AssertionError: OpenAI requires `tool_call_id` #895

AlexEnrique · 2025-02-11T16:05:52Z

Description

Hi. I had run into an issue when switching models.
Basically, I have implemented an API endpoint where I can change models.

Where is what happened:

I started with gemini-1.5-flash, asking it what time is now, which would call my now() tool.
It runs without any problem returning the current datetime
Then I switched to gpt-4o-mini and asked the same question again, passing the message history I got after using Gemini
This causes the following exception: AssertionError: OpenAI requires tool_call_id to be set: ToolCallPart(tool_name='now', args={}, tool_call_id=None, part_kind='tool-call')

[Edit] Minimal working example

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIModel
from pydantic_ai.models.gemini import GeminiModel

from datetime import datetime


open_ai_api_key = ...
gemini_api_key = ...


openai_model = OpenAIModel(
    model_name='gpt-4o-mini',
    api_key=open_ai_api_key,
)

gemini_model = GeminiModel(
    model_name='gemini-2.0-flash-exp',  # could be gemini-1.5-flash also
    api_key=gemini_api_key,
)

agent = Agent(gemini_model)


@agent.tool_plain
def now():
    return datetime.now().isoformat()


r1 = agent.run_sync('what is the current date time?')
print(r1.all_messages_json())

r2 = agent.run_sync(  # this will fail
    'what time is now?',
    model=openai_model,
    message_history=r1.all_messages(),
)
print(r2.all_messages_json())

Message history (stored until call gpt-4o-mini)

[ModelRequest(parts=[SystemPromptPart(content='\nYou are a test agent.\n\nYou must do what the user asks.\n', dynamic_ref=None, part_kind='system-prompt'), UserPromptPart(content='call now', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 5, 628330, tzinfo=TzInfo(UTC)), part_kind='user-prompt')], kind='request'),
 ModelResponse(parts=[TextPart(content='I am sorry, I cannot fulfill this request. The available tools do not provide the functionality to make calls.\n', part_kind='text')], model_name='gemini-1.5-flash', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 6, 59052, tzinfo=TzInfo(UTC)), kind='response'),
 ModelRequest(parts=[UserPromptPart(content='call the tool now', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 14, 394461, tzinfo=TzInfo(UTC)), part_kind='user-prompt')], kind='request'),
 ModelResponse(parts=[TextPart(content='I cannot call a tool.  The available tools are functions that I can execute, not entities that I can call in a telephone sense.  Is there something specific you would like me to do with one of the available tools?\n', part_kind='text')], model_name='gemini-1.5-flash', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 15, 449295, tzinfo=TzInfo(UTC)), kind='response'),
 ModelRequest(parts=[UserPromptPart(content='what time is now?', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 23, 502937, tzinfo=TzInfo(UTC)), part_kind='user-prompt')], kind='request'),
 ModelResponse(parts=[ToolCallPart(tool_name='now', args={}, tool_call_id=None, part_kind='tool-call')], model_name='gemini-1.5-flash', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 24, 151395, tzinfo=TzInfo(UTC)), kind='response'),
 ModelRequest(parts=[ToolReturnPart(tool_name='now', content='2025-02-11T12:55:24.153651-03:00', tool_call_id=None, timestamp=datetime.datetime(2025, 2, 11, 15, 55, 24, 153796, tzinfo=TzInfo(UTC)), part_kind='tool-return')], kind='request'),
 ModelResponse(parts=[TextPart(content='The current time is 2025-02-11 12:55:24 -03:00.\n', part_kind='text')], model_name='gemini-1.5-flash', timestamp=datetime.datetime(2025, 2, 11, 15, 55, 24, 560881, tzinfo=TzInfo(UTC)), kind='response')]

