Python: [Bug]: Using response_format with AzureAIClient failed

[damienaicheh]

Description

I noticed that using response_format with AzureAIClient failed compared to AzureAIAgentClient. I recreated a repro sample.

Code Sample

import asyncio
import os
from random import randint
from typing import Annotated

from agent_framework_azure_ai import AzureAIClient
from azure.identity.aio import AzureCliCredential
from dotenv import load_dotenv
from pydantic import BaseModel, Field

# Gpt 4.1 mini used for this repro
# The only difference between the working and failing agents is the response_format parameter
#
# "agent-framework>=1.0.0b260106",
# "agent-framework-azure-ai>=1.0.0b260106",
# "agent-framework-core>=1.0.0b260106",
# "agent-framework-devui>=1.0.0b260106",
#
load_dotenv()


class WeatherResponse(BaseModel):
    """
    Structured response from the Weather agent.
    """
    temperature: float = Field(..., description="The temperature in Celsius")
    location: str = Field(...,
                          description="The location for the weather report")


def get_weather(
    location: Annotated[str, Field(description="The location to get the weather for.")],
) -> str:
    """Get the weather for a given location."""
    conditions = ["sunny", "cloudy", "rainy", "stormy"]
    return f"The weather in {location} is {conditions[randint(0, 3)]} with a high of {randint(10, 30)}°C."


async def working_agent_interaction():
    async with (
        AzureCliCredential() as credential,
        AzureAIClient(project_endpoint=os.environ["AZURE_AI_PROJECT_ENDPOINT"],
                      model_deployment_name=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
                      credential=credential).create_agent(
            name="WeatherWorkingAgent",
            instructions="You are a helpful weather agent.",
            tools=get_weather,
        ) as agent,
    ):
        query = "What's the weather like in Portland?"
        print(f"User: {query}")
        print(f"Agent ID: {agent.id}")
        print("Agent: ", end="", flush=True)
        async for chunk in agent.run_stream(query):
            if chunk.text:
                print(chunk.text, end="", flush=True)
        print("\n")


async def failing_agent_interaction():
    async with (
        AzureCliCredential() as credential,
        AzureAIClient(project_endpoint=os.environ["AZURE_AI_PROJECT_ENDPOINT"],
                      model_deployment_name=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
                      credential=credential).create_agent(
            name="WeatherFailingAgent",
            instructions="You are a helpful weather agent.",
            tools=get_weather,
            response_format=WeatherResponse
        ) as agent,
    ):
        query = "What's the weather like in Portland?"
        print(f"User: {query}")
        print(f"Agent ID: {agent.id}")
        print("Agent: ", end="", flush=True)
        async for chunk in agent.run_stream(query):
            if chunk.text:
                print(chunk.text, end="", flush=True)
        print("\n")


async def main() -> None:
    await working_agent_interaction()
    await failing_agent_interaction()

if __name__ == "__main__":
    asyncio.run(main())

Error Messages / Stack Traces

# Agent: Traceback (most recent call last):
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/openai/_responses_client.py", line 137, in _inner_get_streaming_response
#     async with client.responses.stream(**run_options) as response:
#                ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/openai/resources/responses/responses.py", line 2502, in stream
#     raise ValueError("model must be provided when creating a new response")
# ValueError: model must be provided when creating a new response

# The above exception was the direct cause of the following exception:

# Traceback (most recent call last):
#   File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 81, in <module>
#     asyncio.run(main())
#     ~~~~~~~~~~~^^^^^^^^
#   File "/usr/local/lib/python3.13/asyncio/runners.py", line 195, in run
#     return runner.run(main)
#            ~~~~~~~~~~^^^^^^
#   File "/usr/local/lib/python3.13/asyncio/runners.py", line 118, in run
#     return self._loop.run_until_complete(task)
#            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^
#   File "/usr/local/lib/python3.13/asyncio/base_events.py", line 725, in run_until_complete
#     return future.result()
#            ~~~~~~~~~~~~~^^
#   File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 78, in main
#     await failing_agent_interaction()
#   File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 70, in failing_agent_interaction
#     async for chunk in agent.run_stream(query):
#         if chunk.text:
#             print(chunk.text, end="", flush=True)
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/observability.py", line 1408, in trace_run_streaming
#     async for streaming_agent_response in run_streaming_func(self, messages=messages, thread=thread, **kwargs):
#         yield streaming_agent_response
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_agents.py", line 1033, in run_stream
#     async for update in self.chat_client.get_streaming_response(
#     ...<18 lines>...
#         )
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_tools.py", line 2011, in streaming_function_invocation_wrapper
#     async for update in func(self, messages=prepped_messages, **filtered_kwargs):
#         all_updates.append(update)
#         yield update
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/observability.py", line 1160, in trace_get_streaming_response
#     async for update in func(self, messages=messages, **kwargs):
#         yield update
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_middleware.py", line 1419, in _stream_generator
#     async for update in original_get_streaming_response(self, messages, **kwargs):
#         yield update
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_clients.py", line 676, in get_streaming_response
#     async for update in self._inner_get_streaming_response(
#     ...<2 lines>...
#         yield update
#   File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/openai/_responses_client.py", line 153, in _inner_get_streaming_response
#     raise ServiceResponseException(
#     ...<2 lines>...
#     ) from ex
# agent_framework.exceptions.ServiceResponseException: <class 'agent_framework_azure_ai._client.AzureAIClient'> service failed to complete the prompt: model must be provided when creating a new response

Package Versions

1.0.0b260106

Python Version

3.13

Additional Context

No response

[Dezzley]

Same issue when passing response_format to the .run_stream method.

    chat_agent = ChatAgent(chat_client=chat_client)
    async for chunk in chat_agent.run_stream("What is advil?", response_format=MedicineInfo):
        ...

