memory parameter uses in the Agent

[mallesh-sourcefuse]

Is the memory parameter in CrewAI's Agent class limited to a boolean option for default memory, or can it also accept a Langchain memory object for more advanced configurations?

research_agent = Agent( role="Researcher", goal="Gather relevant information.", backstory="Expert researcher.", llm=llm, memory=ConversationBufferWindowMemory(k=3, memory_key="chat_history", return_messages=True), verbose=True )

By passing chat history to the agent to understand the conversation flow and respond more accurately. Does CrewAI support passing custom chat history through Langchain memory objects?

@lucasgomide Can you please check this ??

[KeepALifeUS]

The memory parameter in CrewAI Agent is primarily boolean-based for built-in memory. For custom memory patterns, consider this approach:

State-Based Memory Pattern

from crewai import Agent, Task, Crew

# External memory state
memory_state = {
    "conversation_history": [],
    "key_facts": {},
    "context_window": []  # Last N interactions
}

def update_memory(agent_output, max_window=5):
    memory_state["conversation_history"].append(agent_output)
    memory_state["context_window"] = memory_state["conversation_history"][-max_window:]
    return memory_state

# Inject memory into agent context
agent = Agent(
    role="Assistant",
    goal="Help user with context awareness",
    backstory=f"""You have access to conversation history:
    {memory_state["context_window"]}
    
    Key facts remembered: {memory_state["key_facts"]}""",
    memory=True  # Enable built-in + your custom context
)

Why This Works

1. Flexibility: Control exactly what gets remembered
2. Window Size: Implement your own sliding window (like ConversationBufferWindowMemory)
3. Persistence: Easy to save/load from Redis, JSON, or DB
4. Composable: Works alongside CrewAI's built-in memory

For multi-agent scenarios where agents need shared memory, external state becomes essential - each agent reads/writes to the same memory store.

More on state-based patterns: https://github.com/KeepALifeUS/autonomous-agents