The MemoryManager class enables agents to create, update, and recall long-term memories about users across sessions.
Constructor from agno.memory import MemoryManager from agno.db.sqlite import SQLiteDb db = SQLiteDb( db_file = "agent.db" ) memory_manager = MemoryManager( model = "gpt-4o" , db = db )
Parameters model
Model | str
default: "None"
Model used for memory extraction and management. Defaults to gpt-4o if not provided.
db
BaseDb | AsyncBaseDb
required
Database to store memories.
Custom system message for the memory manager.
memory_capture_instructions
Custom instructions for capturing memories.
Additional instructions appended to the system message.
Whether to allow adding new memories.
Whether to allow updating existing memories.
Whether to allow deleting memories.
Whether to allow clearing all memories.
Methods get_user_memories() Get all memories for a user.
memories = memory_manager.get_user_memories( user_id = "user_1" ) for memory in memories: print ( f " { memory.topic } : { memory.content } " )
aget_user_memories() Async version of get_user_memories().
memories = await memory_manager.aget_user_memories( user_id = "user_1" )
add_user_memory() Add a new memory.
from agno.db.schemas import UserMemory memory = UserMemory( content = "User prefers dark mode" , topic = "preferences" ) memory_id = memory_manager.add_user_memory( memory = memory, user_id = "user_1" )
delete_user_memory() Delete a specific memory.
memory_manager.delete_user_memory( memory_id = "mem_123" , user_id = "user_1" )
clear_user_memories() Clear all memories for a user.
memory_manager.clear_user_memories( user_id = "user_1" )
clear() Clear all memories from the database.
Agent Integration from agno import Agent from agno.memory import MemoryManager from agno.db.sqlite import SQLiteDb db = SQLiteDb( db_file = "agent.db" ) agent = Agent( model = "gpt-4o" , db = db, memory_manager = MemoryManager( db = db), update_memory_on_run = True , add_memories_to_context = True ) # Agent automatically creates and recalls memories response = agent.run( "My favorite color is blue" , user_id = "user_1" ) response = agent.run( "What's my favorite color?" , user_id = "user_1" )
Example Usage Basic Usage
Custom Instructions
Manual Memory Management
Multi-user
from agno import Agent from agno.memory import MemoryManager from agno.db.sqlite import SQLiteDb db = SQLiteDb( db_file = "memories.db" ) agent = Agent( model = "gpt-4o" , db = db, memory_manager = MemoryManager( db = db), update_memory_on_run = True , add_memories_to_context = True ) # Conversation 1 response = agent.run( "I love pizza" , user_id = "alice" ) # Conversation 2 (different session, same user) response = agent.run( "What should I have for dinner?" , user_id = "alice" , session_id = "session_2" ) # Agent recalls: "I remember you love pizza!"
UserMemory Schema Memories are stored as UserMemory objects:
memory_id: Unique identifieruser_id: User the memory belongs tocontent: The memory contenttopic: Optional topic/categorymetadata: Additional metadataupdated_at: Last update timestamp
Best Practices
Enable memory : Set update_memory_on_run=True for automatic memory creationAdd to context : Set add_memories_to_context=True for memory recallUser IDs : Always provide user_id for multi-user applicationsPrivacy : Implement proper access controls for user memoriesCleanup : Provide users a way to view and delete their memoriesTopics : Use topics to organize different types of memories 