Microsoft Agent Framework

4 months ago

Microsoft Agent Framework is an open-source development kit for building AI agents and multi-agent workflows for .NET and Python.
It combines strengths from Semantic Kernel and AutoGen projects, adding new capabilities.
Two primary capabilities: AI agents (individual agents using LLMs) and workflows (graph-based multi-agent tasks).
AI agents support model providers like Azure OpenAI, OpenAI, and Azure AI.
Workflows support type-based routing, nesting, checkpointing, and human-in-the-loop scenarios.
Agent Framework is the successor to Semantic Kernel and AutoGen, merging their features and introducing new ones.
Common AI agent use cases: customer support, education/tutoring, code generation, and research assistance.
AI agents are not suitable for highly structured tasks with predefined rules.
Workflows solve complex, multi-step tasks involving multiple agents and external systems.
Key workflow benefits: modularity, agent integration, type safety, flexible flow, and checkpointing.
Agent Framework is currently in public preview, with feedback welcomed via GitHub.
Warning: Using third-party servers/agents carries risks; data compliance is the user's responsibility.
Installation available for Python (`pip install agent-framework --pre`) and .NET (`dotnet add package Microsoft.Agents.AI`).

Hasty Briefsbeta