tuple or namedtuple as a tool parameter type generates invalid schema

[markns]

Describe the bug

Using tuple or namedtuple as a function_tool parameter type

LatLong = namedtuple('LatLong', "lat long")

@function_tool
def get_weather(lat_long: LatLong) -> Weather:

@function_tool
def get_weather(lat_long: tuple[float, float]) -> Weather:

leads to the following error during function schema creation:

Error getting response: Error code: 400 - 
{'error': {'message': "Invalid schema for function 'get_weather': In context=(), 'minItems' is not permitted.", 
'type': 'invalid_request_error', 'param': 'tools[0].parameters', 'code': 'invalid_function_parameters'}}. 

Debug information

• Agents SDK version: v0.0.6
• Python version = 3.13

Repro steps

import asyncio
from collections import namedtuple
from pydantic import BaseModel

from agents import Agent, Runner, function_tool

class Weather(BaseModel):
    lat_long: tuple[float, float]
    temperature_range: str
    conditions: str

LatLong = namedtuple('LatLong', "lat long")

@function_tool
def get_weather(lat_long: LatLong) -> Weather:
    return Weather(lat_long=lat_long, temperature_range="14-20C", conditions="Sunny with wind.")

agent = Agent(
    name="Hello world",
    instructions="You are a helpful agent.",
    tools=[get_weather],
)

async def main():
    result = await Runner.run(agent, input="What's the weather in Tokyo?")
    print(result.final_output)
    # The weather in Tokyo is sunny.

await main()

Expected behavior

Using a tuple or namedtuple as type annotation should work.

I've previously used this function to normalize a schema generated by LLMEasyTools, but not sure how to make that work with the agents sdk yet.

def normalize_for_openai(obj: dict):
    if isinstance(obj, dict):
        # Create a new dict to avoid mutating while iterating
        new_obj = {}
        for k, v in obj.items():
            if k in ("minItems", "maxItems"):
                continue
            elif k == "prefixItems":
                # Assumes all prefixItems are the same type
                first_type = v[0].get("type", "number")
                new_obj["items"] = {"type": first_type}
            else:
                new_obj[k] = normalize_for_openai(v)
        return new_obj
    elif isinstance(obj, list):
        return [normalize_for_openai(i) for i in obj]
    else:
        return obj

[rm-openai]

Looking into it, thanks!

[rm-openai]

Looks like the OpenAI API doesn't support prefixItems, which was introduced in JSON Schema Draft 2020-12. And Pydantic (which we use under the hood for tool schema generation) turns tuple[float, float] into prefixItems.

I can ask internally if we can fix this. Until then, unfortunately your best bet is to use 2 floats args, an object, a list, or some other type.

[rm-openai]

Reported this internally. Hopefully we support this in the API soon. Closing this out - feel free to create a new issue/reopen if needed!