Agent application - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center

You can quickly create agent applications with zero code in the Alibaba Cloud Model Studio console and integrate retrieval-augmented generation and plugin capabilities. After the application is created, you can use it through the console or API.

Important

Only users who created Model Studio applications before April 21, 2025, can access the Application tab and use all its features, including applications (Agent applications, Workflow applications, Agent orchestration applications), components (Prompt, Plug-in), data (Knowledge base, Application data) features and related APIs. This feature is in preview. Use with caution in production environments.

Why use agent applications

Large language models have powerful language understanding and generation capabilities. However, they have certain limitations, such as:

They cannot answer questions about private domains, such as company policies and personnel information.
They cannot obtain the latest information, such as real-time weather and match results.
They cannot accurately answer professional questions, such as mathematical calculations and image generation.

You can use agent applications to solve these problems. An agent application is an application based on large language models that integrates specific external capabilities according to business scenarios.

External capabilities here refer to capabilities that the large language model itself does not have, such as obtaining real-time information and answering questions in private domains.

You can create agent applications with zero code in the Alibaba Cloud Model Studio console and integrate external capabilities into the applications to solve your business problems.

Agent applications have many use cases, such as:

Private domain knowledge Q&A
You only need to prepare knowledge base files to quickly create a private domain knowledge Q&A application in the Alibaba Cloud Model Studio console, which can be applied to Q&A scenarios such as company policies and personnel information.
Intelligent assistant
By introducing retrieval-augmented generation capabilities, and custom plugins, you can build an intelligent assistant to help you improve work efficiency, such as processing emails and writing weekly reports.

Step 1: Create an agent application

1. Create an application

Visit the Alibaba Cloud Model Studio console Application Management, click Add Application, and on the Agent Application tab, click Create Now.

If you have already created an application before, click Add Application in the upper right corner.

2. Select a large language model and configure parameters

After entering the agent application management interface, you need to select a large language model and configure parameters. You can refer to the image on the right, click Settings, and select a model from the Model Selection dropdown menu. Here we select Qwen-Max. You can configure model parameters according to your needs. For parameter details, move your mouse over to view. If you have no specific requirements, you can keep the default settings.

Note

DeepSeek-R1 series models do not support setting temperature coefficient.

3. Test the application

After selecting a large language model, you have completed creating an agent application in Alibaba Cloud Model Studio. You can enter questions to test it.

Step 2: Extend the capabilities of the agent application

In the agent application management interface, you can extend the capabilities of the agent application by setting prompts, adding RAG and plugins.

System prompt: Guide the direction of content generation

You can enter a prompt in the agent application management interface to guide the direction of content generation. For example, set the prompt to:

Please imitate Li Bai's tone.

Enter "hello", and the output of the agent application is shown on the right.

Note

For DeepSeek R1 series models, DeepSeek official does not recommend setting system prompts. The original text is as follows: "Avoid adding a system prompt; all instructions should be contained within the user prompt."

Retrieval-augmented generation (RAG): Enhance private domain knowledge Q&A capabilities

RAG (Retrieval-Augmented Generation) is a retrieval technology that automatically retrieves relevant document content and inputs this information as context to the agent, thereby improving the accuracy of the agent's responses. Especially when dealing with private knowledge or vertical domain Q&A, RAG can significantly enhance the effectiveness of the agent.

Tip: The text retrieved by RAG also occupies the context window length of the large language model. Therefore, you need to adjust the retrieval strategy and text length according to the actual situation to fully utilize the context window and avoid exceeding the limit.

1. Upload data

Visit Application Data, click Import Data. Here we select Model Studio Series Mobile Product Introduction.docx as the knowledge base file. In the Import Data interface, import the file via Local Upload.

2. Create a knowledge base

Visit Knowledge Base, click Create Knowledge Base. In the Create Knowledge Base interface, enter the Knowledge Base Name and Knowledge Base Description. Here we set the knowledge base name as: Model Studio Mobile Product Introduction, and the knowledge base description as: This knowledge base contains detailed introductions to Model Studio mobile products. Keep other options as default and click Next.
Click Select File, click Default Category for the category location, and select Model Studio Series Mobile Product Introduction for the File Name. Click Next.
For data processing, use the default Intelligent Chunking. Click Import Complete.
When the status is Parsing Complete, it indicates that the knowledge base creation is complete.

