Bring Agentic Coding On-Device with Foundry Local
Microsoft, Thursday, November 5th, 2026
Foundry Local runs Qwen3.8-27B on Nvidia GPUs and connects to GitHub Copilot for private, offline agentic coding.
Microsoft's Foundry Local brings agentic coding to developer hardware by running Qwen3.8-27B locally and connecting it to GitHub Copilot through Bring Your Own Key.
Local inference keeps code on the machine, makes costs predictable and works offline.
Powered by ONNX Runtime, Foundry Local optimizes prefill, decoding, memory, long-context processing and concurrent requests on Nvidia GPUs.
In tests on an RTX 5090 and DGX Spark using real GitHub issues, it delivered higher generation speed than llama.cpp at every prompt length, with the gap widening beyond 131K tokens, and it requires a GPU with at least 24 GB of VRAM and Foundry Local CLI 0.13.0.