DOC: Custom Tool Error Handling is not Clear

[Hsgngr]

Issue with current documentation:

I cannot handle my custom_tool's raised errors.

I have this weatherTool:

class GetCurrentWeatherTool(BaseTool):
    name = "get_current_weather"
    description = "Get the current weather for a specified location"
    args_schema: Type[WeatherInput] = WeatherInput

    def _run(
        self, 
        location: str,
        run_manager: Optional[CallbackManagerForToolRun] = None
    ) -> str:
        """Use the tool."""
        return get_current_weather(location)

    async def _arun(
        self, 
        location: str,
        run_manager: Optional[AsyncCallbackManagerForToolRun] = None
    ) -> str:
        """Use the tool asynchronously."""
        # You can implement the async version of the method here or raise a NotImplementedError if async is not supported
        raise NotImplementedError("get_current_weather does not support async")

    def _handle_error(error: ToolException) -> str:
        return (
            "The following errors occurred during tool execution:"
            + error.args[0]
            + "Location couldn't be found, please try another tool."
        )

And I raise ToolException when the API call gets an error:
raise ToolException("location has been found.") inside of get_current_weather

Now I am adding this tool to my tools when I'm creating my agent:

self.weather = GetCurrentWeatherTool()
            Tool(
                 name="Weather",
                 func=self.weather.run,
                 description="useful for weather related questions",
                 return_direct=True,
             )

In the documentation Custom Tools it explains how to add it and gives an example:

    Tool.from_function(
        func=search_tool1,
        name="Search_tool1",
        description=description,
        handle_tool_error=True,
    ),

However I couldnt find the equal of that for Subclassing the BaseTool.

Idea or request for content:

It would be great to give an example for Subclassed CustomTool(BaseTool) when defining the tools.

[dosubot]

🤖

Hello,

Thank you for bringing this to our attention. I understand that you're having trouble handling errors for your custom tool that subclasses BaseTool. The handle_tool_error attribute is indeed used to handle exceptions that occur during the execution of a tool. It can be set to a boolean, a string, or a callable function.

Here's an example of how you can handle errors in your custom tool:

from langchain.tools.base import BaseTool, ToolException

class GetCurrentWeatherTool(BaseTool):
    name = "get_current_weather"
    description = "Get the current weather for a specified location"
    args_schema: Type[WeatherInput] = WeatherInput
    handle_tool_error = True  # or a string, or a callable

    def _run(
        self, 
        location: str,
        run_manager: Optional[CallbackManagerForToolRun] = None
    ) -> str:
        """Use the tool."""
        try:
            return get_current_weather(location)
        except Exception as e:
            # If an error occurs, raise a ToolException
            raise ToolException("An error occurred in GetCurrentWeatherTool") from e

    async def _arun(
        self, 
        location: str,
        run_manager: Optional[AsyncCallbackManagerForToolRun] = None
    ) -> str:
        """Use the tool asynchronously."""
        # You can implement the async version of the method here or raise a NotImplementedError if async is not supported
        raise NotImplementedError("get_current_weather does not support async")

In the run and arun methods of BaseTool, there is a try-except block that catches ToolException. If handle_tool_error is False, the exception is re-raised. If handle_tool_error is True, the first argument of the exception is used as the observation. If handle_tool_error is a string, that string is used as the observation. If handle_tool_error is a callable, it is called with the exception as an argument, and its return value is used as the observation.

If handle_tool_error is a callable, it should take one argument (the exception) and return the observation. Here's an example:

class GetCurrentWeatherTool(BaseTool):
    handle_tool_error = lambda e: f"An error occurred: {e}"
    # ... other methods and properties ...

In this example, if a ToolException is raised during the execution of the tool, the observation will be a string that includes the message of the exception.

I hope this helps! If you have any other questions, feel free to ask.

Best,
Dosu

Sources

• Custom Agent Doesn't Accept Custom Tool
• libs/langchain/langchain/tools/base.py
• libs/langchain/langchain/tools/graphql/tool.py
• libs/langchain/langchain/tools/file_management/file_search.py
• libs/langchain/tests/unit_tests/tools/test_base.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[ahmadaii]

@Hsgngr How did you implemented this in pydantic based tool? I have similar question and till curious how to do that.

[greeshmaPremarajNathan]

I have added
except Exception as e:
# If an error occurs, raise a ToolException
raise ToolException("An error occurred") from e
except ValidationError as e:
# If an error occurs, raise a ToolException
raise ToolException("An error occurred") from e
and also put
handle_tool_error=True
but getting
File "/home/greeshma/anaconda3/envs/py3_10/lib/python3.10/site-packages/langchain_core/tools.py", line 262, in _parse_input
result = input_args.parse_obj(tool_input)
File "pydantic/main.py", line 526, in pydantic.main.BaseModel.parse_obj
File "pydantic/main.py", line 342, in pydantic.main.BaseModel.init
pydantic.error_wrappers.ValidationError: 4 validation errors for CreateCalenderEvent
start_date_time -> 1
invalid datetime format (type=value_error.datetime)
start_date_time -> 2
invalid datetime format (type=value_error.datetime)
end_date_time -> 1
invalid datetime format (type=value_error.datetime)
end_date_time -> 2
invalid datetime format (type=value_error.datetime)

[Abinandan314]

@Hsgngr Did you find any solution. If so, can you please post it here!