3. Integrate RAG into the application

Visit Application Management, click the Manage button on the created application card to enter the agent application management interface. Click the Knowledge Base button, and instructions for the large language model to reference the knowledge base will be automatically filled in the Prompt. Click +Knowledge Base, then select Model Studio Mobile Product Introduction from the list, and keep other options as default.
Knowledge Base Configuration (optional):
If you need to configure fragment assembly strategies, set knowledge scope, display answer sources, you can further modify the retrieval configuration. For specific configurations, see Step 4: Use Knowledge Base.
Sample Library (optional):
When enabled, the ${samples} variable is added to the prompt to implement few-shot prompt example introduction, improving generation effects.
- Sample Library: Add cases.
- Configuration: Set the number of retrieved fragments.
Enter a question in the input box:
I like playing games, which phone do you recommend?
As you can see from the image on the right, after integrating RAG capabilities, the agent application can answer questions based on the uploaded knowledge base file.
Agent applications currently only support generating text content.

Plugins: Access images, videos, code execution, and more efficiency tools

Plugins can make agent applications more efficient and professional in business scenarios. Alibaba Cloud Model Studio provides various official plugins. You can click Select Plugin in the agent application management interface to select one or more plugins from the Plugins column.

You can also add custom plugins. For details, see Plugin Overview.

1. Add a plugin

Taking the Image Generation plugin as an example, select the corresponding plugin and click Add To Agent.

2. Test the plugin effect

After adding the plugin, enter the following in the input box:

Draw a cat.

The agent application's response is shown on the right. You can see that the agent application called the image generation plugin (i.e., text_to_image).

3. Use more plugins

You can select up to 10 plugins in the same agent application. The agent application will choose to use one or more plugins based on the input. For example, ask:

Check the weather in Hangzhou tomorrow, and draw West Lake with this weather.

The agent application will output the content shown on the right.

Step 3: Experience the agent application

You can experience the agent application using text.

Video conversation is limited to Qwen VL series models

Step 4: Publish the agent application

Published applications can be called by API, and can also be shared with RAM users under the same Alibaba Cloud account through a web page. You can click the Publish button in the upper right corner of the agent application management interface. If your application is not being published for the first time, a popup will show you the modifications to the current agent application compared to the last published version. After confirming the modifications, click Confirm Publication.

Note

If you use a RAM account to create an application, please confirm that you have the service-linked role permission ram:CreateServiceLinkedRole before publishing the application. For details, see Service-linked Role.

If your application is being published for the first time, there will be no popup information.

Call via API

You can click View API to the right of API Call in the Publication Channels tab of the agent application to view how to call the agent application via API.

Note: You need to replace YOUR_API_KEY with your API KEY to initiate the call.

Share channel application

Click Generate Sharing Link to the right of the official channel. You can share this link with RAM users under the same Alibaba Cloud account as needed.

Step 5: Delete and copy the agent

You can find the published application card in Application Management, and perform More > Copy Application/Delete ApplicationDelete and Copy Agent, Modify Application Name operations in .

Step 7: Version management

If you need to edit historical version descriptions or select and use previously published versions, you can use the version management feature. Follow these steps:

On the Configuration tab of your agent application, click Version Management in the upper-right corner of the navigation bar.
In the historical version list, select your target version:
- If you need to modify the version information, hover your mouse over the icon and click it. In the Edit Version Description dialog box, make your changes as needed, then click OK.
- If you need to use this version, click Overwrite Current Draft. In the confirmation dialog box, click Confirm.
  This version content will overwrite the current draft content.

FAQ

Is there an additional charge for knowledge base retrieval?
The retrieval-augmented generation feature is free. However, the retrieved content is charged as model input tokens.
Is there a timeout limit for custom plugins?
Yes, the timeout limit is 5 seconds.
Can I create agent applications through API?
You can use the Assistant API to create model applications with features similar to agent applications.
Note: Applications created with the Assistant API cannot be managed in the console. For more information, see the Assistant API reference.