What is ADA

Updated at:
Copy as MD

AI-Native Big Data Service (ADA) is an enterprise-grade full-stack big data intelligence service provided by Alibaba Cloud. Delivered through an agent-based paradigm, ADA unifies data analysis, data engineering, data governance, and intelligent O&M into a single conversational interface, allowing you to complete data retrieval, analysis, development, and O&M tasks using natural language.

In traditional big data workflows, data retrieval requests depend on manual scheduling, metric definitions are scattered across team documents, and cross-engine development requires switching between multiple consoles. ADA delegates task execution to agents, unifies business definitions through a centralized knowledge base, and hands off periodic jobs to a scheduling service, reducing manual effort across these stages.

Compared with using engine consoles directly, ADA stands out in three aspects. Tasks are executed by agents, and repetitive jobs can be completed on schedule by the scheduler. Business definitions come from a centrally maintained knowledge base, ensuring that the same metric returns consistent results across different sessions. Engine capabilities are subscribed to as official agents, so adding a new engine does not require deploying an additional console.

ADA adopts a layered architecture that includes, from bottom to top, the data connection layer, the knowledge layer, the agent layer, and the interaction layer. The AI capability center and enterprise management span across all layers.

Layer

Description

Data connection layer

Connects to engine instances through connectors. Supported engines include MaxCompute, Hologres, Serverless StarRocks, ClickHouse, SelectDB, and RDS MySQL.

Knowledge layer

Maintains semantic models, knowledge graphs, and memory in the knowledge base, providing business definitions and context for agents.

Agent layer

Consists of the primary agent Ada and domain-specific agents. Responsible for intent understanding, task decomposition, and collaborative execution.

Interaction layer

Provides two types of entry points: the AdaWorks web console and IM channels.

AI capability center

Provides a discovery and subscription portal for official agents, skills, and connectors.

Enterprise management and subscription billing

Provides management capabilities for data permissions, usage, and costs.

Key concept

The concepts in ADA are organized around three themes: execution entities, business definitions, and capability sources. The following table lists the core terms that appear frequently.

Term

Description

ADA

AI-Native Big Data Service. Refers to the product as a whole.

Ada

The primary agent of ADA and the unified conversational interface. Ada understands intents, decomposes tasks, dispatches work to the appropriate domain-specific agents, and provides subscription recommendations when capability gaps are identified.

AdaWorks

The web-based collaborative console of ADA, used to complete data tasks through conversations.

Agent

An execution unit with a defined role, system prompt, tools, and skill configurations. Agents are classified into official agents and custom agents.

Domain-specific agent

A built-in agent designed for a specific work scenario. Domain-specific agents include the data analysis expert, semantic modeling expert, data engineering expert, data governance expert, intelligent O&M expert, real-time data warehouse management expert, streaming data management expert, search analytics expert, and data intelligence solution consultant.

Official agent

An agent published by an engine product team and available for subscription in the capability marketplace, such as the MaxCompute agent, DataWorks data agent, and Hologres agent.

Skill

A reusable capability unit that stores a fixed workflow, input requirements, and output format. Skills are invoked by agents.

MCP

Model Context Protocol, an open protocol that allows agents to access external tools and resources. An MCP service exposes external system capabilities to agents.

Connector

The configuration for connecting to data sources and external capabilities. Big data engine connectors connect directly to engine instances, and MCP service connectors connect to MCP Servers.

Knowledge asset

A collective term for semantic models, knowledge graphs, memories, and other content in a knowledge base. Knowledge assets serve as the source of context and criteria for agent outputs.

Knowledge base

A centralized repository of business terms, metrics, and statistical criteria, organized by semantic models, knowledge graphs, and memories. Knowledge bases ensure the accuracy of natural language data retrieval.

Knowledge graph

A graph asset formed by extracting entities, relationships, and attributes from table schemas and business documents. Knowledge graphs support visual exploration.

Memory

A memory asset that stores preferences and context accumulated during conversations. Memories reflect individual usage habits.

Scheduled task

A task that is persistently scheduled by the system and executed on a predefined schedule in an independent session.

Artifact

Deliverables produced by tasks, such as reports and data tables. Artifacts can be browsed, searched, shared, and exported.

AI Capability Center

A central hub for discovering, subscribing to, and managing official agents, skills, and connectors.

