.
Before using MaxAgent, a standard version is required. If the page alert is not yet operational, please direct the standard version according to the page; if it is already open, the dialogue can start directly.
Credits is MaxAgent's use of quota. The standard version of $0 is open and provides 300 Credits per month, which can be used to complete the experience with free quota before purchasing the Credits packages as needed.
Confirm account number permission
Opening up services and visiting project resources are two things. MaxAgent makes a request as a requester when it involves reading basic information, a log or a table metadata of the project. The identity of the applicant requires permission to have access to corresponding items, logs or table information.
If the requesting person is a RAM user, the corresponding authorization will need to be confirmed by contacting the main account number or the administrator to configure permission for the actual task. Attention, access to MaxAgent does not mean access to all items.
This paper asks for generic product knowledge and does not need to read a project log or table data.
Finish your first question.
For the first time, a problem that does not depend on production data can be selected. For example, the use and limitations of external tables are known before they are prepared to be used. Send in session input box:
Please describe the applicable scenes, premises and common limitations of the MaxCompute external table.
Please distinguish the generic description from the special requirements of different data sources and attach the relevant official documents.
If a restriction needs to be judged in relation to data sources or formats, please specify.
This inquiry clarifies what you want to know, the basis for what you need and the operational boundaries. The responses are expected to explain what the external tables are appropriate for, what needs to be identified before they are used, and what limitations depend on the specific data source or format.
If the answer is still too broad, add context to the same session:
My data is stored in OSS and, for the time being, only for reading options.
Please narrow down to the OSS external table, indicating the authorizations and data formats to be confirmed.
“Is the answer correct?” to verify a fact
When a response is received, the choice of a conclusion that would affect subsequent decision-making is returned to the official document for verification. For example, you can check: "Will you copy the source data and deposit them in MaxCompute, using external tables?"
MaxCompute Official Document, An Introduction to the External Table, states that data from external tables are stored in the external system and that when external tables are used, they are not copied and stored in MaxCompute.
Looking back at the historical conversation.
On the left side of the MaxAgent stand-alone page, history sessions can be found and click back, and key answers and verifications can be based on copying text or screenshots.
Credits: sharing quota, personal quota and usage
Credits is a unit of MaxAgent's usage, mainly covering the computing resource consumption from the Token and MaxAgent examples. The input questions, context and the resulting answers are consumed; the costs submitted to the MaxCompute engine for tasking are not included in Credits. This paper is for knowledge questions and answers only; the actual submission of SQLs using DataAgent for SQL queries or follow-up is subject to separate consideration of MaxCompute for resource consumption.
A tenant, i.e. a master account number and related identity under it, shares Credits available to the tenant. The administrator can set a general tenant limit and assign a default personal limit or a separate limit to the identity of RAM User, RAM Role, etc. The personal limit is only the upper limit in the share of quota and does not mean that a separate Credits is reserved for that status; failure to meet any of the available balances of tenants, the general limit of tenants or any of the individual limits would block the new MaxAgent request.
Credits Limits Management, which can be operated by the "Intelligent Assistant > MaxAgent Management > Limits Management" at the MaxCompute Console.
You need to see quota and usage, access to the MaxCompute Conservator's “Intelligence Assistant > Credits Details”, see Credits General quota, quota and the remaining quota, as well as details.
If you have a problem, let's see which level the card is on.
- If the page alert is not operational, confirm that the standard version is open.
- In the absence of permission, confirm the current login identity and the required authorization; in the case of specific projects, check the availability of the corresponding resource permission, without circumventing the issue through the extension of the authorization.
- Distinguishing between three situations, if Credits is not enough, not bypassing quota controls by switching unrelated accounts.
- Credits are purchased or supplemented by administrators with permission in accordance with the current site rules when the tenant share balance is exhausted.
- Tenants are still available at quota, but the total limit has reached above limited-time to check whether the general limit for tenants needs to be adjusted.
- The default personal limit or a separate limit for that status is adjusted only when the current status reaches the individual limit.
- If the problem lacks project information, operational information or time frame, add context; this type of input is incomplete and is not the same as permission or Z81Z.
next step: Full question.
Now open MaxAgent, copy the example to finish the question, check a conclusion in an official document, and look back at the history session and the quota page. Upon completion of this exercise, you can choose to continue to experience an issue that is relevant to your work and does not require changes in resources.
maxcompute.console.aliyun.com/
Auto-fetched and translated by AI Atlas. All rights belong to the original author.