All Products
Search
Document Center

Alibaba Cloud Model Studio:Model Studio application overview

Last Updated:Mar 10, 2026

Large language models (LLMs) excel at understanding and generating text, but they have limitations, such as difficulty handling private data, accessing the latest information, following fixed processes, and planning multi-step tasks on their own. Model Studio addresses these gaps with three application types -- agent applications, workflow applications, and agent orchestration -- that add knowledge base retrieval, workflow design, and agent collaboration on top of the base model. Agent orchestration has been replaced by workflow applications.

If you are new to AI application development, start with an agent application. Switch to a workflow application when you need deterministic control over each step.

Important

Agent application

An agent application (labeled Assistant in the console) is a conversational AI application that completes tasks through contextual dialogue, autonomous decision-making, and tool calling.

When to use an agent application:

  • You need a chat-based interface for natural-language interactions.

  • Prompt engineering is the primary mechanism for shaping model behavior.

  • You want to connect knowledge bases or call external tools (official or custom plug-ins) without designing a multi-step pipeline.

Agent applications require no code. Define prompts, attach a knowledge base, enable plug-ins -- the model handles the rest. The simple configuration process and intuitive conversational output format make agent applications ideal for beginners to quickly create single-function AI applications.

For setup instructions, see Agent application.

Workflow application

A workflow application (labeled Workflow in the console) is a procedural AI application that breaks tasks into discrete subtasks represented as nodes, giving you explicit control over each step.

When to use a workflow application:

  • The process must be fixed and repeatable, with each step producing a defined input and output.

  • The task involves branching logic, text conversion, or integration with external services such as Function Compute.

  • Deterministic control matters more than prompt flexibility.

Workflow applications require no code. Drag and drop nodes onto a canvas, configure each node, and connect them to build your task flow.

For setup instructions, see Workflow application.

Feature comparison

Feature

Agent application

Workflow application

Creation method

No-code

No-code

Interaction model

Conversational

Procedural

Prompt dependency

High -- the model relies on prompts for behavior, branching, and planning

Low -- node configuration defines behavior

Knowledge base

Supported

Supported

Tools

Supported (official and custom plug-ins)

Supported (plug-in node, API node)

Text conversion

Limited -- depends on prompt instructions

Supported (text conversion node, script conversion node)

External service integration

Supported (custom plug-ins)

Supported (Function Compute node)

Branching

Limited -- depends on prompt instructions

Supported (logical branching, intent recognition)

Automatic planning

Limited -- depends on prompt instructions

Supported (agent group node)

Agent applications offer a faster path to a working prototype through conversational interaction and prompt tuning. Workflow applications provide deterministic control, making them better suited for production pipelines that require repeatable, auditable steps.

Access restrictions for International Edition

The following restrictions apply to International Edition users:

All features listed above are in preview. Use them with caution in production environments.

Application Development tab in the Model Studio console