Agent

⚠️ Model pricing and consumption information in this document (including screenshots) is subject to actual conditions; the help documentation is for reference only, and the interface and features are subject to the current product version.

Attribute	Value
Node name	Agent
Version	v1.0
Supported scenarios	Chatflow Agent ✓, Workflow Agent ✓
Test run support	✓

Quick Reference

The Agent node has no strategy selection; it always runs in Function Calling mode (the model uses tools via function calls, looping until the task is complete).

Feature Description

The Agent node is one of the most essential nodes on the platform. At its core, it is an LLM loop: within the loop, the LLM autonomously decides the next action until the task is complete, enabling it to handle highly complex functionality, including:

• Planning: decomposing tasks and deciding execution steps
• Answering questions: generating reply content
• Calling tools: using mounted tools to fetch information or perform actions
• Using Skills: invoking mounted Skills to complete specialized tasks
• Launching a sandbox: executing code in a sandbox environment
• Running scripts: executing scripts to process data or logic
• Running HITL: Human-in-the-Loop collaboration, bringing in human confirmation at key steps

It also supports:

• Workflow integration: calling existing Agent Tools, supporting multi-step conversations and business processing

Node Display Content

1. Basic Information Display

• Icon: node-specific icon
• Name: node name
• Model: displayed the same way as the LLM node
• Note: explanatory text added by the user

2. Action Buttons

• More button: click to display the context menu content
• Run button: click to enter the test run interface

Node Edit Panel

See the right side of the screenshot above for the edit panel interface.

1. Model Configuration

• Model selection: same as the LLM node's model configuration
• Parameter settings: model parameters can be configured
• Run mode: no strategy selection; defaults to Function Calling mode

2. Tool List Management

• Tool counts: displays the number of enabled and added tools (enabled count/added count)
• Add tool: an add-tool button; after clicking, multiple tools can be selected from the tool list
• Tool display: displays the list of added tools
  • Shows the tool logo, provider, and tool name
  • Each tool supports on/off toggle configuration
  • Hovering over a tool reveals delete and settings actions
  • All tools in this release are "authorized"; the backend configuration is set up in advance

a. Tool Selection

Click the add button to open the Tool Settings dialog, which supports adding a single tool or adding multiple, searching for tools, and switching between the Tools and OpenAPI tabs. Tools are grouped by category (such as News, Real-time Quotes, etc.):

b. MCP Connectors (Third-Party MCP Tools)

Third-party MCP tools can be connected. Currently, Financial Datasets (real-time stock prices, financial data, SEC filings, and market news) is available:

⚠️ This feature is only available on the Premium plan.

c. Tool Search

• When Tool Search is enabled, in-plan tools not explicitly checked are retrieved on demand via the tool_search meta-tool and loaded at the end of the context
• Effect: significantly reduces prompt size and improves tool selection accuracy
• Suitable for scenarios with many tools—instead of stuffing all tools into the context, let the model fetch them on demand

3. Skills

Skills can be mounted onto an Agent: click the add button to open the Select Skill dialog. The list shows official Skills (such as strategy-us-power-nuclear, financial-report-analysis, etc.), and you can also create a Skill directly from here:

4. Instruction Configuration

• Configuration item: same as the LLM node's system_prompt configuration
• Limitation: cannot be deleted

5. Query Configuration

• Configuration item: same as the LLM node's User Prompt / Assistant Prompt
• Limitation: cannot be deleted

6. Feature Switches

The Agent node provides a set of capability switches; enable them as needed:

a. User Tags

• When enabled, the LLM reads the current user's profile tags and preference tags in scenarios requiring personalized answers—for example, investment goals, risk appetite, analysis style, and other personalized interests/disinterests

b. Knowledge Base

• When Knowledge Base is enabled, the AI will search the selected knowledge bases when necessary

c. Human-in-the-Loop (HITL)

• Allows the model to ask the user questions and wait for answers during execution
• This feature is very powerful; enabling it is recommended

d. Web Search

• When Web Search is enabled, the AI will use the built-in web search tool when necessary
• Strongly recommended to enable

e. To-Dos

• When To-Dos is enabled, the AI automatically identifies tasks in the conversation and decomposes and executes them
• Strongly recommended to enable

f. Output Legality and Compliance Check

• When enabled, the system performs a compliance check on the content as it is output
• If the output content is found to be illegal or non-compliant with financial regulations, the system will immediately stop the output and clear the content already generated

