Custom agent tutorial doesnt handle replacing SerpAPI with google search tool

[paulbatum]

Start with the following tutorial: https://python.langchain.com/en/latest/modules/agents/agents/custom_llm_agent.html

But instead of using SerpAPI, use the google search tool:

from langchain.agents import load_tools
tools = load_tools(["google-search"])

The step that creates the CustomPromptTemplate will encounter a validation error:

ValidationError                           Traceback (most recent call last)
Cell In[36], line 24
     21         kwargs["tool_names"] = ", ".join([tool.name for tool in self.tools])
     22         return self.template.format(**kwargs)
---> 24 prompt = CustomPromptTemplate(
     25     template=template,
     26     tools=tools,
     27     # This omits the `agent_scratchpad`, `tools`, and `tool_names` variables because those are generated dynamically
     28     # This includes the `intermediate_steps` variable because that is needed
     29     input_variables=["input", "intermediate_steps"]
     30 )

File [/site-packages/pydantic/main.py:341]/.venv-jupyter/lib/python3.10/site-packages/pydantic/main.py:341), in pydantic.main.BaseModel.__init__()

ValidationError: 1 validation error for CustomPromptTemplate
tools -> 0
  Tool.__init__() missing 1 required positional argument: 'func' (type=type_error)

The problem appears to be that the result of calling load_tools(["google-search"]) is a BaseTool and not a Tool and doesn't have a func. This can be fixed by modifying the CustomPromptTemplate to use BaseTool instead of Tool.

from langchain.tools import BaseTool

class CustomPromptTemplate(StringPromptTemplate):
    # The template to use
    template: str
    # The list of tools available
    tools: List[BaseTool]
    
    def format(self, **kwargs) -> str:
<SNIP>

However I am not sure if this is the correct fix, or if the problem is that load_tools should create a Tool instead of a BaseTool. i.e. is this a doc issue or a product issue?

[tlazaro11]

I'm having the same dilemma/issue as well :)

[DeadBranches]

To substitute SerpAPI with GoogleSearch we also need to ensure the correct python classes are loaded.

So, instead of using the load_tools() function like we all initially tried (and as paulbatum helpfully demonstrated), let's try building a list of tools in a different way.

The following code works for me.

from langchain import OpenAI, GoogleSearchAPIWrapper, LLMChain
import os

os.environ["GOOGLE_CSE_ID"] = "6832f56b74e622d9f"
os.environ["GOOGLE_API_KEY"] = "RfcvhiFuS4W4nbRpAWn7KdHZfdMA53mzbdS-lDC"

search = GoogleSearchAPIWrapper()
tools = [
    Tool(
        name="Search",
        func=search.run,
        description="useful for when you need to answer questions about current events",
    )
]

Here's what's different:

1. Add GoogleSearchAPIWrapper to the python import statement for the langchain module.
2. Add your Google Search API credentials to the script. In the above example environmental variables are used to store the credentials because the class looks for them there. Alternatively, you could directly pass your credentials as arguments to GoogleSearchAPIWrapper().
3. Create an instance of the GoogleSearchAPIWrapper class and store it in the variable search.
4. Build your list of tools, which is literally a list of Tool objects. Our instance of SearchAPIWrapper is passed to the func parameter of Tool, but only in part. By that I mean just the run method of the search object is passed as an argument to Tool.

[tlazaro11]

How about for human tool? I don't see a wrapper for that

[DeadBranches]

@tlazaro11 Sorry, I don't understand your question. Could you please provide an excerpt of the code and error message or part of documentation you're having difficulty with?

[paulbatum]

@tlazaro11 I'm not clear on what you're trying to do but you can also try the fix I outlined of changing your CustomPromptTemplate to use BaseTool

[dosubot[bot]]

Hi, @paulbatum! I'm Dosu, and I'm here to help the LangChain team manage their backlog. I wanted to let you know that we are marking this issue as stale.

From what I understand, the issue is related to the Custom agent tutorial not handling the replacement of SerpAPI with the Google search tool correctly. There have been some suggestions and discussions in the comments about potential fixes, including modifying the CustomPromptTemplate to use BaseTool instead of Tool. It seems like there was a resolution to this issue by making that modification.

Before we close this issue, we wanted to check with you if it is still relevant to the latest version of the LangChain repository. If it is, please let us know by commenting on the issue. Otherwise, feel free to close the issue yourself, or it will be automatically closed in 7 days.

Thank you for your understanding and contribution to the LangChain project! Let us know if you have any further questions or concerns.

[sudamerushabh]

Hi. I am getting the same error.

my tool is as following:

def listBeneficiaries():
    """Tool to list all the beneficiaries."""
    response = requests.get('https://test-flentas.free.beeceptor.com/beneficiaries')
    return json.dumps(response.json())


class ListBeneficiaries(BaseTool):
    name = "list_beneficiaries"
    description = """
use this tool when you want list all the beneficiaries of the customer. output of this tool is json string
        """
    # args_schema: Type[BaseModel] = StockPercentChangeInput

    def _run(self,):
        response = listBeneficiaries()
        return response

    def _arun(self, ticker: str):
        raise NotImplementedError("This tool does not support async")
    
tools = [ListBeneficiaries()]

then I am writing custom agent prompt and all.

After that I am creating the PromptTemplate:

prompt = CustomPromptTemplate(
    template=template,
    tools_getter=tools,
    # This omits the `agent_scratchpad`, `tools`, and `tool_names` variables because those are generated dynamically
    # This includes the `intermediate_steps` variable because that is needed
    input_variables=["input", "intermediate_steps"],
)

Error I am getting after running this code:

ValidationError: 1 validation error for CustomPromptTemplate tools_getter -> 0 __init__() missing 1 required positional argument: 'func' (type=type_error)

[pvtoan]

Hi,

I got exactly the same problem as @sudamerushabh .

Appreciate any help!

[manilokesh]

refer https://towardsdatascience.com/building-a-math-application-with-langchain-agents-23919d09a4d3