function call is not respecting function args in 'load_memory'

[shu8hamrajput]

I have a agent that uses 'load_memory' to recall previous conversation. the 'load_memory' tool uses a arg 'query' to find relevent conversation. But multiple time the function call doesn't have any args.

root_agent = LlmAgent(
    name="Supervisor",
    model="gemini-2.0-flash",
    description=(
        "You are a supervisor agent that delegates survey modification tasks to other agents."
    ),
    instruction=(
        "You are a supervisor agent that delegates survey question modification tasks to other agents."
        "all queries are about survey discussed in the previous conversation."
        "use 'load_memory' with appropriate `query` when you want to access survey questions discussed in past conversation."
    ),
    tools=[ask_query, load_memory],
)

session = session_service.create_session(
    app_name=APP_NAME,
    user_id=USER_ID,
    session_id=SESSION_ID
)

news_list: list[str] = [
    "What is the latest news in technology?",
    "What are the top headlines in sports?",
    ...
]

session_service.append_event(session, Event(
    invocation_id=SESSION_ID,
    author="user",
    content=Content(
        parts=[
            Part(
                text=news
            )
            for news in news_list
        ]
    )
))

runner = Runner(
    agent=root_agent,
    app_name=APP_NAME,
    session_service=session_service,
    memory_service=memory_service
)
memory_service.add_session_to_memory(session)

on certain queries the load_memory gets successfully called with correct args. but sometime it doesn't

** successfull function call**:

 [Event] Author: Supervisor, Type: Event, Final: False, Content: parts=[Part(video_metadata=None, thought=None, code_execution_result=None, executable_code=None, file_data=None, function_call=FunctionCall(id=<adk-id>, args={'query': 'news about India'}, name='load_memory'), function_response=None, inline_data=None, text=None)] role='model'

** failed function call**:

 [Event]  Author: Supervisor, Type: Event, Final: False, Content: parts=[Part(video_metadata=None, thought=None, code_execution_result=None, executable_code=None, file_data=None, function_call=FunctionCall(id=<adk-id>, args={}, name='load_memory'), function_response=None, inline_data=None, text=None)] role='model'

How do i ensure that function call happens correctly or request again for correct function call?

[shu8hamrajput]

I am not sure if it's right way or not But I fixed this issue for myself by making query a required field in load_model declartion in file google/adk/tools/load_memory_tool.py:60

the updated class looks like:

class LoadMemoryTool(FunctionTool):
  """A tool that loads the memory for the current user."""

  def __init__(self):
    super().__init__(load_memory)

  @override
  def _get_declaration(self) -> types.FunctionDeclaration | None:
    return types.FunctionDeclaration(
        name=self.name,
        description=self.description,
        parameters=types.Schema(
            type=types.Type.OBJECT,
            properties={
                'query': types.Schema(
                    type=types.Type.STRING,
                )
            },
            required=['query'] // this makes the query mandatory field
        ),
    )

  @override
  async def process_llm_request(
      self,
      *,
      tool_context: ToolContext,
      llm_request: LlmRequest,
  ) -> None:
    await super().process_llm_request(
        tool_context=tool_context, llm_request=llm_request
    )
    # Tell the model about the memory.
    llm_request.append_instructions(["""
You have memory. You can use it to answer questions. If any questions need
you to look up the memory, you should call load_memory function with a query.
"""])

[shu8hamrajput]

raised a PR with changes #389