Run AI Locally: Foundry Local + Microsoft Agent Framework | Udaiappa Ramachandran

In this session, Udaiappa Ramachandran (CTO/CSO at Akumina Inc. & Microsoft MVP in AI) walks you through Microsoft Foundry Local — the production-grade on-device AI runtime — and shows how to pair it with the Microsoft Agent Framework (MAF) to build real, privacy-first AI agents that run entirely on your device. You'll learn why on-device AI matters for privacy, latency, cost, and offline use cases, and how Foundry Local compares to Ollama and LM Studio. We cover the Foundry Local architecture including ONNX Runtime, WinML, NPU support, and the model catalog, followed by installation, CLI usage, and SDK setup in both Python and C#. From there we dive into the Microsoft Agent Framework — how it unifies Semantic Kernel and AutoGen — and connect MAF agents to Foundry Local with real working code in both languages. You'll also see tool calling (function calling) on local models like Phi-4-mini, a full live demo from CLI to SDK to agent to tools, and finally how to promote your local agent to Foundry Hosted Agents on Azure with zero code changes. Whether you're building a field app, a code assistant, a healthcare tool, or just want to cut cloud inference costs, this talk gives you a clear, hands-on path from local dev to production. Resources mentioned: Foundry Local docs: learn.microsoft.com/azure/foundry-local GitHub: github.com/microsoft/Foundry-Local Agent Framework: learn.microsoft.com/agent-framework MAF + Foundry Local: aka.ms/maf-foundry-local Presented at the Nashua Cloud .NET User Group