---
title: "Microsoft Agent Framework adds native cross-session memory"
slug: "microsoft-agent-framework-adds-native-cross-session-memory"
published: "2026-09-27"
beat: "Tools"
tags: ["Tools", "Launches"]
creator: "Agentry Newsroom"
editor: "Susanne Sperling, Editor — Human in the Loop"
tools: ["Claude (Anthropic)", "Perplexity Sonar"]
creativeWorkStatus: "verified"
dateReviewed: "2026-09-27"
aiActArticle50: "compliant"
humanView: "https://agentry.news/launches/microsoft-agent-framework-adds-native-cross-session-memory"
agentView: "https://agentry.news/agent/microsoft-agent-framework-adds-native-cross-session-memory"
---# Microsoft Agent Framework adds native cross-session memory

> Microsoft announced a preview integration on September 4, 2026, that equips Agent Framework agents with durable, cross-session memory backed by Azure Cosmos DB. The new capability allows agents to sto

*Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. [AI policy](/ai-policy).*

## Microsoft ships cross-session memory for Agent Framework

Microsoft announced a preview integration on September 4, 2026, that gives Agent Framework agents **durable, cross-session memory backed by Azure Cosmos DB** [Microsoft Developer Blog](https://devblogs.microsoft.com/agent-framework/native-memory-for-microsoft-agent-framework-with-azure-cosmos-db/). The update addresses a core limitation in agentic systems: the inability to maintain and retrieve context across separate user sessions without rebuilding state from scratch.

The new Python package `agent-framework-azure-cosmos-memory` provides a `CosmosMemoryContextProvider` that developers can integrate into their Agent Framework deployments. The component stores conversation turns in Azure Cosmos DB and performs extraction and recall of relevant facts, summaries, and user profiles when an agent resumes interaction in a later session [Microsoft Developer Blog](https://devblogs.microsoft.com/agent-framework/native-memory-for-microsoft-agent-framework-with-azure-cosmos-db/).

## How the memory system works

The integration operates as a persistence layer built directly into Microsoft's agent execution runtime. When an agent completes a conversation, the `CosmosMemoryContextProvider` serializes and stores dialogue history and derived semantic knowledge—facts and user profiles extracted from the conversation—in Cosmos DB's document store. On subsequent sessions, the provider retrieves and surfaces relevant memories to the agent, reducing cold-start latency and maintaining conversational continuity.

This approach differs from naive context-window expansion. Rather than appending full conversation logs to every new prompt, the system performs semantic filtering: only memories rated as relevant to the current user query or task are loaded into the agent's working context. This keeps token usage and latency manageable as session counts grow.

## Developer availability and scope

At the time of publication, the preview is available for Python developers using Microsoft Agent Framework. The announcement specifies this as a **preview integration**, meaning the API and behavior may change before general availability. Developers can adopt it immediately for experimental deployments and provide feedback to the Microsoft team.

The timing reflects broader industry movement toward stateful agents. Prior agent architectures treated each request as isolated; maintaining user context required manual session management or external database wiring. By embedding memory into the framework itself, Microsoft reduces integration friction and standardizes how agents should retain learned information about users and past interactions.

The Azure Cosmos DB backend provides geo-distributed storage, consistency tuning, and autoscale throughput—important for production agent deployments that must handle variable load and comply with data residency requirements.

## What's next

The preview does not announce specific timelines for expanded language support, availability beyond Python, or production SLA commitments. Enterprise deployments will depend on how feedback shapes the feature set and Microsoft's roadmap priorities for agent framework hardening.