Skip to main content
This guide walks you through setting up an agent as a Learning Machine.

​
Prerequisites

  • Python 3.9+
  • A database (PostgreSQL recommended for production)
  • OpenAI API key (or another model provider)

​
Setup

1

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate
2

Install dependencies

uv pip install -U agno openai sqlalchemy psycopg
3

Run PostgreSQL with PgVector

Install Docker Desktop and run:
docker run -d \
  -e POSTGRES_DB=ai \
  -e POSTGRES_USER=ai \
  -e POSTGRES_PASSWORD=ai \
  -e PGDATA=/var/lib/postgresql/data/pgdata \
  -v pgvolume:/var/lib/postgresql/data \
  -p 5532:5432 \
  --name pgvector \
  agnohq/pgvector:16
4

Set your API key

export OPENAI_API_KEY=sk-***

​
Your First Learning Agent

Create a file called learning_agent.py:
learning_agent.py
from agno.agent import Agent
from agno.db.postgres import PostgresDb
from agno.models.openai import OpenAIResponses

# Connect to your database
db = PostgresDb(db_url="postgresql+psycopg://ai:ai@localhost:5532/ai")

# Create an agent with learning enabled
agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=True,
    markdown=True,
)

# Session 1: Share some information
print("=== Session 1 ===")
agent.print_response(
    "Hi! I'm Sarah, I work at Acme Corp as a data scientist. I prefer detailed explanations.",
    user_id="sarah@acme.com",
    session_id="session_1",
)

# Session 2: The agent remembers
print("\n=== Session 2 ===")
agent.print_response(
    "What do you know about me?",
    user_id="sarah@acme.com",
    session_id="session_2",
)
Run it: 
python learning_agent.py
The agent will remember Sarah’s name across sessions.

​
What Just Happened?

With learning=True, the agent:
  1. Extracted profile info from Session 1 (name: Sarah, preferred name if mentioned)
  2. Recalled this profile in Session 2 to answer accurately
Note: learning=True only enables user profile extraction. For user memory, session context, and other stores, use explicit LearningMachine configuration.

​
Customizing What Gets Learned

You can control learning through your choice of learning stores and learning modes. Depending upon which stores are enabled for your agent, and the choice of default learning mode for that store or a selected mode, your agent behavior can be optimized for specific use cases. 
from agno.learn import LearningMachine

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=LearningMachine(
        user_profile=True,      # Structured fields (name, preferences)
        user_memory=True,       # Unstructured observations
        session_context=True,   # Session summary and goals
        entity_memory=False,    # Knowledge about external entities
        learned_knowledge=False # Reusable insights (requires Knowledge base)
    ),
)

​
Using SQLite (Development)

For quick development without PostgreSQL: 
from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.models.openai import OpenAIResponses

db = SqliteDb(db_file="learning.db")

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=True,
)

​
Next Steps

Learning Stores

Deep dive into Learning Stores.

Learning Modes

Control how and when learning happens.