Microsoft explains how to run Qwen3.8-27B locally with Foundry Local and GitHub Copilot
AIMicrosoft explains how to connect Foundry Local to GitHub Copilot using Bring Your Own Key so developers can run Qwen3.8-27B on an NVIDIA GPU with at least 24 GB of VRAM. The required setup is Foundry Local CLI 0.13.0 or later, with CUDA and cuDNN runtimes installed separately. In its comparison with llama.cpp on an RTX 5090 and a DGX Spark, Foundry Local produced faster generation at every tested prompt length, with speedups from 1.36x to 1.95x in the chart.
Source details
Microsoft Foundry Blog · Baiju Meswani, Emma Ning, Tianlei Wu · devblogs.microsoft.comPublished
Original title: Bring Agentic Coding On-Device with Foundry Local