Meta Agent is an enterprise-grade intelligent agent for data asset management in AI-Native Database Service. Powered by large language models (LLMs), Meta Agent provides two core capabilities: Asset Inventory and Asset Q&A. These capabilities help you manage metadata efficiently, build a data asset knowledge system, and discover and analyze data assets through natural language interaction.
Overview
Enterprise data management faces common challenges: metadata is scattered across systems, business descriptions are missing, and data lineage is unclear. These issues severely limit data discovery efficiency and governance quality. Meta Agent leverages LLM technology to transform metadata management from a manual process into an AI-driven workflow, enabling automated cataloging and intelligent management of data assets.
Core capabilities
Meta Agent provides the following two core capabilities:
Asset Inventory
Asset Inventory automates knowledge distillation and system-building for data assets. It serves as the foundation of Meta Agent and supports the following features:
-
Automated metadata scanning: Automatically collects metadata from databases, including table structures, column definitions, indexes, and constraints. No manual entry is required.
-
Intelligent description generation: Analyzes table structures and data samples by using LLMs, and then generates business descriptions in Chinese for tables and columns. This improves the readability and understandability of metadata.
-
SQL comment completion: Generates SQL COMMENT statements for tables and columns that lack comments. You can write comments back to the database in bulk.
-
Data lineage analysis: Analyzes SQL logs and ETL workflows to automatically build table-level and column-level data lineage, which helps you understand data origins and flow directions.
-
Business classification and cataloging: Automatically classifies tables based on their business semantics and generates a multi-level business catalog for organized asset discovery and retrieval.
Asset Q&A
Asset Q&A builds on the knowledge accumulated through Asset Inventory to deliver natural-language-driven intelligent interaction, lowering the barrier for data consumption. Asset Q&A supports the following features:
-
Natural language asset search: Describe your data needs in plain language to locate target data assets. You do not need to know specific database or table structures. For example, enter "Find the tables that store user order information" to locate the relevant tables.
-
Intelligent analysis and recommendations: Based on your query intent, Meta Agent automatically recommends related data tables, associations, and usage suggestions to help you understand the full picture of your data assets.
-
Asset Q&A: Interact with the system in a question-and-answer format to inquire about the business meaning, usage specifications, and quality status of data assets.
Editions
Meta Agent currently supports only the subscription (prepaid) billing method. It provides three editions: Basic Edition, Professional Edition, and Enterprise Edition. Each edition provides two specifications: Database Edition and Data Lakehouse Edition.
Edition description
|
Edition |
Key differences |
Applicable scenarios |
|
Basic Edition |
Supports automatic inventory of data sources. The AI automatically generates business descriptions for tables and fields, which users can accept or modify to form standardized data asset documentation. |
|
|
Professional Edition |
In addition to the capabilities of the Basic Edition, the Professional Edition supports full asset inventory and natural language Q&A. Users ask questions in natural language, and the AI automatically understands the intent based on the knowledge base, generates queries, and returns results and insights, forming an end-to-end data consumption loop. |
|
|
Enterprise Edition |
In addition to the capabilities of the Professional Edition, the Enterprise Edition provides all Meta Agent capabilities, including data lineage management, ontology modeling, approval workflows, knowledge base governance (version control, permissions, and accuracy assurance), and knowledge flywheel evolution (signal-driven automatic corrections and skill accumulation). Private deployment and enterprise-grade auditing are supported. |
|
Supported database types
ApsaraDB RDS for MySQL, ApsaraDB RDS for PostgreSQL, ApsaraDB RDS for SQL Server, PolarDB for MySQL, and PolarDB for PostgreSQL
Supported lakehouse types
AnalyticDB for MySQL, AnalyticDB for PostgreSQL, MaxCompute, MongoDB, and Hive
Feature comparison
|
Category |
Feature details |
Basic Edition |
Professional Edition |
Enterprise Edition |
|
Supported number of instances |
Maximum number of instances |
Up to 5 database instances and 5 lakehouse instances |
Unlimited |
Unlimited |
|
Asset inventory |
Structured data sources |
|
|
|
|
Unstructured data sources |
|
|
|
|
|
Document and SQL code import |
|
|
|
|
|
Automatic metadata scanning |
|
|
|
|
|
