The Azure AI Agent Service supersedes the classic OpenAI Assistants API, providing a production-ready platform for building, managing, and deploying AI agents.
| Feature | Description | Impact |
|---|---|---|
| Hosted Agents (April 2026) | Persistent-state, VM-isolated agent compute with scale-to-zero | Agents resume with filesystem and session identity intact across restarts. Built-in versioning and VNet support for production workloads. |
| Toolbox (Public Preview) | Unified MCP-based tool management across frameworks | Configure and manage tools once, use across Agent Framework, LangGraph, and any MCP-compatible client. |
| Memory Service | Managed long-term memory store (Preview) | Agents can persist and retrieve context across multiple sessions seamlessly without custom DBs. |
| Foundry MCP Server | Cloud-hosted Model Context Protocol | Connect to cloud resources directly from IDEs (like VS Code) without local process management. |
| Voice Live | Native speech-to-speech runtime | Allows agents to converse in real-time with barge-in support. |