Custom Agents

Categories of Local Custom Agents

Creating a Local Custom Agent on the GUI

1. Log in to CodeArts Agent IDE as instructed in Quick Start.
2. Click in the upper right corner to open the global setting page.
3. In the navigation pane on the left, choose Agents.
4. Click Create Agent.
5. Set agent parameters and click Create.

   Table 2 Parameters for creating a local custom agent

   Parameter	Description
   Agent icon	The system provides various preset icons for agents. Click on the icon to select a preferred style.
   Agent Name	Name of the custom agent to be created. Naming convention: 2 to 20 characters, consisting of letters, digits, underscores, and hyphens (-), and must start with a letter.
   Scope	Application scope of the custom agent. Project: The agent is only available within the current project and cannot be called by other projects. User: The agent can be called across all projects of the current user.
   Agent Type	Working mode of the agent, which affects its behavior logic and capability boundary. Primary/Sub-agent: It can be used as a sub-agent to be called by a primary agent or as a primary agent to call a sub-agent. Primary agent: It runs independently to process user requests and cannot be called by other agents. Sub-agent: It can be called by primary agents to perform specific tasks.
   Description	A brief explanation of what the agent does to help users understand its functionality, with a maximum of 5,000 characters.
   Prompt	System instructions that define the role, tone, workflow, and rules/specifications of the agent.
   Skills	Skills that are associated with the agent to enable it to perform certain specialized tasks. NOTE: CodeArts Agent has no hard limit on the number of associated skills, but it is recommended that the number not exceed 20 to ensure optimal user experience.
   MCP Tools	This configuration item is displayed only when Scope is set to Project. External tools or services connected to the agent to enable it to interact with the outside world, such as obtaining real-time information and operating software. Up to 80 MCP tools can be added. It is recommended that no more than eight MCP tools be enabled at the same time, with three being ideal for the optimal user experience.
   Built-in tools	Built-in, out-of-the-box tools provided by the agent. Read: used to retrieve and view files. Edit: used to add, modify, or delete files. Terminal: used to run system commands in the command line environment. Preview: used to preview the generated frontend page in real time. Webfetch: used to capture and read web page content.

6. Verify the custom agent.
   • After the custom agent is created, you can view it on the corresponding category tab (such as Project or User), and its status will be displayed as enabled.
   • In the input box on the chat page, click Agent. Verify that the target agent is displayed in the custom agent list.
     

     If you create a sub-agent, it will not be displayed in the custom agent list.

Creating Custom Agents Using Markdown Files

CodeArts Agent allows you to create local projects or user agents through Markdown files. When project and user agents share the same name, the project agent takes precedence over the user agent.

Using Custom Agents

1. At the bottom of the input box on the chat panel, select a custom agent (for example, Python-debugging) to switch to the custom agent mode. The selected model is displayed on the right. You can select a model as required from the drop-down list.
   

   If the chat panel of CodeArts Agent is not displayed, click (Expand AI Side Bar) in the upper right corner of the top menu bar to open it.

2. In the input box on the chat panel, enter a prompt (for example, a Python code block containing errors) and click or press Enter.
3. View the agent execution result.

   The agent automatically detects syntax or logic errors in the code block and provides accurate correction suggestions and the corrected code.

   Figure 3 Viewing the agent execution result
   

Modifying the Default Model for a Local Custom Agent

Once a custom agent is created, its model usage adheres to the following rules: The primary agent defaults to the model used by the system's built-in agent. For example, if the built-in system agent selects GLM-5.2 as its LLM, the custom agent will also use GLM-5.2. A sub-agent directly inherits the model used by the primary agent that invokes it. If you need to modify the default model used by a custom agent, you can do so by modifying the configuration file of the custom agent.

1. Go to the directory where the custom agent's configuration file is stored.

   For example, to modify a project custom agent, go to the ./.codeartsdoer/agents/ directory of the project on the local machine. You can also locate the corresponding directory in the explorer of CodeArts Agent IDE. For details about the directories for storing Markdown files of other types of custom agents, see Table 1.

2. Open the Markdown file of the custom agent, add the model field, and save the file (Windows/Linux: Ctrl+S; macOS: Command(⌘)+S).

   In CodeArts Agent, you can specify the default model for custom agents by configuring the model field. You can dynamically switch to the optimal model for specific tasks to balance performance and cost. For instance, employ a lightweight, high-speed model during the planning phase to boost response speed, and invoke a high-capability model during the implementation phase to ensure execution quality.

   The value of the model field is in the format of Provider/Model ID, where Provider indicates the model provider (set to inferhub-provider in CodeArts Agent), and Model ID indicates the ID of the specific model.

   

   If the configured model does not exist or you do not have the permission, the primary agent automatically switches to the default model of the built-in agent, and the sub-agent inherits the model of the primary agent.

   ---
   description: focusing on Python code error check and optimization, expected to be professional and concise.
   mode: primary
   model: inferhub-provider/GLM-5.2
   permission:
     skill:
       '*': deny
   disable: false
   scope: project
   avatar: avatar1
   ---

   Table 6 IDs of built-in models in CodeArts Agent

   Model Name	Model ID
   GLM-5.1	GLM-5.1
   GLM-5.2	GLM-5.2

3. Return to the settings page of CodeArts Agent IDE. On the Agents tab, click to refresh the agent configuration.
4. View the default model of the custom agent.

   In the input box on the chat panel, switch to the target custom agent. The system automatically applies the default model (for example, GLM-5.2) defined by the Markdown configuration file.

Managing Local Custom Agents

CodeArts Agent IDE features full-lifecycle management capabilities for custom agents. You can efficiently control and continuously optimize created agents based on actual business needs.