Microsoft released its Agent Framework as open-source on GitHub on April 30, providing a comprehensive multi-language framework for building, orchestrating, and deploying AI agents. The framework supports both Python and .NET implementations, with installation as simple as pip install agent-framework or dotnet add package Microsoft.Agents.AI.
What the Framework Includes
Agent Framework offers two primary capability layers, according to Microsoft’s documentation. Individual agents use LLMs to process inputs, call tools and MCP servers, and generate responses. Graph-based workflows connect agents and deterministic functions for multi-step tasks with type-safe routing, checkpointing, streaming, and human-in-the-loop support.
The framework supports multiple model providers out of the box: Microsoft Foundry, Anthropic, Azure OpenAI, OpenAI, and Ollama. It includes foundational building blocks for chat completions, agent session state management, context providers for memory, middleware for intercepting agent actions, and MCP clients for tool integration.
Migration and Consolidation
The release includes official migration guides from both Semantic Kernel and AutoGen, signaling that Microsoft is consolidating its fragmented agent tooling into a single framework. Semantic Kernel launched in 2023 as Microsoft’s SDK for LLM integration, while AutoGen emerged from Microsoft Research as a multi-agent conversation framework. Agent Framework absorbs capabilities from both.
Positioning Against Closed Ecosystems
The open-source release positions Microsoft against OpenAI’s increasingly proprietary agent ecosystem and Google’s Gemini platform. By supporting multiple LLM providers and offering framework-agnostic tooling, Microsoft is betting that developer adoption scales faster through open infrastructure than through vendor lock-in. The multi-language support (Python and .NET) targets both the AI research community and enterprise .NET shops, a combination no other major agent framework covers natively.
The framework launched alongside a 30-minute introduction video, weekly public community office hours, and a Discord channel, suggesting Microsoft is treating this as a community-driven project rather than a top-down enterprise offering.
The Consolidation Signal
Microsoft now has one canonical answer to “how do I build agents on Microsoft infrastructure,” replacing a landscape where developers had to choose between Semantic Kernel for simple LLM calls, AutoGen for multi-agent patterns, and Azure AI Services for deployment. The migration guides make the intent explicit: this is the framework going forward.
For teams already building on Semantic Kernel or AutoGen, the migration paths reduce the risk of adopting a new framework. For teams evaluating agent infrastructure for the first time, the multi-provider support and open-source licensing remove the primary objection to building on Microsoft tooling.