User Profile: Agentic Mode (Deep Dive)

"""
User Profile: Agentic Mode (Deep Dive)
======================================
Agent-controlled profile updates via explicit tools.

AGENTIC mode gives the agent a tool to update profile fields.
You'll see tool calls in the response - more transparent than ALWAYS mode.

Compare with: 01_always_extraction.py for automatic extraction.
See also: 01_basics/1b_user_profile_agentic.py for the basics.
"""

from agno.agent import Agent
from agno.db.postgres import PostgresDb
from agno.learn import LearningMachine, LearningMode, UserProfileConfig
from agno.models.openai import OpenAIResponses

# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------

db = PostgresDb(db_url="postgresql+psycopg://ai:ai@localhost:5532/ai")

agent = Agent(
    model=OpenAIResponses(id="gpt-5.5"),
    db=db,
    instructions=(
        "You are a helpful assistant. "
        "When users share their name or preferences, use update_user_profile to save it."
    ),
    learning=LearningMachine(
        user_profile=UserProfileConfig(
            mode=LearningMode.AGENTIC,
        ),
    ),
    markdown=True,
)

# ---------------------------------------------------------------------------
# Run Demo
# ---------------------------------------------------------------------------

if __name__ == "__main__":
    user_id = "jordan@example.com"

    # Session 1: Share name - watch for tool calls
    print("\n" + "=" * 60)
    print("SESSION 1: Share name (watch for tool calls)")
    print("=" * 60 + "\n")

    agent.print_response(
        "Hi! I'm Jordan Chen, but everyone calls me JC.",
        user_id=user_id,
        session_id="session_1",
        stream=True,
    )
    agent.learning_machine.user_profile_store.print(user_id=user_id)

    # Session 2: Recall in new session
    print("\n" + "=" * 60)
    print("SESSION 2: Profile recalled in new session")
    print("=" * 60 + "\n")

    agent.print_response(
        "What's my name and what should you call me?",
        user_id=user_id,
        session_id="session_2",
        stream=True,
    )
    agent.learning_machine.user_profile_store.print(user_id=user_id)

    # Session 3: Update preferred name
    print("\n" + "=" * 60)
    print("SESSION 3: Update preferred name")
    print("=" * 60 + "\n")

    agent.print_response(
        "Actually, I'd prefer you call me Jordan from now on.",
        user_id=user_id,
        session_id="session_3",
        stream=True,
    )
    agent.learning_machine.user_profile_store.print(user_id=user_id)