Support Responses API in create_react_agent

[humungasaurus]

The create_react_agent prebuilt is a very useful, and provides most of the functionality out of the box that most folks need but it doesn't support OpenAI's responses api. When you pass in an LLM like this:

ChatOpenAI(model=LLM_MODEL_NAME, use_responses_api=True)

Graph executions start failing.

The biggest issue I am having with this is it means that reasoning output available in newer models cannot be accessed when using the create_react_agent prebuilt.

[ccurme]

Hello, thanks for posting this. Can you clarify some of the errors you're running into? Code, stack traces, error messages would be helpful.

As far as I know the Responses API (including reasoning summaries) is supported. Here is an example:

from langchain.chat_models import init_chat_model
from langgraph.prebuilt import create_react_agent


def get_weather(location: str) -> str:
    """Get the weather at a location."""
    return "It's sunny."

tools = [get_weather]

reasoning = {"effort": "medium", "summary": "auto"}

llm = init_chat_model(
    "openai:o4-mini",
    use_responses_api=True,
    model_kwargs={"reasoning": reasoning}
)

agent = create_react_agent(llm, tools)

input_message = {
    "role": "user",
    "content": (
        "What is the weather now at the city that had the third "
        "highest population in the world in the year 2000?"
    ),
}
for step in agent.stream(
    {"messages": [input_message]},
    stream_mode="values",
):
    message = step["messages"][-1]
    message.pretty_print()
    if reasoning := message.additional_kwargs.get("reasoning"):
        print(f"Reasoning: {reasoning}")

================================ Human Message =================================

What is the weather now at the city that had the third highest population in the world in the year 2000?
================================== Ai Message ==================================

[]
Tool Calls:
  get_weather (call_5pLCOaNOkAna46Mae5b45uKx)
 Call ID: call_5pLCOaNOkAna46Mae5b45uKx
  Args:
    location: New York City
Reasoning: {'id': 'rs_682cfd5ea3548191ad81b25df78f1ec105ed6e9b11c816fc', 'summary': [{'text': "**Determining weather location**\n\nThe user wants to know the current weather in the city that had the third highest population in 2000. I need to identify that city first. Looking back, the largest cities then included Tokyo, Delhi, and Shanghai. So, I’m trying to recall which one ranked third in population at that time. I’m also considering how population figures are typically counted—metropolitan area versus city proper, as this detail can be a bit ambiguous. Let's keep going!", 'type': 'summary_text'}, {'text': "**Identifying the third largest city**\n\nOkay, I'm trying to figure out the third largest city by population in the year 2000. From what I remember, the top three include Tokyo, New York City, and possibly Mexico City or Delhi. I’m focusing on metropolitan areas for clarity, noting some population figures—like Tokyo at around 35 million and New York at about 16.7 million. It seems New York could be the third largest, but I want to confirm this with reliable data from that year. Let’s do this!", 'type': 'summary_text'}, {'text': '**Confirming weather location**\n\nIn the year 2000, Mexico City and New York City both had metropolitan populations around 18.3 million. However, data from the UN suggests Mexico City was second with about 18.2 million, while New York was third with around 17.8 million. So New York City is the city I need to look up for the current weather. I’m going to call for the weather in New York City and structure that function call accordingly. Let’s proceed with that!', 'type': 'summary_text'}], 'type': 'reasoning'}
================================= Tool Message =================================
Name: get_weather

It's sunny.
================================== Ai Message ==================================

[{'type': 'text', 'text': 'The current weather in New York City is sunny.', 'annotations': []}]

(this is with latest langchain-openai and langgraph).

[humungasaurus]

Looking at this now, im thinking that maybe the issue is actually the callback handler and not the create_react_agent implementation?

