17-conversation-management

Conversation Management — Controlling Agent Message History

This tutorial shows how to use conversation managers to control how an agent's message history grows, shrinks, and stays within model context limits. You'll learn the three built-in strategies: sliding window (default), null (no management), and summarizing.

Tutorial Details

Information	Details
Strands Features	SlidingWindowConversationManager, NullConversationManager, SummarizingConversationManager
Agent Pattern	Single agent with different conversation management strategies
Tools	Custom mock tools for demonstrating tool-pair preservation and truncation
Model	Amazon Nova Lite on Amazon Bedrock

How It Works

1. SlidingWindowConversationManager (default) keeps the last N messages, preserving tool-use/tool-result pairs and optionally truncating large tool outputs
2. NullConversationManager does nothing — full history is sent every time, and raises an error on overflow
3. SummarizingConversationManager summarizes old messages instead of dropping them, preserving context while staying within limits

Prerequisites

• Python 3.10 or later
• AWS account with Amazon Bedrock access configured
• Basic familiarity with Strands Agents — see 01-first-agent if needed

Tutorial Structure

17-conversation-management/
├── README.md
├── requirements.txt
└── conversation-management.ipynb

File	Description
conversation-management.ipynb	Step-by-step notebook covering all three conversation managers and their configuration options

What You'll Learn

• Default behavior: how SlidingWindowConversationManager works out of the box with window_size=40
• Tuning window_size: choosing the right size based on your use case
• Tool-pair preservation: how trimming keeps toolUse/toolResult pairs together
• should_truncate_results: truncating large tool outputs before dropping messages
• per_turn parameter: proactive management during tool-heavy agent loops
• NullConversationManager: disabling management and handling overflow manually
• SummarizingConversationManager: summarizing old messages instead of dropping them

Installation

Install the required dependencies:

pip install -r requirements.txt

Then open conversation-management.ipynb from the 17-conversation-management/ directory.

Related Tutorials

• 01-first-agent — creating agents (uses default conversation manager)
• 06-memory — persisting agent memory across sessions
• 02-tools-and-mcp — tools that generate large outputs (where truncation matters)