Kickoff_for_each() throwing errors on memory tools

CrewAI Community Support
1

I’m developing a flow to kick off crews for web research on a long list of companies. This flow works well with just one company as input:

inputs = {
    "company": "[the company]",
    (...other inputs)
}

def run():

    _input = inputs
    
    try:
        MyCrew().crew().kickoff_for(inputs=_input)
    except Exception as e:
        raise Exception(f"An error occurred while running the crew: {e}")

However, when I run a similar flow with a list of multiple inputs using kickoff_for_each instead of kickoff, as in:

inputs = [
    {
        "company": company,
        (...other inputs)
    } for company in companies
]

def run():

    _input = inputs
    
    try:
        MyCrew().crew().kickoff_for_each(inputs=_input)
    except Exception as e:
        raise Exception(f"An error occurred while running the crew: {e}")

I get a litany of errors from pydantic about the memory tools I’m using:

      pydantic_core._pydantic_core.ValidationError: 3 validation errors for Crew
    short_term_memory
      Input should be an instance of ShortTermMemory [type=is_instance_of, input_value={'embedder_config': None,... object at 0x144d6d300>}, input_type=dict]
        For further information visit https://errors.pydantic.dev/2.10/v/is_instance_of
    long_term_memory
      Input should be an instance of LongTermMemory [type=is_instance_of, input_value={'embedder_config': None,... object at 0x144d6d1e0>}, input_type=dict]
        For further information visit https://errors.pydantic.dev/2.10/v/is_instance_of
    entity_memory
      Input should be an instance of EntityMemory [type=is_instance_of, input_value={'embedder_config': None,... object at 0x144d6d2d0>}, input_type=dict]
        For further information visit https://errors.pydantic.dev/2.10/v/is_instance_of

I’ve confirmed that the individual elements of the list comprehension are of the same structure as the version that works successfully with kickoff; if I create the list comprehension and run kickoff(inputs=_input[0]), everything works as expected, so I don’t think it’s a data structure issue.

For reference, this is my memory implementation (which works in single kickoff but not kickoff_for_each):

@crew
def crew(self) -> Crew:

    return Crew(
        agents=self.agents,
        tasks=self.tasks,
        process=Process.hierarchical,
        memory = True,
        
        long_term_memory = LongTermMemory(
            storage=LTMSQLiteStorage(
                db_path="memory/long_term_memory_storage.db"
            )
        ),
        short_term_memory = ShortTermMemory(
            storage = RAGStorage(
                    embedder_config={
                        "provider": "openai",
                        "config": {
                            "model": 'text-embedding-3-small'
                        }
                    },
                    type="short_term",
                    path="memory/"
                )
            ),
        entity_memory = EntityMemory(
            storage=RAGStorage(
                embedder_config={
                    "provider": "openai",
                    "config": {
                        "model": 'text-embedding-3-small'
                    }
                },
                type="short_term",
                path="memory/"
            )
        ),
        verbose=True,
        manager_llm=llm #defined using LLM()
    )

I’ve tried downgrading pydantic as far as I can, but 2.4.2 seems to be the earliest version that is still compatible with crewAI, and it still generates these errors. I’ll also note that if I disable / comment out all three memory tools for the crew, the script then works with kickoff_for_each, but does not produce satisfactory results; therefore, I’d like to figure out how to enable memory with kickoff_for_each. Anyone encountered similar issues?