What Is Claw Mode?

Unduh

Mode fokus

Ukuran font

Terakhir diperbarui: 2026-10-20 16:10:25

Diterjemahkan & Diperiksa oleh AI

Claw mode is one of the application categories supported by the agent development platform. In a Claw-mode application, natural language drives the agent to think autonomously and solve complex problems. The application has an independent workspace and can autonomously write and run code, as well as invoke various capabilities such as Skills / tools / connectors. It is suitable for complex task scenarios such as data analysis, content generation, and automated processing.
Note:
Mode switching is not supported for Claw mode. Because the workspace, Skills, connectors, and other configurations in Claw mode differ significantly from those in the other three modes, applications created in Claw mode can only be edited in Claw mode. To use capabilities from other modes, create a new application. Applications created in the other three modes also cannot be switched to Claw mode.
Key Concepts
Concept
Definition
Model
A large model responsible for thinking, task planning, and tool invocation. It supports autonomous switching, and each sub-Agent can independently select a different model.
Prompt
System instructions used to constrain the task execution process and response methods of Agents. It supports one-click AI generation and optimization, and the platform also provides a preset template library.
Skills
Skills is a feature module for extending professional capabilities of agent applications. By installing and using Skills, agents can gain professional processing capabilities in more practical scenarios, such as document translation, Image Analysis, and TTS. A single Agent can add up to 80 Skills.
Connectors and Tools
Connectors and tools are used to quickly extend the capability boundaries of Agents. A single Agent can add up to 1,000 tools (including connectors), and connectors are counted toward the quantity limit as tools.
Workspace
A dedicated independent execution environment for Claw mode applications. It supports autonomous code writing and execution, as well as file read and write operations, making it suitable for long-running and complex autonomous tasks.
Main Agent
The main Agent is the core scheduler and the entry point for user conversations in Claw mode applications. After a Claw mode application is created, the system creates a main Agent by default, and its name remains consistent with the application name.
Sub-Agent
A sub-Agent is a preconfigurable Agent entity within an application, with an independent name, description, prompt, Skills, tools, and connectors. A single application can have up to 20 main Agents and sub-Agents in total.
Intelligent Assistance Settings
A conversational configuration panel on the right side of the development page. It supports describing requirements in natural language, with AI assisting in optimizing application configurations and continuous iterative improvement.
Advanced Settings
Located at the bottom of the left sidebar on the development page. It includes an "Allow dynamic configuration modification in conversations" switch. After you enable this switch, callers can use the API to dynamically adjust Agent configurations at runtime by user or by session, including changing models, installing and uninstalling Skills, and binding plugin tools, without modifying the application configuration or republishing. Customers can build the same intelligent workbench based on this capability.
Core Workflow
The development process in Claw mode consists of the following stages:
1. Create an application: Select Claw mode on the application development page to create one, or jump from the intelligent workbench after a conversation is completed.
2. Configure the application: Configure the model, prompts, Skills, tools and connectors, workflows, and more in the left panel. You can also use the intelligent assistance settings panel on the right to generate or optimize configurations through conversations.
3. Debug the application: Start a test conversation in the debugging area on the right to verify the feature. Model comparison debugging is also supported.
4. Publish the application: After debugging meets expectations, click Publish to launch the application and obtain API call credentials, or go to the publishing channels to configure distribution entries such as WeCom and WeChat.
Differences from Other Modes
Dimension
Claw Mode
Multi-Agent Mode
Standard Mode
Single-Workflow Mode
Core Positioning
Autonomous thinking and code execution. Has an independent workspace and supports autonomous code writing and execution, as well as calling Skills and tools.
Autonomous planning and multi-agent collaboration.Grants the model a high degree of initiative to autonomously plan paths and flexibly invoke tools.
Dual-model collaboration and high-precision Q&A. Uses preset workflows with "thinking + generation" dual-model collaboration and supports workflow stacking.
Standardized response through fixed processes. Binds a specific workflow to respond to all conversations, ensuring that processes are absolutely standardized.
Scenario
Suitable for complex tasks such as data analysis and automated processing.
Suitable for complex scenarios that require multi-agent collaboration and elastic response.
Suitable for serious Q&A scenarios such as enterprise knowledge services and product consulting.
Suitable for enterprise services with clear business objectives and unified operational processes.
Multi-Agent Support
Supported (20 in total, including main + sub-agents)
Supported (main + sub-agents, 10 in total)
Not supported
Not supported
Independent Workspace
Yes (code execution and file read/write)
None
None
None
Skills Support
Yes (up to 80)
None
None
None
Mode Switching
Switching to other modes not supported.
Supports switching to standard mode and single-workflow mode.
Supports switching between each other.
Supports switching between each other.
Multi-Agent Collaboration
Claw mode supports two multi-agent collaboration methods. For details, see Multi-Agent Collaboration.
Sub-Agent collaboration (one-way delegation): The main Agent serves as the sole decision-making center, automatically routing or decomposing tasks based on user intent and dispatching subtasks to sub-Agents for execution. After a sub-Agent completes its work, it returns the result to the main Agent, which then synthesizes the final response. This method is adopted by default when a Claw application is created.
Multi-Agent team collaboration (bidirectional collaboration): Building on sub-Agent collaboration, this mode extends bidirectional communication between the main Agent and sub-Agents, and sub-Agents can also collaborate directly with each other. It is enabled after you turn on the Allow Multi-Agent Team Collaboration switch in the sub-Agent area of the left panel. Once enabled, the system automatically selects one-way delegation or bidirectional collaboration at runtime based on task complexity, without requiring manual switching.
Choosing a Creation Method
Claw mode provides two entry points to the development page:
Creation Method
Scenario
Operation Path
Create in App Development Page
Create a Claw app directly and configure it from scratch.
App development → Create app → Select Claw mode → Create
Create from Smart Workbench
First explore through conversations in the workbench, and then convert it into a Claw app with one click.
Smart workbench → Describe requirements in conversation → Complete the card "Go to App Development".
Application Example
Creating a Data Analysis Assistant Based on Claw Mode
It is suitable for autonomous task scenarios such as data analysis and content generation that require code execution and file read/write. For the specific setup process, see Creating a "Data Analysis Assistant" Based on "Claw Mode".
Examples of Other Application Modes
To learn about creating applications in other modes, see:
Create a "Marketing Copy Generation Assistant" Based on "Standard Mode"
Create a "Web Content Summarization Assistant" Based on "Single-Workflow Mode"
Create a "Talk Show Material Creation Assistant" Based on "Multi-Agent Mode"
FAQs
Does Claw Mode Support Mode Switching?
Not supported. Because the workspace, Skills, connectors, and other configurations in Claw mode differ significantly from those in the other three modes, applications created in Claw mode cannot be switched to other modes. Conversely, applications created in the other three modes also cannot be switched to Claw mode. To use capabilities from other modes, create a new application.
Does Claw Mode Support Workflows?
Yes. Claw mode supports adding workflows as tools in the tool configuration area, allowing Agents to call them on demand at runtime. The Workflow Management Tab provides an entry point for creating and managing workflows.

