> ## Documentation Index > Fetch the complete documentation index at: https://docs.praison.ai/llms.txt > Use this file to discover all available pages before exploring further. # Learn vs Train > Understanding the difference between passive learning and active training PraisonAI provides two distinct systems for improving agent behavior: **Agent Learn** (passive) and **Agent Train** (active). Understanding when to use each is key to building effective agents. ## Agent Learn (Passive) Automatically captures patterns during normal interactions: ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} graph LR I[💬 User Interaction] --> C[🧠 Auto-Capture] C --> S[(💾 Learn Stores)] S --> R[🔍 Retrieve] R --> E[⚡ Enhanced Response] classDef input fill:#6366F1,stroke:#7C90A0,color:#fff classDef process fill:#10B981,stroke:#7C90A0,color:#fff classDef store fill:#F59E0B,stroke:#7C90A0,color:#fff class I input class C,R,E process class S store ``` ## Agent Train (Active) Explicit iterative improvement with feedback: ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} graph LR S[📋 Scenario] --> A[🤖 Execute] A --> O[📝 Output] O --> G{👤 Grade} G -->|Score + Feedback| I[📈 Improve] I --> A classDef scenario fill:#8B0000,stroke:#7C90A0,color:#fff classDef process fill:#6366F1,stroke:#7C90A0,color:#fff classDef grade fill:#F59E0B,stroke:#7C90A0,color:#fff class S scenario class A,O,I process class G grade ``` *** ## Quick Comparison | Aspect | Agent Learn | Agent Train | | ----------------- | --------------------------------- | --------------------------- | | **Type** | Passive / Automatic | Active / Explicit | | **When** | During normal interactions | Dedicated training sessions | | **Feedback** | Implicit (auto-detected patterns) | Explicit (human/LLM scores) | | **Purpose** | Remember preferences & patterns | Improve specific behaviors | | **Storage** | 7 specialized stores | Scenarios & reports | | **Iterations** | Continuous | Fixed (1-N iterations) | | **Code Location** | `praisonaiagents` (core SDK) | `praisonai` (wrapper) | *** ## When to Use Each ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} graph TD Q1{What do you need?} Q1 -->|Remember user preferences| Learn Q1 -->|Improve specific behaviors| Train Q1 -->|Adapt over time| Learn Q1 -->|Focused improvement session| Train Q1 -->|Automatic capture| Learn Q1 -->|Human feedback| Train Learn[✅ Use Agent Learn] Train[✅ Use Agent Training] classDef learn fill:#10B981,stroke:#7C90A0,color:#fff classDef train fill:#6366F1,stroke:#7C90A0,color:#fff class Learn learn class Train train ``` ### Use Agent Learn When: * **Remembering user preferences** - Dark mode, language, communication style * **Capturing domain knowledge** - Project-specific terms, codebase patterns * **Building context over time** - Session history, conversation threads * **Automatic adaptation** - No intervention needed * **Long-term memory** - Persistent across sessions ### Use Agent Training When: * **Improving specific responses** - Better greetings, more accurate answers * **Quality assurance** - Iterative refinement with scoring * **Human feedback loops** - Expert-in-the-loop improvement * **Benchmarking behavior** - Measurable improvement metrics * **One-time improvement sessions** - Focused training runs *** ## Data Flow Comparison ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} sequenceDiagram participant User participant Agent participant LearnManager participant Stores User->>Agent: "I prefer Python" Agent->>LearnManager: Auto-capture LearnManager->>Stores: Store in PersonaStore Note over Stores: Persisted automatically User->>Agent: "Set up my environment" Agent->>LearnManager: Get context LearnManager->>Stores: Retrieve preferences Stores-->>Agent: "User prefers Python" Agent-->>User: Python-focused setup ``` **Key Points:** * Happens automatically during `agent.start()` * No explicit feedback required * Stores in 7 specialized stores * Retrieved automatically for future interactions ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} sequenceDiagram participant User participant Trainer as AgentTrainer participant Agent participant Grader User->>Trainer: Add scenario Trainer->>Agent: Execute Agent->>Trainer: Response Trainer->>Grader: