.NET Multi-Agent Orchestration Framework

A lightweight, enterprise-ready multi-agent orchestration framework designed for the .NET ecosystem.

Python-centric tools like CrewAI or LangChain are great, but many enterprise systems run on .NET. This framework fills that gap by providing native, idiomatic C# abstractions for building autonomous agents, equipping them with tools, and orchestrating complex workflows.

Features

• Native .NET: Built targeting .NET 8/9 with minimal external dependencies.
• Agent Abstraction: Define agents with distinct personas (Role, Goal, Backstory) and give them specific tools to accomplish their objectives.
• Extensible Orchestration: Chain agents together via SequentialWorkflow to pass conversational context, memory, and state smoothly from one agent to the next.
• LLM Agnostic: The ILlmProvider interface makes it trivial to swap backends (OpenAI, Semantic Kernel, local Ollama) by simply changing the provider injection.
• RAG Ready: Built-in simple IVectorStore interfaces with an InMemoryVectorStore example to get you started with Retrieval-Augmented Generation immediately.

Getting Started

1. Define your LLM Provider

Implement ILlmProvider or use the included OpenAILlmProvider.

var llmProvider = new OpenAILlmProvider("sk-your-openai-api-key");

2. Equip Agents with Tools

Tools implement the ITool interface. Give your agents access to search, APIs, or databases.

var vectorStore = new InMemoryVectorStore(); // Load with data
var searchTool = new VectorSearchTool(vectorStore);

3. Create Agents

Assign roles and tools to your agents.

var researcherAgent = new Agent(
    name: "Research Agent",
    role: "Senior AI Researcher",
    goal: "Gather information about AI orchestration frameworks.",
    backstory: "You are an expert in AI frameworks and provide summaries for developers.",
    llmProvider: llmProvider,
    tools: new[] { searchTool }
);

4. Orchestrate

Link agents into a workflow.

var workflow = new SequentialWorkflow()
    .AddAgent(researcherAgent)
    .AddAgent(developerAgent)
    .AddAgent(reviewerAgent);

var context = new DefaultOrchestrationContext();
context.AddMessage(new ChatMessage(ChatRole.User, "Build an Agent class."));

var finalContext = await workflow.RunAsync(context);

Publishing

All four library projects (AgentFramework.Core, .Memory, .Orchestration, .Providers.OpenAI) carry full NuGet metadata and are packed/pushed as a set by .github/workflows/publish-nuget.yml whenever a v* tag is pushed:

git tag v0.1.0
git push origin v0.1.0

Requires a NUGET_API_KEY repo secret with push rights on nuget.org. AgentFramework.Examples is intentionally excluded — it's a console demo, not a library.

Contributing

We welcome contributions! Please see our Contributing Guide for more details.

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Machine-readable repository context for AI agents and crawlers lives in llms.txt.