The Agent Development Platform supports building applications in Multi-Agent Mode. In this mode, you can orchestrate multiple Agents within one application, enabling them to collaborate and complete complex tasks.
This guide walks you through the process of creating a Multi-Agent application using the example of a “Stand-Up Comedy Material Assistant.”
Prerequisites
You have a registered Tencent Cloud account with identity authentication completed.
2.2 Click New Application, enter application name and application description, then select Multi-Agent Mode for Application mode.
2.3 Click Create. You will be redirected to the application settings page.
Step 2: Create and Configure Agents
To build the “Stand-Up Comedy Material Assistant,” you’ll need four Agents, each with a specific role:
Agent Name
Agent Feature
Stand-up Comedy Content Coordinator
Main Agent. Handles user input, decides when to transfer tasks, and delegates to sub-agents.
Fetch
Sub-Agent. Fetches and parses webpage content.
Summarizer
Sub-Agent. Summarizes content into key points.
Joy
Sub-Agent. Expands summaries into full stand-up comedy routines.
2.1 Setting Up the Main Agent
When switching to Multi-Agent Mode, the system automatically creates a Main Agent with the same name as your application. You can edit this name at any time.
Note:
For single-Agent apps, a neutral name is fine. For Multi-Agent apps, choose meaningful names so the Main Agent can accurately delegate tasks.
2.1.1 Set Model
Click Model Settings and choose a reasoning model that can understand user queries and decide when to hand off tasks to sub-agents.
2.1.2 Set Transfer Description
The transfer description explains this Agent’s purpose and helps the model decide when to route tasks.
Example of a transfer description:
The Stand-Up Content Coordinator is the core dispatcher. It manages multi-agent orchestration and handles ambiguous queries.
2.1.3. Set prompt content
In the Prompt section, provide clear instructions to guide the model.
Click Template to use an automatically generated prompt content, making it easier start and standardize the input format.
You can also use AI one-click optimization to optimize the prompt content entered by restructure your input prompts into a more model-friendly format.
Note:
The one-click optimization feature will consume token resources.
You can refer to the following prompt:
# Task objective
You are the core dispatcher. Your sole responsibility is managing multi-agent orchestration and handling ambiguous queries. You do not perform business-specific tasks.
# Limit
If the user input is not a URL, immediately respond: “Input error. Please provide a valid link.”
2.2 Setting Up the Sub Agent Fetch
The Sub-Agent Crawler is responsible for fetching webpage content and return it to the Summarizer.
2.2.1 Adding a Sub Agent
Click Add Agent and select Create.
2.2.2 Setting Up a Sub Agent
2.2.2.1 Fill in the Agent name, model, transfer description, and prompt.
2.2.2.2 Search for the Fetch plug-in and add it.
Example transfer description:
Responsible for fetching webpage content. Parses input links and retrieves corresponding data.
Example Prompt:
Enter / in the prompt content to introduce the added plugin.
# Task objective
As the content fetching agent, fetch the webpage corresponding to the user’s link and pass it to the Summarizer Agent.
# Task flow
1. Use the fetch plugin to retrieve webpage content.
2. Provide the content to the Summarizer Agent.
2.3 Setting Up the Sub Agent “Summarizer”
Summarizer is responsible for summarizing Summarize the content from the Crawler into concise, humorous points.
2.3.1 Adding the Sub Agent
Click Add Agent and select Create.
2.3.2 Setting Up the Sub Agent
Fill in the Agent name, model, transfer description, and prompt.
example transfer description:
Responsible for summarizing input content.
Refer to the following prompt content:
# Task objective
As the comedy material summarizer, transform the Crawler’s output into witty, humorous highlights, and send them to the Humorist Agent.
# Restriction
Keep it concise. No more than 100 words.
2.4 Setting Up Sub Agent "Joy"
Joy is responsible for expanding summaries into stand-up routines.
2.4.1 Adding the Sub Agent
Click Add Agent and select Create.
2.4.2 Setting Up the Sub Agent
Fill in the Agent name, model, transfer description, and prompt.
Example transfer description:
Responsible for generating stand-up routines by adapting input content.
Refer to the following prompt content:
# Task objective
As the stand-up routine creator, expand on the Summarizer’s points. Incorporate internet slang (“fam”, “let’s be real”), audience interaction (“ever happened to you guys?”), and punchy contrasts (“It’s not that… it’s that…”). Deliver a full routine.
# Restriction
Aim for humor and completeness. Output no more than 200 words.
2.5 Setting Up Agent Transfer Relationship
When multiple Agents exist, you must explicitly configure their transfer relationships. By default, additional Agents will not be active until the Main Agent is linked to them.
After completing these settings, you may also add welcome words, voice, or avatars to enhance user experience.
Step 3: Test the Application
Use the debugging panel on the right to verify that your application behaves as expected.
Step 4: Release the Application
Once testing meets expectations:, click Release to release the application.
After the application is released, go to Application Release > Service Status to get the Experience Link and API call information.
To publish your application on platforms such as WeChat or WeCom, go to Application Release - Release Channel Settings. For details, refer to Application Release Overview.
You have now created a "Stand-up Comedy Content Creation Assistant" in "Multi-Agent Mode". For instructions on creating applications in Standard mode and Single Workflow mode, refer to:
