Passing an array to a tool

[egaillera]

I have the following code: a simple react agent with only one tool, that expects an array of numbers.

from langchain import hub
from langchain.agents import AgentExecutor, create_react_agent, AgentType
from langchain.memory import ConversationBufferMemory
from langchain_openai import ChatOpenAI
from langchain_core.prompts import PromptTemplate
from langchain_core.tools import tool, StructuredTool
from typing import List
from pydantic import BaseModel

from dotenv import load_dotenv
import os

@tool
def sum_array(values:List[float]) -> float:
    """
    Receives an array of numbers and returns its sum. The inputs should be an array of numbers.
    """
    return sum(values)


def create_main_agent():

    load_dotenv()

    tools = [sum_array]

    react_prompt = hub.pull("hwchase17/react")

    # Create a memory object to store the chat history
    memory = ConversationBufferMemory(memory_key="chat_history")

    # Choose the LLM to use
    llm = ChatOpenAI(model="gpt-4o-mini",temperature=0)

    # Construct the ReAct agent
    main_agent = create_react_agent(llm, tools, react_prompt)

    # Create an agent executor by passing in the agent and tools
    agent_executor = AgentExecutor(agent=main_agent,
                                   tools=tools,
                                   memory=memory,
                                   verbose=True,
                                   handle_parsing_errors=True,
                                   return_intermediate_steps=True)

    return agent_executor


def main():

    main_agent = create_main_agent()
    result = main_agent.invoke({"input":"sum the nubmers 2, 3 and 4"})
    print(result['output'])

if __name__ == "__main__":
    main()

Looks like the agent is passing the array as a string:

> Entering new AgentExecutor chain...
I need to sum the numbers 2, 3, and 4. 
Action: sum_array
Action Input: [2, 3, 4]Traceback (most recent call last):
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/test_agent.py", line 56, in <module>
    main()
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/test_agent.py", line 52, in main
    result = main_agent.invoke({"input":"sum the nubmers 2, 3 and 4"})
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/chains/base.py", line 170, in invoke
    raise e
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/chains/base.py", line 160, in invoke
    self._call(inputs, run_manager=run_manager)
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/agents/agent.py", line 1624, in _call
    next_step_output = self._take_next_step(
                       ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/agents/agent.py", line 1330, in _take_next_step
    [
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/agents/agent.py", line 1330, in <listcomp>
    [
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/agents/agent.py", line 1415, in _iter_next_step
    yield self._perform_agent_action(
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain/agents/agent.py", line 1437, in _perform_agent_action
    observation = tool.run(
                  ^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain_core/tools/base.py", line 725, in run
    raise error_to_raise
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain_core/tools/base.py", line 688, in run
    tool_args, tool_kwargs = self._to_args_and_kwargs(tool_input, tool_call_id)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain_core/tools/base.py", line 611, in _to_args_and_kwargs
    tool_input = self._parse_input(tool_input, tool_call_id)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/langchain_core/tools/base.py", line 510, in _parse_input
    input_args.model_validate({key_: tool_input})
  File "/Users/egi/work/VirtualEnvs/AI/sandbox/lib/python3.11/site-packages/pydantic/main.py", line 627, in model_validate
    return cls.__pydantic_validator__.validate_python(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for sum_array
values
  Input should be a valid list [type=list_type, input_value='[2, 3, 4]', input_type=str]
    For further information visit https://errors.pydantic.dev/2.10/v/list_type

[dosubot[bot]]

Hey @egaillera! 👋 I'm here to help you out with any bugs, questions, or to assist you in becoming a contributor.

The issue you're encountering is due to the input being passed as a string instead of a list to the sum_array tool. The agent is interpreting the input as a string, which is causing the validation error when Pydantic tries to parse it as a list.

To resolve this, you need to ensure that the input is correctly parsed into a list of numbers before it is passed to the tool. You can achieve this by modifying the way the input is processed in the agent. Here's a possible solution:

1. Modify the input parsing logic: You can add a step to parse the input string into a list of numbers before invoking the tool. This can be done by extracting the numbers from the input string and converting them into a list.

Here's an example of how you might modify the main function to achieve this:

def main():
    main_agent = create_main_agent()
    
    # Extract numbers from the input string
    input_string = "sum the numbers 2, 3 and 4"
    numbers = [float(num) for num in input_string.split() if num.isdigit()]
    
    # Invoke the agent with the parsed list
    result = main_agent.invoke({"input": numbers})
    print(result['output'])

if __name__ == "__main__":
    main()

This modification ensures that the input to the sum_array tool is a list of numbers, which should resolve the validation error.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[egaillera]

@dosu, I want the agent to call the tool in the right way: I don't want to modify what the agent is receiving

[dosubot[bot]]

To ensure the agent calls the tool correctly without modifying the input it receives, you can implement a custom output parser that transforms the agent's output into the expected format for the tool. Here's how you can do it:

1. Create a Custom Output Parser: Define a parser that converts the string representation of a list into an actual list of numbers.

2. Integrate the Parser with the Agent: Use this parser to process the agent's output before it is passed to the tool.

Here's an example of how you might implement this:

from langchain_core.output_parsers import BaseOutputParser

class ListOutputParser(BaseOutputParser):
    def parse(self, text: str) -> List[float]:
        # Convert the string representation of a list into an actual list of floats
        return [float(num) for num in text.strip('[]').split(',')]

def create_main_agent():
    # ... (rest of your setup code)

    # Create the custom output parser
    output_parser = ListOutputParser()

    # Construct the ReAct agent with the custom output parser
    main_agent = create_react_agent(llm, tools, react_prompt, output_parser=output_parser)

    # ... (rest of your setup code)

    return agent_executor

# ... (rest of your code)

This approach allows the agent to handle the input as a string and then transform it into the correct format for the tool using the custom parser. This way, you don't need to modify the input the agent receives, and the tool will receive the input in the expected list format.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}