Understanding agent memory

Note:

This feature is currently available in preview mode only for AWS and IBM Cloud deployments.

Agent memory enables AI agents to retain, recall, and use information about you across chat sessions, creating more personalized and efficient interactions over time.

When agent memory is enabled, the AI agent learns from your conversations and remembers key details such as your preferences, work context, and frequently used information. This capability allows the agent to provide more relevant responses without requiring you to repeat information in every conversation.

How agent memory works

Agent memory operates by extracting and storing relevant information from your conversations with AI agents. The system identifies and retains facts that can improve future interactions, such as:

Personal details (name, role, location)
Work preferences and context
Frequently requested information
Communication style preferences

The agent uses this stored information to:

Provide contextually relevant responses
Reduce the need for repetitive explanations
Improve response accuracy over time
Personalize interactions based on your preferences

Agent memory requires configuration at both the builder and user levels:

Agent memory: Builders enable the agent's capability to store and use memories by setting the memory_enabled parameter to true.
User memory: Users control whether their conversations contribute to memory by enabling the Use memories in conversations toggle in the Orchestrate Chat interface.

Agent memory must be enabled first to give the agent the capability to save memory
User memory must then be enabled by each user to activate memory for their conversations

Prerequisites

To use agent memory, the following requirements must be met:

The builder or administrator must set the memory_enabled parameter for the agent to true.
You must enable memory in the chat interface using the Use memories in conversations toggle.
Agent memory is supported only for native agents that use gpt-oss as the AI model.

Getting started with agent memory

Follow these topics to understand, enable, and use agent memory effectively:

Managing agent memory: Set up memory for the first time with step-by-step instructions. Learn the prerequisites, how to enable memory at both the agent and user levels, and verify that memory is working correctly.
Use memory in conversations: Discover practical examples and best practices for using agent memory effectively. Learn how to instruct the agent to remember or forget information, and see real-world scenarios showing memory in action.
Memory behavior and limits: Understand what information is stored, how memory is scoped, and the governance policies that protect your data. Learn about privacy controls, data retention, and compliance standards.
Retention policy: Learn how long memory is kept, how the Least Recently Used (LRU) strategy works, and what happens when you disable memory. Understand the 30-day idle interval and how memory items are automatically removed.
Frequently asked questions: Find quick answers to common questions about agent memory functionality, behavior, and management.