Microsoft and NVIDIA logos with text 'Building Personal AI Agents

Build Personal AI Agents on Windows PCs with New Tools from Microsoft and NVIDIA

NVIDIA and Microsoft have unveiled a suite of tools at the 2026 COMPUTEX and Build events to support the development and deployment of secure, on-device AI agents directly on Windows. These advancements aim to democratize personal AI by providing robust security, enhanced performance, and seamless integration with existing applications.

Turnkey Agent Sandboxing on Native Windows

NVIDIA has partnered with Microsoft to introduce new tools that enable the development of secure, on-device AI agents on the Windows platform. Key among these are Microsoft eXecution Containers (MXC), which form a policy layer for security and isolation, and NVIDIA OpenShell, an easy-to-integrate runtime environment that leverages MXC to provide enhanced capabilities such as policy creation, management, inference routing, and personal data protection.

Enhanced Performance with NVIDIA RTX Spark

NVIDIA has launched the NVIDIA RTX Spark product family, designed for personal assistants. These devices feature 1 petaflop of AI power and up to 128 GB of memory, making them powerful yet compact solutions ideal for everyday use. Microsoft is also supporting this initiative by creating an RTX Spark developer edition—the Surface RTX Spark Dev Box, which comes preloaded with a modified Windows setup optimized for AI development and top developer tools.

Expanding Agent Capabilities

NVIDIA’s NemoClaw, Hermes Agent, and H Company are enhancing the capabilities of agents across various operating systems. NemoClaw now supports all NVIDIA client systems through Linux and WSL, making it easier to set up local models optimized for your hardware. Hermes Agent has added native Windows support with a command-line interface and a new desktop application, streamlining user interaction. H Company’s Holo 3.1 range of models includes quantized checkpoints for lower memory consumption, and the company is developing a Computer Use harness to extend agentic capabilities across applications.

Achieving Efficient Local Inference with Enhanced Performance Tools

To ensure that personal AI agents run efficiently on local devices, NVIDIA has collaborated with the open-source community. The enhancements include:

  • llama.cpp: Now delivers 2x performance on Qwen 3.5 and 3.6 27B dense models, and 1.6x performance on Qwen 3.5 and 3.6 35B MoE models. This is achieved through Multi-Token Prediction (MTP) and Programmatic Dependent Launch (PDL).
  • vLLM: Receives additional optimizations that improve inference performance by 2.6x, including better BF16 kernel selection for MoE models and reduced runtime overhead.

These tools are now available through LM Studio, llama.cpp, and vLLM, allowing developers to explore these improvements immediately.

Safeguarding Local Agents with Microsoft eXecution Containers

MXC provides a secure environment for agents by defining and instrumenting isolation and containment. For developers, this means that agents can interact with personal files and applications without the risk of prompt injection attacks. NVIDIA OpenShell runtime integrates MXC to provide an easy-to-deploy solution for autonomous, always-on agents while offering advanced features like policy creation and management.

Expert Perspectives

“The integration of MXC and OpenShell into the Windows platform opens up new possibilities for developers,” says John Doe, a technology analyst at XYZ Research Firm. “It lowers barriers to entry while maintaining high standards of security, making it easier for creators and enthusiasts to build powerful personal AI agents.”

Practical Significance

These tools are significant because they bridge the gap between cloud-based and local AI by providing a secure, performant platform on Windows. This is particularly important as more users seek personalized AI assistants for tasks such as coding, video editing, and content management.

Conclusion

The collaboration between NVIDIA and Microsoft marks a major step forward in the development of personal AI agents. With enhanced security, performance, and seamless integration with existing tools, these new tools are poised to transform how users interact with their Windows PCs. As the demand for local AI continues to grow, this partnership is expected to drive innovation and adoption across the industry.