Microsoft has gone all-out on AI in its annual Build developer conference, and now, the company lifted the veil off Windows AI Foundry, a new platform replacing the former Copilot Runtime.
“At Microsoft, we believe the future of AI is being built right now across the cloud, on the edge and on Windows. Windows is, and will remain, an open platform that empowers developers to do their best work, offering the ultimate flexibility,” the company announced in an official blog post. “Windows AI Foundry, an evolution of Windows Copilot Runtime, offers a unified and reliable platform supporting the AI developer lifecycle from model selection, optimization, fine-tuning and deployment across client and cloud. Windows AI Foundry includes several capabilities,” it added.
Windows AI Foundry comes with the ability to run diverse AI models directly on user machines, including open-source options and Microsoft’s own inbox models. The platform supports third-party contributions from firms such as Nvidia, notably with the inclusion of NVIDIA NIM (Nvidia Inference Microservices). These models are fully optimized to operate across different hardware configurations, including CPUs, GPUs, and dedicated NPUs (Neural Processing Units) found in Microsoft’s Copilot+ PCs.
Windows AI Foundry includes a suite of APIs that offer immediate access to commonly used AI functions, including text intelligence, object detection, image description, and text recognition. These features are powered by Microsoft’s integrated models on Copilot+ devices and are aimed at supporting both general-purpose and domain-specific applications. Another of the standout features is support for LoRA (Low-Rank Adaptation), which allows developers to fine-tune Microsoft’s inbox model, Phi Silica, using custom datasets. This customization capability is particularly relevant for organizations aiming to adapt AI to niche or proprietary use cases without requiring large-scale retraining.
To enable richer user interactions, Microsoft has also introduced APIs for semantic search and knowledge retrieval. These functions support the implementation of retrieval-augmented generation (RAG) workflows, enabling applications to perform context-sensitive queries using both structured and unstructured data sources.
Microsoft is also simplifying local deployment through Foundry Local, a module that can be installed via a simple terminal command. Developers can leverage command-line tools (CLI) and software development kits (SDKs) to test, debug, and deploy AI models from their local environments. These tools will be fully integrated into Windows 11 and the Windows App SDK, promoting native support for AI-driven applications across the Windows ecosystem. The integration of Windows ML further enhances deployment efficiency by removing the need for developers to manually include ML runtimes or hardware drivers in their application packages.
The success of Windows AI Foundry will depend in part on Microsoft’s partnerships with major hardware vendors. The company said it is actively working with AMD, Intel, Nvidia, and Qualcomm to ensure seamless model performance across a variety of chipsets and devices. These collaborations are expected to help standardize AI deployment on Windows, making the platform attractive not only for individual developers but also for enterprise software teams.
