How do we use `stream_events()` when `tool_choice` = "function_name" ?

[takasuza]

Question

If you use stream_event() after specifying the function name in tool_choice as follows,

• 3 validation errors for Response tool_choice.literal['none','auto','required'].

What should I do?

import random

@function_tool
def how_many_jokes() -> int:
    return random.randint(1, 10)

agent = Agent(
    name="Joker",
    instructions="First call the `how_many_jokes` tool, then tell that many jokes.",
    model=OpenAIChatCompletionsModel(
        model="gpt-4o-mini",
        openai_client=azure_openai_client,
    ),
    tools =[how_many_jokes],
    model_settings=ModelSettings(tool_choice ="how_many_jokes"),
    )


result = Runner.run_streamed(
    agent,
    input="Hello",
)
print("=== Run starting ===")

try:
    async for event in result.stream_events():
        # We'll ignore the raw responses event deltas
        if event.type == "raw_response_event":
            continue
        elif event.type == "agent_updated_stream_event":
            print(f"Agent updated: {event.new_agent.name}")
            continue
        elif event.type == "run_item_stream_event":
            if event.item.type == "tool_call_item":
                print("-- Tool was called")
            elif event.item.type == "tool_call_output_item":
                print(f"-- Tool output: {event.item.output}")
            elif event.item.type == "message_output_item":
                print(f"-- Message output:\n {ItemHelpers.text_message_output(event.item)}")
            else:
                pass  # Ignore other event types

except Exception as e:
    print(e)

print("=== Run complete ===")

[anasbaigmughal]

The error occurs because the tool_choice parameter in ModelSettings does not support specifying a particular function name like "how_many_jokes".

Issue:

You have:

model_settings=ModelSettings(tool_choice ="how_many_jokes")

But according to OpenAI Agents SDK, ModelSettings expects tool_choice to be one of:

• "none"
• "auto"
• "required"

It does not accept function names as values.

Fix:

Remove the tool_choice parameter from ModelSettings or use a valid value:

model_settings=ModelSettings(tool_choice="auto")  # or "required" if you need it

Alternatively, omit model_settings entirely, as ModelSettings is optional and has a default configuration:

agent = Agent(
    name="Joker",
    instructions="First call the `how_many_jokes` tool, then tell that many jokes.",
    model=OpenAIChatCompletionsModel(
        model="gpt-4o-mini",
        openai_client=azure_openai_client,
    ),
    tools=[how_many_jokes]
)

[takasuza]

Thank you for the information.

I assumed that it would be possible to specify the function name, as tool_choice has the following structure.

tool_choice: (
    Literal["auto", "required", "none"] | str | None
) = None

However, in reality, it seems that something like the following cannot be used, right...?
(Am I just imagining things when it seems to work with run()?)

model_settings=ModelSettings(tool_choice ="how_many_jokes")

[anasbaigmughal]

You're right that tool_choice appears to allow a str, but in practice, it only accepts "auto", "required", or "none".

The difference you're noticing between run() and run_streamed() might be due to internal handling differences in how tool selection is validated. If run() seems to work, it could be implicitly correcting or ignoring invalid values, whereas run_streamed() enforces stricter validation.

To ensure compatibility, it's best to stick with "auto", "required", or "none" or simply omit model_settings

[takasuza]

https://openai.github.io/openai-agents-python/agents/#forcing-tool-use

According to the above, it says that you can specify the tool name.
Therefore, I believe it's probably something like "auto", "required", "none", or the "function name (as a string)".

That's exactly why I believe the following code works.

@function_tool
def get_weather(location:str):
    """get_weather function"""
    return f"{location} is sunny"

test_agent = Agent(
    name="Assistant",
    instructions="Provide users with information about the weather",
    tools=[get_weather,get_info],
    model=OpenAIChatCompletionsModel(
        model="gpt-4o",
        openai_client=azure_openai_client,
    ),
    model_settings=ModelSettings(tool_choice ="get_weather"),
)

result = await Runner.run(
    test_agent, 
    input="hello",
)

[rm-openai]

Yes you should be able to specify a specific function name. Is it only during streaming? Does it also happen if you use OpenAI directly instead of azure?

[takasuza]

The code runs without any issues when not using streaming, but an error occurs only during streaming.
I tested it with the OpenAI API, and it worked fine.

Could this error be happening only when using the Azure OpenAI API?

[rm-openai]

Hmm yeah I'm not sure why it would be this way. Can you try directly hitting Azure with a curl and seeing if it still happens? Could be worth reaching out to azure directly if so

[takasuza]

Sorry, I'm not familiar with how to use curl. However, I’ve heard that others using Azure have experienced similar issues as well...

When we forcibly used Swarm on Azure, we also noticed that the streaming responses from OpenAI and Azure OpenAI were slightly different. By handling those differences, we were able to get Swarm working. So, I suspect that the discrepancy in streaming behavior might be the cause.

Here is the code we are using:

import random

@function_tool
def how_many_jokes() -> int:
    return random.randint(1, 10)

agent = Agent(
    name="Joker",
    instructions="First call the `how_many_jokes` tool, then tell that many jokes.",
    model=OpenAIChatCompletionsModel(
        model="gpt-4o-mini",
        openai_client=azure_openai_client,
    ),
    tools =[how_many_jokes],
    model_settings=ModelSettings(tool_choice ="how_many_jokes"),
    )


result = Runner.run_streamed(
    agent,
    input="Hello",
)
print("=== Run starting ===")
try:
    async for event in result.stream_events():
        # We'll ignore the raw responses event deltas
        if event.type == "raw_response_event":
            continue
        elif event.type == "agent_updated_stream_event":
            print(f"Agent updated: {event.new_agent.name}")
            continue
        elif event.type == "run_item_stream_event":
            if event.item.type == "tool_call_item":
                print("-- Tool was called")
            elif event.item.type == "tool_call_output_item":
                print(f"-- Tool output: {event.item.output}")
            elif event.item.type == "message_output_item":
                print(f"-- Message output:\n {ItemHelpers.text_message_output(event.item)}")
            else:
                pass  # Ignore other event types

except Exception as e:
    print(e)

print("=== Run complete ===")

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 3 days since being marked as stale.