Production agents need to remember. In this episode, we dive into Memory in Foundry Agent Service — a managed, long-term memory layer that turns stateless LLM calls into stateful, continuous agents. Lewis Liu and Amy Boyd walk through how memory is natively integrated with Microsoft Agent Framework and LangGraph, enabling agents to persist user preferences, conversation context, and task outcomes across sessions — with zero infrastructure overhead. See how per-user memory scoping, automatic memory extraction, and CRUD APIs give developers full control while keeping enterprise governance built in.

✅ Chapters:
00:17 Introduction MAF - Microsoft Agent Framework and Foundry Memory
00:45 Why we need Foundry Memory
01:40 What's Foundry Memory
03:20 How Memory helps Agents - Use Cases
07:31 Foundry Memory Key concepts and Features
14:33 Foundry Memory - What's under development
16:55 Foundry Memory and MAF - Demo 1
20:25 Foundry Memory and MAF - Demo 2
24:24 Foundry Memory and MAF - Demo 3
30:50 Foundry Memory Public Preview and What's Next

✅ Resources:
Foundry Toolkit for VS Code: https://aka.ms/foundrytk
Microsoft Agent Framework on GitHub: https://aka.ms/AgentFramework
AI Agents for Beginners: https://aka.ms/ai-agents-beginners
Blog post: https://aka.ms/DeployingAgents-blog
MAF 1.0 Announcement: https://aka.ms/AgentFramework1.0-blog

📌 Let's connect:
Jorge Arteiro | https://www.linkedin.com/in/jorgearteiro
Amy Boyd | https://www.linkedin.com/in/amykatenicho/
Lewis Liu | https://www.linkedin.com/in/lewisxl/


Subscribe to the Open at Microsoft: https://aka.ms/OpenAtMicrosoft

Open at Microsoft Playlist: https://aka.ms/OpenAtMicrosoftPlaylist

📝Submit Your OSS Project for Open at Microsoft https://aka.ms/OpenAtMsCFP

New episode on Tuesdays!

Microsoft Developer

Microsoft Agent Framework (MAF) has reached general availability and now integrates natively with Azure AI Foundry's managed memory service (currently in public preview). Foundry memory provides a secure, long-term memory store for AI agents, supporting two memory types: conversation summaries and user preferences. Developers can configure memory scopes (per-user/per-agent, shared across agents, team-level, or multi-store) to control data isolation. The integration requires only a few lines of code via a context provider pattern. Key features include automatic conflict detection and resolution in stored facts, TTL-based memory decay, and upcoming procedural memory for workflow improvement. A live demo showcases a customer support scenario where agents greet users by name, recall past issues, and hand off context between a concierge and resolution agent without requiring users to repeat information.

Using Microsoft Agent Framework with Foundry managed memory