Credits

The unit of measurement in ADA that tracks agent invocations, model consumption, and usage of official capabilities.

Seat

The unit of subscription specification. The number of seats determines the total credits allocated each month, without limiting the number of members.

Tenant

The isolation boundary for resources and data, which corresponds to the Alibaba Cloud account that activated the service. When a resource is set to be visible to all members within the tenant, all members of that tenant can use it.

Service-linked role

An Alibaba Cloud service-linked role (SLR) that allows ADA to access resources in other Alibaba Cloud services based on granted permissions.

The following three pairs of concepts are often confused during selection. The following tables describe their differences.

Skills vs. MCP services

Dimension

Skill

MCP service

Purpose

Defines how an agent completes a specific type of work.

Provides callable external tools to agents.

Content

Fixed workflows, constraints, input requirements, and output formats.

Capability interfaces of external systems.

Typical examples

Daily/weekly report generation, data quality inspection.

Hologres MCP Server, self-managed business system MCP Server.

Key configuration points

Descriptions, input requirements, and failure handling must be complete.

You must complete connectivity detection and verify actual calls.

Official agents and custom agents

Comparison dimension

Official agent

Custom agent

Publisher

Platform or engine product teams

Tenant members

How to obtain

Subscribe in the Capability Gallery

Create in the Workbench, or copy and customize a built-in agent

Updates and maintenance

Continuously updated by the publisher

Maintained by the creator

Typical use cases

In-depth engine O&M and development assistance

Custom agents tailored to team workflows

Conversations and scheduled tasks

Comparison dimension

Conversation

Scheduled task

Trigger method

Manually initiated

Automatically triggered on a schedule

Interaction method

Multi-turn conversation with follow-up questions at any time

Unattended execution with reviewable session history. After an IM channel is configured, results can be delivered to the IM channel.

Applicable scenarios

One-time exploration and metric definition verification

Recurring tasks and continuous monitoring

Use case

Metric anomaly investigation

When business metrics fluctuate, manual scheduling for data retrieval delays conclusions. Describe the issue directly to Ada, and the data analysis expert agent identifies the anomaly scope, breaks down dimensions, and provides root cause clues with metric definition references. This scenario requires knowledge base configuration in advance to ensure consistent metric definitions across multi-turn follow-up questions. For the operation procedure, see Start a conversation.

Automated daily and weekly report generation

The data retrieval and formatting process for daily and weekly reports is stable but time-consuming. Save the process as a skill, package it as a scheduled task that runs on a periodic basis, and deliver the output through an IM channel. The delivery requires no manual intervention and does not occupy dedicated resources. For the configuration method, see Scheduled tasks.

Cross-engine intelligent O&M

When the platform spans offline data warehouses, real-time data warehouses, search engines, and machine learning platforms, the entry points for job diagnostics and query optimization are isolated from each other. Subscribe to official agents by engine in the Capability Gallery. Diagnostic capabilities are automatically available as the engine product team releases updates, and O&M operations are centralized in a single conversation entry point. For the subscription procedure, see AI Capability Center.

For more scenarios and selection recommendations, see Scenarios.

FAQ

  • Q: What is the relationship between Ada and scenario-specific agents?

    A: Ada serves as the unified entry point, responsible for intent understanding and task dispatching. Scenario agents handle execution in specific domains. When you ask Ada a question directly, Ada dispatches the task to the matching scenario agent. You can also use @ in the input box to specify an agent for direct conversation.

  • Q: What is the difference between knowledge assets and data connections?

    A: Data connections address the data sourcing problem by establishing access paths to data sources. Knowledge assets address the data comprehension problem by defining metrics, entities, and business semantics. Without either, natural language data retrieval cannot produce reliable results.

  • Q: Compared with using the engine console directly, what problems does ADA solve?

    A: The engine console is designed for users who are already proficient in SQL and engine features, and organizes operations by tool boundaries. ADA organizes execution by business goals. Cross-engine data retrieval, metric alignment, result formatting, and periodic scheduling are completed within a single conversation. Every tool invocation is logged for post-hoc verification. When a knowledge base is associated, conclusions also include metric references.

  • Q: What is the relationship between credits and seats?

    A: A seat is a subscription specification that determines the total number of credits allocated each month. Credits are the unit of measurement for actual consumption. The number of seats does not limit the number of members. For more information, see the billing overview.