This example demonstrates how to configure agents to search through previous sessions and limit the number of historical sessions included in context. This helps manage context length while maintaining relevant conversation history.

​
Code

cookbook/agents/state/last_n_session_messages.py
import os

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.models.openai import OpenAIChat

# Remove the tmp db file before running the script
os.remove("tmp/data.db")

agent = Agent(
    model=OpenAIChat(id="gpt-5-mini"),
    user_id="user_1",
    db=SqliteDb(db_file="tmp/data.db"),
    add_history_to_context=True,
    num_history_runs=3,
    search_session_history=True,  # allow searching previous sessions
    num_history_sessions=2,  # only include the last 2 sessions in the search to avoid context length issues
)

session_1_id = "session_1_id"
session_2_id = "session_2_id"
session_3_id = "session_3_id"
session_4_id = "session_4_id"
session_5_id = "session_5_id"

agent.print_response("What is the capital of South Africa?", session_id=session_1_id)
agent.print_response("What is the capital of China?", session_id=session_2_id)
agent.print_response("What is the capital of France?", session_id=session_3_id)
agent.print_response("What is the capital of Japan?", session_id=session_4_id)
agent.print_response(
    "What did I discuss in my previous conversations?", session_id=session_5_id
)  # It should only include the last 2 sessions

​
Usage

1

Create a virtual environment

Open the Terminal and create a python virtual environment.
python3 -m venv .venv
source .venv/bin/activate
2

Install libraries

pip install -U agno openai
3

Run Agent

python cookbook/agents/state/last_n_session_messages.py