Microsoft Agent Framework
App : A product you use directly
Microsoft open-source framework for .NET and Python, successor to Semantic Kernel and AutoGen: agents, graph-based workflows, A2A and MCP.
Version 1.0 stable since April 2026, with long-term support: the reference for .NET teams and the Azure ecosystem. Semantic Kernel and AutoGen move to maintenance mode.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Building AI agents that move from prototype to production is complex: it requires managing multi-agent orchestration, workflow durability, compatibility with multiple model providers, and observability. Microsoft Agent Framework (MAF) addresses these needs in a single open-source SDK under the MIT license.
The framework provides a unified abstraction for creating agents in Python, .NET, or Go, connecting them to model providers (Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, Google Gemini, Ollama), and composing them into graph-based workflows: sequential, parallel, handoff, or group chat. Workflows support checkpointing, restart after interruption, and human-in-the-loop approvals.
MAF targets development teams building agent systems intended for production, who need advanced orchestration and want the freedom to switch model providers without major rewrites. Integration with the Azure and Microsoft Foundry ecosystem is built in.
This framework is not suited for people looking for a no-code graphical interface, or for those who simply want to test a language model on a one-off basis. It requires Python or .NET development skills and an understanding of agent and workflow concepts.
What it offers
- ·Python, .NET, and Go support with consistent APIs across languages.
- ·Native connectors for Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, Google Gemini, and Ollama.
- ·Graph-based workflow engine supporting sequential, parallel, handoff, and group chat patterns, with checkpointing and restart.
- ·Configurable middleware to intercept and transform agent behavior at every stage (content filters, logging, compliance policies).
- ·Pluggable memory and context: conversational history, persistent key-value state, vector-based retrieval via Foundry Agent Service, Mem0, Redis, Neo4j, or a custom store.
- ·Declarative agent and workflow definition in YAML for versioning and rapid deployment.
- ·Support for A2A (Agent-to-Agent) and MCP (Model Context Protocol) protocols for cross-framework collaboration.
- ·Built-in observability via OpenTelemetry for distributed tracing, monitoring, and debugging.
In practice
- Price
- Free
- Commercial use
- Commercial use: yes
- Available on
- cli
- Licence
- MIT
- FR interface
- No
Security and data
- Rating
- Security: one caveat, see the card
- Your data
- Local or cloud depending on use
- Maintenance
- Active
- Last release
- 15 October 2026
Links and repository checked on 18 October 2026 · added on 27 September 2026
How to install Microsoft Agent Framework
pip install agent-framework