Skip to main content

​
Parameters

ParameterTypeDefaultDescription
memoryOptional[Memory]NoneThe configured Memory object from the parent agent

​
Functions

​
get_message_history

Provides the prepared message history (full session memory) to the agent’s core run method. Returns:
  • List[Any]: The prepared message history

​
get_context_injection

Provides the prepared context string (e.g., session summary) to the ContextManager. Returns:
  • str: The prepared context injection string

​
get_system_prompt_injection

Provides the prepared system prompt string (e.g., user profile) to the SystemPromptManager. Returns:
  • str: The prepared system prompt injection string

​
process_response

Captures the final model response from the LLM call, making it available for the memory update process on exit. Parameters:
  • model_response (Any): The model response to process
Returns:
  • Any: The processed model response

​
manage_memory

Asynchronous context manager for orchestrating memory operations throughout a task’s lifecycle. Returns:
  • AsyncContextManager: Context manager that handles memory operations
Usage: 
async with memory_manager.manage_memory() as manager:
    # Get prepared inputs
    history = manager.get_message_history()
    context = manager.get_context_injection()
    system_prompt = manager.get_system_prompt_injection()
    
    # Execute task
    response = await model.request(...)
    manager.process_response(response)
    # Memory updates handled automatically
Features:
  • On Entry: Calls memory.prepare_inputs_for_task() to prepare all necessary inputs (history, summaries, profiles)
  • On Exit: Calls memory.update_memories_after_task(model_response) to process the LLM response and update all relevant memories
  • Memory Integration: Seamlessly integrates with the Memory orchestrator
  • Async Processing: Handles memory updates asynchronously to avoid blocking
  • Context Preparation: Provides prepared inputs for other context managers
⌘I