The Learning system enables agents to learn from interactions, build knowledge over time, and improve their performance. It combines memory, knowledge updates, and adaptive behavior.
LearningMachine The LearningMachine class provides unified learning capabilities for agents.
from agno import Agent from agno.learn.machine import LearningMachine from agno.db.sqlite import SQLiteDb db = SQLiteDb( db_file = "agent.db" ) agent = Agent( model = "gpt-4o" , db = db, learning = True , # Enable learning add_learnings_to_context = True )
Enabling Learning learning
bool | LearningMachine
default: "None"
Enable learning. Use True for default learning or provide a LearningMachine instance for custom configuration.
If True, adds learnings to the agent’s context.
How Learning Works
Experience Collection : Agent interactions are stored in the databasePattern Recognition : System identifies patterns in successful interactionsKnowledge Extraction : Useful information is extracted and storedContext Enhancement : Learnings are added to future interactions
Example Usage Basic Learning
Custom LearningMachine
Multi-User Learning
Learning with Memory
Learning from Feedback
Domain-Specific Learning
from agno import Agent from agno.db.sqlite import SQLiteDb db = SQLiteDb( db_file = "learning.db" ) agent = Agent( model = "gpt-4o" , db = db, learning = True , add_learnings_to_context = True ) # Agent learns from interactions response = agent.run( "I prefer concise answers" , user_id = "user_1" ) # In future interactions, agent remembers preference response = agent.run( "Explain quantum computing" , user_id = "user_1" ) # Agent provides concise answer based on learned preference
Learning Types 1. User Preferences Learn communication style, detail level, formatting preferences:
agent.run( "I prefer bullet points over paragraphs" , user_id = "user_1" ) # Future responses use bullet points
2. Domain Knowledge Learn facts and information from interactions:
agent.run( "Our API rate limit is 1000 requests/hour" , user_id = "admin" ) # Agent remembers and uses this information
3. Task Patterns Learn successful approaches to tasks:
agent.run( "When analyzing code, always check for edge cases first" ) # Agent prioritizes edge cases in future code analysis
4. Error Corrections Learn from mistakes:
agent.run( "That calculation was wrong. Always use parentheses for order of operations" ) # Agent improves future calculations
Accessing Learnings from agno import Agent agent = Agent( model = "gpt-4o" , db = db, learning = True ) # Run some interactions agent.run( "Training data..." , user_id = "user_1" ) # Access learning machine if agent.learning_machine: # Get learnings for user learnings = agent.learning_machine.get_learnings( user_id = "user_1" ) for learning in learnings: print (learning.content)
Combining Learning Features 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 for facts memory_manager = MemoryManager( db = db), update_memory_on_run = True , add_memories_to_context = True , # Learning for patterns and preferences learning = True , add_learnings_to_context = True , # History for context add_history_to_context = True , ) # Agent uses all three: # - Memory: Remembers user facts # - Learning: Learns preferences and patterns # - History: Maintains conversation context
Best Practices
Enable database : Learning requires a database to store patternsUser IDs : Use consistent user IDs for personalized learningFeedback loops : Encourage users to provide feedbackPrivacy : Implement proper data governance for learned informationMonitoring : Review what the agent is learningCombining features : Use with memory and history for best resultsClear learnings : Provide option to clear learned patterns
Use Cases
Personal assistants : Learn user preferences and habitsCustomer support : Learn effective resolution patternsEducational tutors : Adapt to student learning styleCode assistants : Learn project conventions and patternsSales agents : Learn effective sales approachesHealthcare : Learn patient communication preferences
Limitations
Requires database for persistence
Learning quality depends on interaction volume
May need periodic review and cleanup
Privacy considerations for learned data
Requires user IDs for personalized learning
