On this page
Microsoft Agent Framework: agent building features and setup
Microsoft's development framework for agents and multi-agent workflows. It provides orchestration, checkpoints for resuming work, and execution observability.
Tool by the Otomatzia editorial team · source content checked on
Overview
Microsoft Agent Framework is the framework Microsoft currently recommends to new users of AutoGen and Semantic Kernel. It supports patterns such as sequential execution, parallel execution, and task handoffs between agents. For Python and .NET teams, it provides a foundation for integrating models and tools while retaining the ability to monitor and intervene. It is development infrastructure to integrate into an application, not a customer service system that can be launched without implementation work.
Features
- Graph-based agent orchestration with sequential execution, parallelism, and task handoffs.
- Checkpoints for resuming work, result streaming, and human approval.
- Support for different model providers and OpenTelemetry integration.
Business uses
- An enterprise agent that combines several information tools with an approval process.
- Migrating an existing system based on AutoGen to the current framework.
Getting started
- Choose the language and provider setup that fits the existing system.
- Set up a small agent example and add one test tool.
- Check resuming after a pause, execution logs, and permissions before connecting to real systems.
What to check
- When migrating from an earlier library, use the migration guide and recheck behavior, not just package names.
- An orchestration mechanism does not replace user authentication or limits on what each tool can do.
Sources to check next
Vendor features and prices change. Confirm details on the linked official pages before you rely on them.