[moonbox3]

Hi there, thanks for filing. I am not able to reproduce the error ValueError: model must be provided when creating a new response with your sample code. I was able to reproduce a schema failure related to the BaseModel missing model_config with extra="Forbid", which can be resolve by adding:

model_config = ConfigDict(extra="forbid")

To your BaseModel:

from pydantic import BaseModel, ConfigDict, Field

class WeatherResponse(BaseModel):
    """
    Structured response from the Weather agent.
    """

    temperature: float = Field(..., description="The temperature in Celsius")
    location: str = Field(..., description="The location for the weather report")
    model_config = ConfigDict(extra="forbid")

This is a requirement from the OpenAI/Azure AI API for Structured Outputs.

From the OpenAI documentation:

additionalProperties: false must always be set in objects

Structured Outputs only supports generating specified keys/values, so we require developers to set additionalProperties: false to opt into Structured Outputs.

When you set extra="forbid" in Pydantic's ConfigDict, it causes model_json_schema() to include "additionalProperties": false in the generated JSON schema. Without it, Pydantic omits this field entirely, and the API rejects the schema.

TL;DR: Any Pydantic model used as a response_format must have model_config = ConfigDict(extra="forbid") to satisfy the API's schema requirements.

After adding that to your sample code, I see:

User: What's the weather like in Portland?
Agent ID: <id>
Agent: The weather in Portland is currently stormy, with a high temperature of 13°C. If you're heading outside, it's a good idea to be prepared for wet and windy conditions!

User: What's the weather like in Portland?
Agent ID: <id>
Agent: {"temperature":21,"location":"Portland"}

[moonbox3]

Can you also please test with 1.0.0b260107? There was a regression fix in that release related to response format schema handling. Would be good to see your behavior in that release, so I can properly investigate if you continue to see a stack trace. Thanks.

[damienaicheh]

Hi Evan,

with the latest version package I have this stacktrace:

User: What's the weather like in Portland?
Agent ID: 8a538645-ef61-4bda-87ac-2ef3d2967fbb
Agent: The weather in Portland is currently stormy with a high of 12°C. Would you like to know the forecast for the next few days?

User: What's the weather like in Portland?
Agent ID: 91d3e0ff-b88a-40f5-804e-0035111229e6
Agent: Traceback (most recent call last):
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/openai/_responses_client.py", line 132, in _inner_get_streaming_response
    async for chunk in await client.responses.create(stream=True, **run_options):
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/openai/resources/responses/responses.py", line 2320, in create
    return await self._post(
           ^^^^^^^^^^^^^^^^^
    ...<42 lines>...
    )
    ^
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/openai/_base_client.py", line 1794, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/openai/_base_client.py", line 1594, in request
    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'WeatherResponse': In context=(), 'additionalProperties' is required to be supplied and to be false.", 'type': 'invalid_request_error', 'param': 'text.format.schema', 'code': 'invalid_json_schema'}}

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 143, in <module>
    asyncio.run(main())
    ~~~~~~~~~~~^^^^^^^^
  File "/usr/local/lib/python3.13/asyncio/runners.py", line 195, in run
    return runner.run(main)
           ~~~~~~~~~~^^^^^^
  File "/usr/local/lib/python3.13/asyncio/runners.py", line 118, in run
    return self._loop.run_until_complete(task)
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/usr/local/lib/python3.13/asyncio/base_events.py", line 725, in run_until_complete
    return future.result()
           ~~~~~~~~~~~~~^^
  File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 140, in main
    await failing_agent_interaction()
  File "/workspaces/repro/src/./test_tmp/bug_response_format_repro.py", line 132, in failing_agent_interaction
    async for chunk in agent.run_stream(query):
        if chunk.text:
            print(chunk.text, end="", flush=True)
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/observability.py", line 1408, in trace_run_streaming
    async for streaming_agent_response in run_streaming_func(self, messages=messages, thread=thread, **kwargs):
        yield streaming_agent_response
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_agents.py", line 1033, in run_stream
    async for update in self.chat_client.get_streaming_response(
    ...<18 lines>...
        )
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_tools.py", line 2011, in streaming_function_invocation_wrapper
    async for update in func(self, messages=prepped_messages, **filtered_kwargs):
        all_updates.append(update)
        yield update
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/observability.py", line 1160, in trace_get_streaming_response
    async for update in func(self, messages=messages, **kwargs):
        yield update
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_middleware.py", line 1419, in _stream_generator
    async for update in original_get_streaming_response(self, messages, **kwargs):
        yield update
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/_clients.py", line 676, in get_streaming_response
    async for update in self._inner_get_streaming_response(
    ...<2 lines>...
        yield update
  File "/workspaces/repro/src/.venv/lib/python3.13/site-packages/agent_framework/openai/_responses_client.py", line 148, in _inner_get_streaming_response
    raise ServiceResponseException(
    ...<2 lines>...
    ) from ex
agent_framework.exceptions.ServiceResponseException: <class 'agent_framework_azure_ai._client.AzureAIClient'> service failed to complete the prompt: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'WeatherResponse': In context=(), 'additionalProperties' is required to be supplied and to be false.", 'type': 'invalid_request_error', 'param': 'text.format.schema', 'code': 'invalid_json_schema'}}

If I add on top the Pydantic option I have (as you can see the agent answered twice, I ran it multiple time, sometimes it answer once, up to 3 times):

User: What's the weather like in Portland?
Agent ID: 757874b3-65b7-466c-8f75-99c1e706e591
Agent: The weather in Portland is stormy with a high of 19°C.

User: What's the weather like in Portland?
Agent ID: 9480069b-67d3-468c-af38-ea3f594edd55
Agent: {"temperature":16,"location":"Portland"}{"temperature":16,"location":"Portland"}