Grade (score 7/10) Grader->>Trainer: "Needs more detail" Trainer->>Agent: Improved prompt Agent->>Trainer: Better response Trainer->>Grader: Grade (score 9/10) Trainer->>User: Training report ``` **Key Points:** * Explicit training session * Human or LLM provides scores * Iterative improvement loop * Generates measurable reports *** ## Storage Comparison ### Agent Learn Stores | Store | Purpose | Auto-Captured | | ------------------ | ---------------------------- | ------------- | | `PersonaStore` | User preferences, profile | ✅ Yes | | `InsightStore` | Observations, learnings | ✅ Yes | | `ThreadStore` | Session/conversation context | ✅ Yes | | `PatternStore` | Reusable knowledge patterns | Optional | | `DecisionStore` | Decision logging | Optional | | `FeedbackStore` | Outcome signals | Optional | | `ImprovementStore` | Self-improvement proposals | Optional | ### Agent Training Storage | Data | Purpose | Format | | ------------------- | ----------------------- | ------ | | `TrainingScenario` | Input + expected output | JSON | | `TrainingIteration` | Per-iteration results | JSON | | `TrainingReport` | Summary with scores | JSON | *** ## Code Examples ```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} from praisonaiagents import Agent # Enable learning with simple shorthand agent = Agent( name="Assistant", instructions="You are a helpful assistant", learn=True # Enables AGENTIC mode (auto-learning) ) # Learnings captured automatically agent.start("I prefer concise answers") agent.start("Always use Python examples") # Later - agent remembers preferences agent.start("How do I sort a list?") # Agent gives concise Python answer ``` ```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} from praisonai.train.agents import AgentTrainer, TrainingScenario, apply_training from praisonaiagents import Agent agent = Agent( name="Assistant", instructions="You are a helpful assistant" ) # Create explicit training session trainer = AgentTrainer( agent=agent, iterations=5, human_mode=True # Get human feedback ) trainer.add_scenario(TrainingScenario( id="greeting", input_text="Hello!", expected_output="Friendly, welcoming response" )) # Run training with feedback report = trainer.run() print(f"Score improved: {report.improvement:+.1f}") # Apply the best iteration at runtime apply_training(agent, session_id=report.session_id) # Now agent uses learned improvements response = agent.start("Hello!") ``` *** ## Using Both Together Agent Learn and Agent Training are **complementary**. Use them together for best results: ```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} from praisonaiagents import Agent from praisonai.train.agents import AgentTrainer, TrainingScenario # Create agent with learning enabled agent = Agent( name="Assistant", instructions="You are a helpful assistant", learn=True # Passive learning always on ) # Periodically run training sessions trainer = AgentTrainer(agent=agent, iterations=3, human_mode=True) trainer.add_scenario(TrainingScenario( id="quality_check", input_text="Explain machine learning", expected_output="Clear, accurate explanation" )) # Training improves specific behavior report = trainer.run() # Agent continues learning passively # AND benefits from training improvements ``` ```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}} graph LR subgraph "Combined Approach" A[🤖 Agent] --> L[📚 Learns Continuously] A --> T[🏋️ Trains Periodically] L --> B[Better Context] T --> B B --> A end classDef agent fill:#6366F1,stroke:#7C90A0,color:#fff classDef system fill:#F59E0B,stroke:#7C90A0,color:#fff classDef result fill:#10B981,stroke:#7C90A0,color:#fff class A agent class L,T system class B result ``` *** ## Summary | Feature | Agent Learn | Agent Training | | --------------- | ------------------------ | ------------------------ | | **Enable** | `learn=True` | `AgentTrainer(agent)` | | **Trigger** | Automatic | Manual | | **Feedback** | None needed | Score + suggestions | | **Best for** | Preferences, context | Quality improvement | | **Persistence** | Learn stores | Training reports | | **CLI** | `praisonai memory learn` | `praisonai train agents` | *** ## Related Passive continuous learning Active iterative training Agent memory systems CLI reference