.NET AI and Machine Learning

The .NET team's skill for adding AI and machine learning to .NET 8+ apps. It picks the right technology for the task first (classic ML, an LLM, an agent, local inference or vector search), then delivers only what you asked for: a plan when you ask for a plan, code when you ask for code. What you can do with it - Add classification, regression, clustering, anomaly detection or recommendations with ML.NET, trained with a fixed seed and evaluated on a held-out set - Add text generation, summarization or reasoning through Microsoft.Extensions.AI's IChatClient, with retries, timeouts and keys kept out of your code - Build agentic workflows with tool calling on Microsoft Agent Framework, with an iteration limit and a cost ceiling - Build RAG pipelines: document ingestion and chunking, cached embeddings and vector search through Microsoft.Extensions.VectorData, with providers such as Azure AI Search, Qdrant, Postgres (pgvector) or Redis - Run pre-trained models with ONNX Runtime or local models through Ollama (OllamaSharp), or build GitHub Copilot extensions with the Copilot SDK What it steers you away from - Using an LLM for tabular predictions that ML.NET handles faster and more cheaply - LLM calls without retries or timeouts, and API keys committed in appsettings.json - Hand-rolled agent loops, the archived Accord.NET, and defaulting to Semantic Kernel for new work without a reason How to install it In Claude Code or GitHub Copilot CLI, run /plugin marketplace add dotnet/skills, then /plugin install dotnet-ai@dotnet-agent-skills. The dotnet/skills marketplace also works in VS Code (preview), Cursor and Codex. Good to know - It targets .NET 8 or later, not .NET Framework. - It isn't meant for pure data engineering, or for training custom deep-learning models: for those it points you to Python, then ONNX Runtime for inference in .NET.

• “Add a support ticket classifier to this .NET Web API, trained on our CSV, with evaluation metrics and a prediction endpoint.”

  Copy

• “Add a document summarization endpoint to this .NET 10 Web API.”

  Copy

• “Flag customers likely to churn from this CSV, and tell me which technology you chose and why.”

  Copy

• “Build a .NET console app with an agent that researches a topic using web search and note-taking tools.”

  Copy

• “Plan a Blazor chatbot that answers customers from our product PDFs, with Postgres for storage.”

  Copy