Bantuan dan Dukungan

Apakah halaman ini membantu?

Anda juga dapat Menghubungi Penjualan atau Mengirimkan Tiket untuk meminta bantuan.

masukan

tencent cloud

Tencent Cloud Agent Development Platform

What Is Claw Mode?

Key Concepts

Core Workflow

Differences from Other Modes

Multi-Agent Collaboration

Choosing a Creation Method

Application Example

Creating a Data Analysis Assistant Based on Claw Mode

Examples of Other Application Modes

FAQs

Does Claw Mode Support Mode Switching?

Does Claw Mode Support Workflows?

Bantuan dan Dukungan

Concept	Definition
Model	A large model responsible for thinking, task planning, and tool invocation. It supports autonomous switching, and each sub-Agent can independently select a different model.
Prompt	System instructions used to constrain the task execution process and response methods of Agents. It supports one-click AI generation and optimization, and the platform also provides a preset template library.
Skills	Skills is a feature module for extending professional capabilities of agent applications. By installing and using Skills, agents can gain professional processing capabilities in more practical scenarios, such as document translation, Image Analysis, and TTS. A single Agent can add up to 80 Skills.
Connectors and Tools	Connectors and tools are used to quickly extend the capability boundaries of Agents. A single Agent can add up to 1,000 tools (including connectors), and connectors are counted toward the quantity limit as tools.
Workspace	A dedicated independent execution environment for Claw mode applications. It supports autonomous code writing and execution, as well as file read and write operations, making it suitable for long-running and complex autonomous tasks.
Main Agent	The main Agent is the core scheduler and the entry point for user conversations in Claw mode applications. After a Claw mode application is created, the system creates a main Agent by default, and its name remains consistent with the application name.
Sub-Agent	A sub-Agent is a preconfigurable Agent entity within an application, with an independent name, description, prompt, Skills, tools, and connectors. A single application can have up to 20 main Agents and sub-Agents in total.
Intelligent Assistance Settings	A conversational configuration panel on the right side of the development page. It supports describing requirements in natural language, with AI assisting in optimizing application configurations and continuous iterative improvement.
Advanced Settings	Located at the bottom of the left sidebar on the development page. It includes an "Allow dynamic configuration modification in conversations" switch. After you enable this switch, callers can use the API to dynamically adjust Agent configurations at runtime by user or by session, including changing models, installing and uninstalling Skills, and binding plugin tools, without modifying the application configuration or republishing. Customers can build the same intelligent workbench based on this capability.

Dimension	Claw Mode	Multi-Agent Mode	Standard Mode	Single-Workflow Mode
Core Positioning	Autonomous thinking and code execution. Has an independent workspace and supports autonomous code writing and execution, as well as calling Skills and tools.	Autonomous planning and multi-agent collaboration.Grants the model a high degree of initiative to autonomously plan paths and flexibly invoke tools.	Dual-model collaboration and high-precision Q&A. Uses preset workflows with "thinking + generation" dual-model collaboration and supports workflow stacking.	Standardized response through fixed processes. Binds a specific workflow to respond to all conversations, ensuring that processes are absolutely standardized.
Scenario	Suitable for complex tasks such as data analysis and automated processing.	Suitable for complex scenarios that require multi-agent collaboration and elastic response.	Suitable for serious Q&A scenarios such as enterprise knowledge services and product consulting.	Suitable for enterprise services with clear business objectives and unified operational processes.
Multi-Agent Support	Supported (20 in total, including main + sub-agents)	Supported (main + sub-agents, 10 in total)	Not supported	Not supported
Independent Workspace	Yes (code execution and file read/write)	None	None	None
Skills Support	Yes (up to 80)	None	None	None
Mode Switching	Switching to other modes not supported.	Supports switching to standard mode and single-workflow mode.	Supports switching between each other.	Supports switching between each other.

Creation Method	Scenario	Operation Path
Create in App Development Page	Create a Claw app directly and configure it from scratch.	App development → Create app → Select Claw mode → Create
Create from Smart Workbench	First explore through conversations in the workbench, and then convert it into a Claw app with one click.	Smart workbench → Describe requirements in conversation → Complete the card "Go to App Development".