Sampling and profiling statistics |
|
|
|
|
|
Asset catalog generation |
|
|
|
|
|
Table and field description generation |
|
|
|
|
|
SQL template generation |
|
|
|
|
|
Metric definition generation |
|
|
|
|
|
Usage guide generation |
|
|
|
|
|
Ontology entity, property, and relationship generation |
|
|
|
|
|
Cross-table and cross-source relationship inference |
|
|
|
|
|
Sensitive data scanning and identification |
|
|
|
|
|
Inventory tasks |
Manual targeted inventory |
|
|
|
|
Question-driven on-demand inventory |
|
|
|
|
|
Scheduled inventory tasks |
|
|
|
|
|
Triggered inventory (signal-triggered) |
|
|
|
|
|
Change monitoring |
|
|
|
|
|
Review and collaboration |
Knowledge acceptance and modification |
|
|
|
|
Reasoning basis display |
|
|
|
|
|
Correction guidance and related impact preview |
|
|
|
|
|
Batch delivery |
|
|
|
|
|
Multi-person approval workflow |
|
|
|
|
|
Knowledge base management |
Viewing and modifying field and table descriptions |
|
|
|
|
Viewing and modifying business entities and relationships |
|
|
|
|
|
Viewing and modifying metric definitions |
|
|
|
|
|
Adding custom knowledge types |
|
|
|
|
|
Knowledge version management and rollback |
|
|
|
|
|
Knowledge graph management |
|
|
|
|
|
Intelligent Q&A and governance |
Natural language Q&A |
|
|
|
|
Automatic SQL and API generation for queries |
|
|
|
|
|
Result presentation |
|
|
|
|
|
Result feedback loop |
|
|
|
|
|
External agent integration |
|
|
|
|
|
Semantic search and precise matching |
|
|
|
|
|
Call authentication |
|
|
|
|
|
Call logging and auditing |
|
|
|
|
|
Usage control |
|
|
|
|
|
Conflict detection |
|
|
|
|
|
Redundant knowledge identification |
|
|
|
|
|
Knowledge status management |
|
|
|
|
|
Knowledge expiration detection |
|
|
|
|
|
Governance rule configuration |
|
|
|
|
|
Governance metrics dashboard |
|
|
|
|
|
Governance reports |
|
|
|
|
|
Knowledge flywheel evolution |
Explicit feedback collection |
|
|
|
|
Implicit signal collection |
|
|
|
|
|
Signal dashboard |
|
|
|
|
|
Signal analysis |
|
|
|
|
|
Update strategy recommendations |
|
|
|
|
|
Automatic knowledge update and notification |
|
|
|
|
|
Skill and action accumulation |
|
|
|
|
|
Roles and permissions |
Number of agents |
Not supported |
Only 1 |
Unlimited |
|
Knowledge permissions |
|
|
|
|
|
Role definition and assignment |
|
|
|
|
|
Single sign-on |
|
|
|
|
|
Operation audit logs |
|
|
|
Supported regions
-
China (Hong Kong), Singapore, and Malaysia (Kuala Lumpur).
Access Meta Agent
To access Meta Agent, perform the following steps:
-
Log on to the AI-Native Database Service console.
-
In the left-side navigation pane, in the AI Data Assets section, click Meta Agent.
-
On the Meta Agent page, select Asset Inventory or Asset Q&A from the feature cards to start using the corresponding feature.
Before you use Meta Agent for the first time, register your data sources. For more information, see the data source management documentation.
Scenarios
Integration scenarios
The business semantic knowledge produced by Meta Agent serves as the knowledge foundation for AI-powered data applications. By incorporating the business semantic layer from Meta Agent, you can significantly improve AI accuracy and effectiveness in the following scenarios:
|
Module |
Without Meta Agent |
With Meta Agent |
|
Analytics Agent (data analysis) |
Abbreviated column names may cause incorrect SQL generation. The system must scan many tables to locate the target data. |
Business descriptions enable precise identification of target tables and columns, which significantly improves NL2SQL accuracy. |
|
DAS Agent (database O&M) |
Detects only that a SQL query is slow and prioritizes optimization by technical metrics alone. |
Identifies that a query belongs to a critical transaction path and prioritizes issues based on business importance. |
|
NL2SQL applications |
Requires manually writing extensive prompts to describe table structures. Business terms such as "revenue" cannot be mapped to columns like sale_amt. |
Directly uses structured business descriptions generated by Asset Inventory. Column mapping and enumeration value interpretation are automated. |
|
Data governance |
Requires manual classification and sensitivity labeling for each table and column. This approach is costly and has low coverage. |
Automatically identifies sensitive columns based on business descriptions and uses data lineage to assess the impact scope of changes. |
Data privacy and security
|
Security commitment |
Description |
|
Security hosting |
Database metadata is accessed through the security hosting mechanism of AI-Native Database Service. Database connection information is never directly exposed. |
|
Knowledge isolation |
Inventory knowledge and Q&A content are isolated between different accounts. |
|
Metadata-only processing |
Asset Inventory analyzes only table structures and metadata. It does not read business data content. |
|
Access control |
Access is controlled based on the RAM permission system. |