Foundry Local

What if your AI assistant ran entirely on your device but still had access to up-to-date Microsoft docs?

I built a small app that combines Foundry Local (on-device inference with phi-4-mini) and Learn MCP Server (documentation retrieval). You ask a question, the local model calls a search tool, gets doc content from Learn, and synthesizes a grounded answer. No API keys, no cloud LLM costs.

User Question → Foundry Local (phi-4-mini) → Learn MCP Server → Grounded Answer

It’s adapted from the Foundry Local tutorial-tool-calling sample, replacing the weather tool with a docs search tool. The key design choice: tool_choice: required, which means the model always searches before answering.

Repo

Discussion

Thanks for reading! :-)