So, this works:

def _get_llm(queue: Queue | None, agent_name: str):
    if queue:
        emit_handler = EmitHandler(queue, agent_name)
        cb_manager = CallbackManager([emit_handler])
        return ChatOpenAI(streaming=True, model=LLM_MODEL_NAME, callback_manager=cb_manager)
    else:
        return ChatOpenAI(streaming=True, model=LLM_MODEL_NAME)

def create_simple_agent(organization_id: str, connection_pool, queue: Queue = None):
    # Initialize checkpointer (for saving graph state)
    checkpointer = get_checkpointer(connection_pool)
    
    # redacted stuff related to the prompt
    
    return create_react_agent(
        _get_llm(queue, agent_name="simple_agent"),
        [key_definitions_search, data_dictionary_search, code_search, execute_bigquery, plot_data, ask_human],
        prompt=PROMPT.format(**prompt_variables),
        checkpointer=checkpointer,
        state_schema=State
    )

But this fails (same create_simple_agent):

def _get_llm(queue: Queue | None, agent_name: str):
    reasoning = {
        "effort": "medium",  # 'low', 'medium', or 'high'
        "summary": "auto",  # 'detailed', 'auto', or None
    }
    if queue:
        emit_handler = EmitHandler(queue, agent_name)
        cb_manager = CallbackManager([emit_handler])
        return ChatOpenAI(streaming=True, model=LLM_MODEL_NAME, callback_manager=cb_manager, use_responses_api=True, model_kwargs={"reasoning": reasoning})
    else:
        return ChatOpenAI(streaming=True, model=LLM_MODEL_NAME, use_responses_api=True, model_kwargs={"reasoning": reasoning})

Heres the stack trace:

  File "/app/routers/chat.py", line 55, in run_agent
    async for state in simple_agent.astream(
    ...<13 lines>...
        queue.put_nowait({"type": "state_update", "data": serialize_state(state)})
  File "/usr/local/lib/python3.13/site-packages/langgraph/pregel/__init__.py", line 2274, in astream
    async for _ in runner.atick(
    ...<7 lines>...
            yield o
  File "/usr/local/lib/python3.13/site-packages/langgraph/pregel/runner.py", line 444, in atick
    await arun_with_retry(
    ...<7 lines>...
    )
  File "/usr/local/lib/python3.13/site-packages/langgraph/pregel/retry.py", line 128, in arun_with_retry
    return await task.proc.ainvoke(task.input, config)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langgraph/utils/runnable.py", line 583, in ainvoke
    input = await step.ainvoke(input, config, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langgraph/utils/runnable.py", line 359, in ainvoke
    ret = await asyncio.create_task(coro, context=context)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langgraph/prebuilt/chat_agent_executor.py", line 696, in acall_model
    response = cast(AIMessage, await model_runnable.ainvoke(state, config))
                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/runnables/base.py", line 3075, in ainvoke
    input = await coro_with_context(part(), context, create_task=True)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/runnables/base.py", line 5429, in ainvoke
    return await self.bound.ainvoke(
           ^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<3 lines>...
    )
    ^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 393, in ainvoke
    llm_result = await self.agenerate_prompt(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<8 lines>...
    )
    ^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 967, in agenerate_prompt
    return await self.agenerate(
           ^^^^^^^^^^^^^^^^^^^^^
        prompt_messages, stop=stop, callbacks=callbacks, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 925, in agenerate
    raise exceptions[0]
  File "/usr/local/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 1093, in _agenerate_with_cache
    result = await self._agenerate(
             ^^^^^^^^^^^^^^^^^^^^^^
        messages, stop=stop, run_manager=run_manager, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/usr/local/lib/python3.13/site-packages/langchain_openai/chat_models/base.py", line 1126, in _agenerate
    return await agenerate_from_stream(stream_iter)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 182, in agenerate_from_stream
    chunks = [chunk async for chunk in stream]
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/langchain_openai/chat_models/base.py", line 2436, in _astream
    async for chunk in super()._astream_responses(*args, **kwargs):
        yield chunk
  File "/usr/local/lib/python3.13/site-packages/langchain_openai/chat_models/base.py", line 814, in _astream_responses
    async for chunk in response:
    ...<9 lines>...
            yield generation_chunk
  File "/usr/local/lib/python3.13/site-packages/openai/_streaming.py", line 147, in __aiter__
    async for item in self._iterator:
        yield item
  File "/usr/local/lib/python3.13/site-packages/openai/_streaming.py", line 193, in __stream__
    raise APIError(
    ...<3 lines>...
    )
openai.APIError: There was an issue with your request. Please check your inputs and try again```