Build Personal AI Agents on Windows PCs with New Tools from Microsoft and NVIDIA
NVIDIA Technical, Tuesday, June 2nd, 2026
Microsoft and NVIDIA introduce security tools and hardware for building secure, on-device AI agents on Windows with 2x faster inference.
NVIDIA and Microsoft have unveiled new tools at COMPUTEX 2026 to enable developers to build secure AI agents on Windows PCs, including Microsoft eXecution Containers (MXC) for sandbox security and NVIDIA OpenShell for runtime integration.
The NVIDIA RTX Spark product family delivers 1 petaflop of AI performance with up to 128 GB memory, while Microsoft offers a special developer edition with preloaded tools. Updates to NVIDIA NemoClaw, Hermes Agent, and H Company's Holo 3.1 models expand agent capabilities across Windows and Linux, with improvements including 2x faster inference through Multi-Token Prediction and Programmatic Dependent Launch techniques.
Enhanced multi-GPU support in llama.cpp and ComfyUI enables RTX PC users to access greater memory capacity and compute performance for running larger models locally.