AI Building Blocks for .NET: Add intelligence to your C# Apps

TL;DR

• 1

  One app, many AI building blocks: Bruno opens with a real support center app that analyzes incidents, generates images, explains its feedback, and coordinates a Blazor UI, a Microsoft Agent Framework agent, and an Nvidia NeMo agent through Aspire traces.

• 2

  Microsoft.Extensions.AI is the foundation: The core abstraction is IChatClient, which lets the same C# code talk to Azure AI Foundry, other hosted providers, or local runtimes like Ollama on port 11434, with Microsoft recommending Entra ID and CLI credentials over API keys.

• 3

  RAG is presented as a pipeline, not magic: Bruno breaks retrieval into reader, chunker, enricher, embeddings, and vector store steps, showing both a toy movie search example and a more realistic markdown ingestion pipeline with semantic chunking, summaries, and a SQLite vector store.

• 4

  MCP gives agents live access to external tools: Using the Microsoft Learn MCP server at learn.microsoft.com/mcp, the model fails to answer the latest Microsoft Agent Framework version on its own, then succeeds once tool calling is enabled and MCP tools are passed into chat options.

• 5

  Agents in .NET are thin wrappers with useful structure: In Microsoft Agent Framework, an agent is basically a chat client plus name, description, instructions, and optional tools, and Bruno shows how quickly that grows into workflows like writer-editor chains, MCP-backed agents, and image-generation agents.

• 6

  Interoperability is a big part of the story: The support app connects a Microsoft agent and an Nvidia NeMo agent through A2A, with Bruno calling out agent cards at /.well-known/agent-card.json and noting A2A's Google origins and current Linux Foundation stewardship.