Tool Introduction

This document introduces the usage and development process of the Tool class. Please refer to the Configuration for detailed parameter configuration.

1. Tool Usage

1.1. Direct External Call

Tools uniformly utilize the .call(params) interface for calling, where you pass in the necessary parameters for the tool. The tool then returns the results after execution, with the return type being Union[str, list, dict].

For example, to directly call a image generation tool:


from qwen_agent.tools import ImageGen
 
tool = ImageGen()
res = tool.call(params = {'prompt': 'a cute cat'})
print(res)

1.2. Internal call by Agent

In the Agent, the _call_tool(...) function is used to call tools, with each Agent instance capable of calling the tools that were initialized and assigned to it. The tools can be passed in through function_list: Optional[List[Union[str, Dict, BaseTool]]] = None parameter. The supported input types include the tool name, tool configuration file, or the tool object itself. For instance, you can use code_interpreter, {'name': 'code_interpreter', 'timeout': 10}, or CodeInterpreter().

Note that for the convenience of inputting tool results into LLMs, the Agent’s _call_tool(...) interface converts all returned results from the tools into string type. For more details, see the Agent class.

2. Tool Development

Qwen-Agent provides a mechanism for registering tools. For example, to register your own image generation tool:

Specify the tool’s name, description, and parameters. Note that the string passed to @register_tool('my_image_gen') is automatically added as the .name attribute of the class and will serve as the unique identifier for the tool.
Implement the call(...) function.


import urllib.parse
import json5
import json
from qwen_agent.tools.base import BaseTool, register_tool
# Add a custom tool named my_image_gen：
@register_tool('my_image_gen')
class MyImageGen(BaseTool):
    description = 'AI painting (image generation) service, input text description, and return the image URL drawn based on text information.'
    parameters = {
        'type': 'object',
        'properties': {
            'prompt': {
                'description':
                    'Detailed description of the desired content of the generated image, such as details of characters, environment, actions, etc., in English.',
                'type':
                    'string',
            }
        },
        'required': ['prompt'],
    }
 
    def call(self, params: str, **kwargs) -> str:
        prompt = json5.loads(params)['prompt']
        prompt = urllib.parse.quote(prompt)
        return json.dumps(
            {'image_url': f'https://image.pollinations.ai/prompt/{prompt}'},
            ensure_ascii=False)

Once the tools are registered, they can be used as mentioned above.

If you prefer not to use the registration method, you can also directly define a tool class and then pass the tool object to the Agent (unregistered tools do not support passing the tool name or configuration file).


import urllib.parse
import json5
import json
from qwen_agent.tools.base import BaseTool
 
class MyImageGen(BaseTool):
    name = 'my_image_gen'
    description = 'AI painting (image generation) service, input text description, and return the image URL drawn based on text information.'
    parameters = {
        'type': 'object',
        'properties': {
            'prompt': {
                'description':
                    'Detailed description of the desired content of the generated image, such as details of characters, environment, actions, etc., in English.',
                'type':
                    'string',
            }
        },
        'required': ['prompt'],
    }
 
    def call(self, params: str, **kwargs) -> str:
        prompt = json5.loads(params)['prompt']
        prompt = urllib.parse.quote(prompt)
        return json.dumps(
            {'image_url': f'https://image.pollinations.ai/prompt/{prompt}'},
            ensure_ascii=False)