Learning Modes

Control when and how agents learn.

Learning modes control when and how a Learning Machine captures information. Each store can use a different mode.

Mode	How it works	Tradeoff
Always	Extraction runs automatically after each response	Extra LLM call per interaction
Agentic	Agent receives tools and decides what to save	May miss implicit information
Propose	Agent proposes learnings, user confirms before saving	Requires user interaction

Always Mode

Extraction happens automatically in the background. No agent tools involved.

1from kern.agent import Agent
2from kern.db.postgres import PostgresDb
3from kern.learn import LearningMachine, LearningMode, UserProfileConfig
4from kern.models.openai import OpenAIResponses
5
6agent = Agent(
7    model=OpenAIResponses(id="gpt-5.2"),
8    db=PostgresDb(db_url="postgresql+psycopg://ai:ai@localhost:5532/ai"),
9    learning=LearningMachine(
10        user_profile=UserProfileConfig(mode=LearningMode.ALWAYS),
11    ),
12)
13
14# Profile info extracted automatically - no tool calls visible
15agent.print_response(
16    "I'm Alice Chen, but please call me Ali.",
17    user_id="alice@example.com",
18)

Best for: User Profile, User Memory, Session Context, Entity Memory

Agentic Mode

The agent receives tools and decides when to save.

1from kern.learn import LearningMachine, LearningMode, UserProfileConfig
2
3agent = Agent(
4    model=OpenAIResponses(id="gpt-5.2"),
5    db=db,
6    learning=LearningMachine(
7        user_profile=UserProfileConfig(mode=LearningMode.AGENTIC),
8    ),
9)
10
11# Agent decides to call update_profile tool
12agent.print_response(
13    "Remember that I prefer dark mode interfaces.",
14    user_id="alice@example.com",
15)

Best for: Learned Knowledge, Decision Log

Tools by Store

Store	Tools
User Profile	`update_profile`
User Memory	`update_user_memory`
Entity Memory	`search_entities`, `create_entity`, `update_entity`, `add_fact`, `update_fact`, `delete_fact`, `add_event`, `add_relationship`
Learned Knowledge	`search_learnings`, `save_learning`
Decision Log	`log_decision`, `record_outcome`, `search_decisions`

Propose Mode

The agent proposes learnings. User must confirm before saving.

1from kern.learn import LearningMachine, LearningMode, LearnedKnowledgeConfig
2
3agent = Agent(
4    model=OpenAIResponses(id="gpt-5.2"),
5    db=db,
6    learning=LearningMachine(
7        knowledge=knowledge,
8        learned_knowledge=LearnedKnowledgeConfig(mode=LearningMode.PROPOSE),
9    ),
10)
11
12# Agent proposes, user confirms
13agent.print_response(
14    "That's a great insight about API rate limits - we should remember that.",
15    user_id="alice@example.com",
16)

Note: Propose mode is currently intended for Learned Knowledge.

Best for: High-stakes knowledge, regulated environments, quality control

Combining Modes

Use different modes for different stores:

1from kern.learn import (
2    LearningMachine,
3    LearningMode,
4    UserProfileConfig,
5    UserMemoryConfig,
6    LearnedKnowledgeConfig,
7)
8
9agent = Agent(
10    model=OpenAIResponses(id="gpt-5.2"),
11    db=db,
12    learning=LearningMachine(
13        user_profile=UserProfileConfig(mode=LearningMode.ALWAYS),     # Automatic
14        user_memory=UserMemoryConfig(mode=LearningMode.ALWAYS),       # Automatic
15        learned_knowledge=LearnedKnowledgeConfig(mode=LearningMode.AGENTIC),  # Agent-driven
16    ),
17)

Defaults by Store

Store	Default mode	Reason
User Profile	Always	Names and preferences should be captured consistently
User Memory	Always	Observations accumulate passively
Session Context	Always	Session state needs continuous tracking
Entity Memory	Always	Continuous extraction captures entity facts/events from normal conversations
Learned Knowledge	Agentic	Agent decides what insights are worth saving
Decision Log	Always (`DecisionLogConfig()`), Agentic (`decision_log=True`)	Supports both automatic logging and explicit logging workflows

Choosing a Mode

Scenario	Mode
Capture user names and preferences	Always
Build user memory automatically	Always
Track session progress	Always
Agent-driven knowledge capture	Agentic
Build entity knowledge graphs	Always
Audit agent decisions	Agentic
High-value collective knowledge	Propose
Compliance-sensitive learning	Propose