Understand Built-in Agents

Built-in Agents are ready-to-use Agents provided by the platform. End users can use them from the chat page for modeling, querying, data operations, and business analysis. Developers can also use them as reference designs for combining ReActAgent, DeepAgent, Skills, file spaces, data model tools, service function tools, privacy protection, and role permissions.

This page focuses on how to choose the right built-in Agent and what design patterns developers can reuse.

Choose by Task

Task	Recommended Agent	Output
You have business requirements but no existing tables	Requirement-based Model Generation and Maintenance	Data model creation or maintenance
You already have database tables or views	Database-based Model Generation and Maintenance	Data model creation or maintenance
You want to inspect table or view structures first	Database Table/View Schema Profile	Schema profile file
You already have a schema profile and want model configuration	Generate Model Configuration from Schema Profile	Model configuration submitted to the platform
You want to query, create, update, or delete business data	Data Operation/Query Agent	Daily data operations
You want to turn a business question into clear metric definitions	Data Metric Design	Metric definition and query requirements
You have clear query requirements and need data results	Data Retrieval Execution	Query result files
You want to turn an existing model into a reusable query Skill	Model-based Query Skill Development	Skill element installable in other Agents
You want to answer an open-ended business question	Business Data Insight Analysis	Analysis report

Data Operations and Query Agents

This group works on existing data models and lets users process data through natural language. It includes two levels:

• Data Operation/Query Agent
• Data Retrieval Execution

Data Operation/Query Agent is for daily business actions such as query, create, update, and delete. Data Retrieval Execution is for structured query tasks where the target, fields, filters, and metrics are relatively clear.

Use Data Operation/Query Agent when the user wants to perform a business action, such as creating a customer, updating a status, or deleting a record. Use Data Retrieval Execution when the user already has a clear query goal and needs stable retrieval results.

This separation keeps daily operations simple while allowing structured data retrieval to stay stable enough for reuse by other Agents and query Skills.

Query Skill Development Agent

Model-based Query Skill Development is for developers. Use it when a data model already exists and you want to capture common queries, metric definitions, and TQL templates as a reusable Skill.

This Agent does not return a one-off query result. Its output is a Skill element that usually contains:

• SKILL.md: when to use the query Skill and which references to read.
• references/data-model.md: full field definitions, field types, enum values, null rules, and type risks.
• references/query-template.md: verified detail query, combined filter, aggregate, and TopN templates.

The Agent works by checking existing Skills, asking Data Retrieval Execution to generate and verify templates, writing only verified TQL templates, saving the Skill element into the current app's skills directory, and reading it back for final checks.

You can prompt it like this:

Build a customer order query Skill based on models.sales.CustomerOrder.
Cover order detail queries, customer/status/time filters, order amount summaries, status distribution, and customer order amount TopN.

After the Skill is generated, install it in the target business assistant or data query Agent. Use the same configuration flow as other Skills: add the generated Skill to the target Agent's Skill list. The target Agent can then read the generated field definitions and query templates when answering related questions.

Use this Agent when repeated questions should become maintainable Skill content. If there is no model yet, use a modeling Agent first. If you only need a one-time data result, use Data Retrieval Execution.

Related docs:

• Build Query Skills
• Use Skills in Agents
• Skill Best Practices

Reusable Design Patterns

Built-in Agents repeatedly use several platform capabilities:

Platform capability	Typical role
ReActAgent	Handles bounded single-agent tasks
DeepAgent	Coordinates multi-step workflows and sub-Agents
Skills	Captures stable reusable professional methods
Data model tools	Read and write business data
Service function tools	Execute complex business actions
Tool exposure	Controls which candidate tools are visible to the model
Role permissions	Controls what the current user can actually execute
Privacy protection	Masks or hashes sensitive information when needed

The most important lesson is not a particular prompt. It is the separation of responsibilities: use bounded scene Agents for clear tasks, use expert Agents for open-ended analysis, capture stable professional methods as Skills, and keep permissions separate from model-visible tool exposure.