Optional schema for tool calling agent

[yos1p]

Hi, I have been using LangChain AgentExecutor for a while and I noticed that when LLM generated not all of the schema, the tool call will fail.

For example, I have a tool / function with 3 variables (a, b, c) and the c variable to be optional. If the LLM generated both function argument with a, b, c, then the tool call works. But if the argument is only a, b, the tool call doesn't work.

Is there a way for us to let LangChain AgentExecutor to execute the tool even though c was not provided?

[dosubot[bot]]

Hey @yos1p, I'm here to help you out with any bugs, questions, or contributions you have regarding LangChain AgentExecutor. Let's tackle this issue together!

I found a similar issue titled "AgentExecutor fails to use StructuredTool" which might be related to your problem. This issue is closed and was last updated on October 27, 2024. You can check it out for any insights or solutions that might apply to your situation here ^[1].

To continue talking to Dosu, mention @dosu.

[feijoes]

@yos1p Are you using StructuredTool? You can pass the args_schema and create the schema parameters yourself if is not working. tool docs

[yos1p]

Yes, I used the args_schema already. But what I noticed is that even after I mark the variable as optional, it still doesn't work.

Does LangChain recognise "Optional" mark in the schema? I don't think so.

[feijoes]

@yos1p Maybe it's a dependency issue? If you can share the code for how you're implementing the tool, I can help.

[yos1p]

@dosu Does StructuredTool schema allows Optional variable?

[dosubot[bot]]

Could you provide more details about how you're defining the args_schema in your StructuredTool? Specifically, how are you marking the variable as optional in your schema? This information will help in understanding the issue better.

To continue talking to Dosu, mention @dosu.

[yos1p]

@dosu

For example:

from pydantic import BaseModel
from typing import Optional

class PersonSchema(BaseModel):
    name: str
    birth_year: Optional[int] = None

So, that if I only give parameter name but not birth_year, Langchain should still call the function.

[dosubot[bot]]

Could you share how you're integrating this PersonSchema with the StructuredTool in LangChain? Specifically, how are you using this schema within the tool's implementation? This will help in understanding how the optional parameter is being handled.

To continue talking to Dosu, mention @dosu.