7. Maximum Iterations

• Range: configurable from 1 to 99
• Purpose: controls the maximum number of Agent executions
• Recommended value: generally set it to 99 to give the Agent enough loop headroom to complete complex tasks

8. Output Variables

• text: the content generated by the agent (text)
• think: the model's reasoning process. Some models support think output, which can be used with the "Reasoning Process Output" node output (Chatflow only)
• JSON generated by the agent: the structured output result

9. Next Step Configuration

• Adding a next node is supported

Execution Logic

The execution logic of the Agent node is implemented by the backend to ensure the Agent executes and calls tools correctly.

Context Menu Actions

The context menu contains the following options:

1. Change node: change the node type
2. Copy: copy the node
3. Duplicate: duplicate the node content
4. Delete: delete the node
5. Help link: jump to the help documentation
6. Run node: test-run the current node

Test Run

• Support status: test run supported
• Change node: changing the node type is supported
• Copy features: copy, cut, and delete operations are supported
• Help features: help is supported; About is not supported

Run Mode

The Agent node uniformly runs in Function Calling mode, with no strategy selection required:

• Characteristics: the model uses predefined tools via function calls, autonomously planning, calling, and evaluating within a loop until the task is complete
• Advantages: precise control and strong predictability, while retaining the ability to handle complex multi-step tasks

Tool Management

1. Tool Types

• API tools: call external API endpoints
• Calculation tools: perform mathematical calculations
• Data processing tools: process and analyze data
• Custom tools: user-defined tools

2. Tool Configuration

• Authorization management: all tools in this release are already authorized
• Toggle control: specific tools can be enabled or disabled
• Parameter settings: tool parameters can be configured

3. Tool Usage

• Automatic selection: the Agent automatically selects appropriate tools based on the task
• Manual configuration: the tools to use can be specified manually
• Tool chains: combining multiple tools is supported

Use Cases

1. Intelligent Assistants

• Multi-turn conversations
• Task execution
• Question answering

2. Automated Processing

• Data collection
• Information organization
• Report generation

3. Decision Support

• Data analysis
• Risk assessment
• Recommendation of options

4. Workflow Integration

• Calling external services
• Executing complex logic
• Multi-step processing

Best Practices

1. Tool Configuration

• Tool selection: choose tools that match the task
• Permission management: set tool permissions appropriately
• Performance optimization: avoid using too many tools

2. Iteration Control

• Count settings: set the iteration count based on task complexity
• Timeout handling: set reasonable timeouts
• Error handling: configure an error handling strategy

Configuration Steps

1. Basic Configuration

1. Configure the large language model
2. Set the instruction and query configuration

2. Tool Configuration

1. Add the required tools
2. Configure tool parameters
3. Set the tool toggle states

3. Advanced Configuration

1. Set the maximum iterations
2. Configure output variables
3. Set the error handling strategy

4. Testing and Validation

1. Test using the test run feature
2. Check the Agent's execution results
3. Adjust configuration parameters

Considerations

1. Tool management: manage the number of tools and their permissions sensibly
2. Iteration control: avoid infinite loops; set a reasonable iteration count
3. Performance considerations: consider the performance and cost of Agent execution
4. Error handling: configure appropriate error handling mechanisms

FAQ

Q: Does the Agent node require selecting a run strategy?

A: No. The Agent node has no strategy selection; it always uses Function Calling mode.

Q: How do I add tools?

A: Click the add tool button and select the tools you need from the tool list.

Q: How should I set the maximum iterations?

A: The range is 1-99; generally 99 is recommended, giving the Agent enough loop headroom to complete complex tasks.

Q: How does the Agent choose tools?

A: The Agent automatically selects appropriate tools based on the task requirements and tool capabilities.

Q: How do I optimize Agent performance?

A: Configure tools sensibly (enable Tool Search when there are many tools) and set an appropriate iteration count.