Run the Right AI Model for the Right Copilot Task — Local or Cloud!

This article is part 1 of the CopilotHarness series, demonstrating how to use local AI models for simple coding tasks like renaming variables or writing docstrings instead of relying on cloud-based models. It explains the Bring Your Own Model (BYOK) mechanism in VS Code, providing step-by-step instructions to configure three minimal proxies: one for Ollama (local), one for Foundry Local (offline/NPU), and one for Azure OpenAI (cloud). Each proxy is a single ASP.NET Core Minimal API file that translates OpenAI-compatible requests to local backends. The article also covers how to unwrap Copilot's message envelope to extract user requests correctly. Aimed at developers, it emphasizes running models locally for speed, privacy, and cost savings.