Memory Patterns Common patterns for managing memory in AI agents and LLM applications.
In-Memory Storage Store conversation in application memory.
from agno import Agent from agno.memory import Memory agent = Agent( model = OpenAI( id = "gpt-4o" ), memory = Memory(), add_history_to_messages = True ) # Conversation persists across runs response1 = agent.run( "My name is John" ) response2 = agent.run( "What is my name?" ) # Remembers "John"
Database Storage Persist conversations to database.
from agno.memory.db.sqlite import SqliteMemory agent = Agent( memory = SqliteMemory( table_name = "agent_sessions" , db_file = "conversations.db" ), session_id = "user-123" )
Semantic Memory (Mem0) Store and retrieve memories by semantic similarity.
from mem0 import Memory import qdrant_client # Initialize Mem0 with Qdrant memory = Memory.from_config({ "vector_store" : { "provider" : "qdrant" , "config" : { "host" : "localhost" , "port" : 6333 } } }) # Add memories memory.add( "User prefers Python over JavaScript" , user_id = "user-123" ) # Search memories results = memory.search( "What programming language?" , user_id = "user-123" )
Shared Memory Share memory across multiple agents.
from agno.memory import Memory shared_memory = Memory() agent1 = Agent( memory = shared_memory, name = "Researcher" ) agent2 = Agent( memory = shared_memory, name = "Writer" ) # Both agents access same conversation history research = agent1.run( "Research quantum computing" ) article = agent2.run( "Write article based on research" )
Memory Configuration Store user-specific memories
Generate summaries of conversations
update_user_memories_after_run
Update memories after each interaction
Best Practices Use in-memory storage for short sessions, database storage for production applications.
Be mindful of PII and sensitive data in memory stores. Implement appropriate data retention and privacy policies.
Memory Management Learn advanced memory techniques
AI Agents Build stateful agents
