Articles
How to connect Visual Studio 2026 Github Copilot to models running on llama.cpp
GitHub Copilot's Ollama provider expects endpoints and metadata that llama.cpp's own server doesn't implement. Here's how LLM Router bridges that gap so Visual Studio 2026's Copilot Chat can drive a fully local llama.cpp model.
Integrating Umbraco with Local LLMs via llama.cpp and Umbraco.AI
A practical guide to connecting Umbraco v18's AI Copilot to a local llama.cpp model — either directly, or (recommended) through LLM Router, which turns the fiddly --alias/--jinja flags into two preset fields and eliminates the previous_response_id error entirely.
Keeping Intel GPU VRAM Alive — A SYCL Keepalive Solution for Local AI on Windows
On Windows, headless Intel GPUs unload their VRAM after 60–70 seconds of inactivity — a major problem when using the GPU for local AI with llama.cpp SYCL. This article presents a tiny C++ keepalive program that prevents the driver from unloading your loaded model.