Traceback

Traceback (most recent call last):
  File "/app/agents/_agents/_wrapper.py", line 125, in run_stream
    async with self._agent.run_stream(
               ~~~~~~~~~~~~~~~~~~~~~~^
        user_prompt=user_prompt,
        ^^^^^^^^^^^^^^^^^^^^^^^^
    ...<2 lines>...
        deps=self.deps,
        ^^^^^^^^^^^^^^^
    ) as result:
    ^
  File "/usr/local/lib/python3.13/contextlib.py", line 214, in __aenter__
    return await anext(self.gen)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/agent.py", line 595, in run_stream
    async with node.run_to_result(GraphRunContext(graph_state, graph_deps)) as r:
               ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/contextlib.py", line 214, in __aenter__
    return await anext(self.gen)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/_agent_graph.py", line 415, in run_to_result
    async with ctx.deps.model.request_stream(
               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        ctx.state.message_history, model_settings, model_request_parameters
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ) as streamed_response:
    ^
  File "/usr/local/lib/python3.13/contextlib.py", line 214, in __aenter__
    return await anext(self.gen)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 160, in request_stream
    response = await self._completions_create(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        messages, True, cast(OpenAIModelSettings, model_settings or {}), model_request_parameters
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 203, in _completions_create
    openai_messages = list(chain(*(self._map_message(m) for m in messages)))
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 267, in _map_message
    tool_calls.append(self._map_tool_call(item))
                      ~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/models/openai.py", line 284, in _map_tool_call
    id=_guard_tool_call_id(t=t, model_source='OpenAI'),
       ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic_ai/_utils.py", line 200, in guard_tool_call_id
    assert t.tool_call_id is not None, f'{model_source} requires `tool_call_id` to be set: {t}'
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: OpenAI requires `tool_call_id` to be set: ToolCallPart(tool_name='now', args={}, tool_call_id=None, part_kind='tool-call')

The text was updated successfully, but these errors were encountered:

AlexEnrique · 2025-02-11T16:12:09Z

Also, I was wondering if exceptions raised by a tool (not ModelRetry) deserve a type for it and should go into the ModelMessage, to be in the message history

sydney-runkle · 2025-02-14T20:41:51Z

Thanks for the report. Might be a good first issue for someone looking to help out with a bug fix.

Perhaps we should just use a dummy value here so that you can pass messages through in a model agnostic way...

AlexEnrique · 2025-02-15T14:31:33Z

@sydney-runkle
I was looking at the code, and I mapped the places where the tool_call_id may be or is None:

For Gemini

pydantic_ai.models.gemini:
- _process_response_from_parts (line 514)

For Mistral

pydantic_ai.models.mistral:
- MistralModel._map_mistral_to_pydantic_tool_call (line 332)
- MistralStreamedResponse._try_get_result_tool_from_text (line 537)

Since we have models that expects tool_call_id to be not None, I would suggest to change the ToolCallPart to have non-nullable tool_call_id, and generate dummy ids in all of these places.

I'll try to make a PR for this

sydney-runkle added the bug Something isn't working label Feb 13, 2025

sydney-runkle added the good first issue Good for newcomers label Feb 14, 2025

AlexEnrique linked a pull request Feb 15, 2025 that will close this issue

Add dummy tool call id for models that doesn't return it #933

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

AssertionError: OpenAI requires `tool_call_id` #895

AssertionError: OpenAI requires `tool_call_id` #895

AlexEnrique commented Feb 11, 2025 •

edited

Loading

AlexEnrique commented Feb 11, 2025 •

edited

Loading

sydney-runkle commented Feb 14, 2025

AlexEnrique commented Feb 15, 2025

AssertionError: OpenAI requires tool_call_id #895

AssertionError: OpenAI requires tool_call_id #895

Comments

AlexEnrique commented Feb 11, 2025 • edited Loading

Description

[Edit] Minimal working example

Message history (stored until call gpt-4o-mini)

Traceback

AlexEnrique commented Feb 11, 2025 • edited Loading

sydney-runkle commented Feb 14, 2025

AlexEnrique commented Feb 15, 2025

AssertionError: OpenAI requires `tool_call_id` #895

AssertionError: OpenAI requires `tool_call_id` #895

AlexEnrique commented Feb 11, 2025 •

edited

Loading

AlexEnrique commented Feb 11, 2025 •

edited

Loading