Microsoft Unveils Windows AI Foundry: A Leap Forward in Local AI Development
Microsoft has taken significant strides in enhancing AI capabilities on Windows with the introduction of Windows AI Foundry, announced at Build 2025. Leveraging the existing Windows Copilot Runtime, this new unified platform facilitates local AI app development, integrating various AI functionalities through the use of Windows AI APIs and machine learning models that operate seamlessly in the background on Copilot+ PCs.
Introducing Windows AI Foundry
Windows AI Foundry merges Windows Copilot Runtime with a suite of innovative tools designed for developers. This powerful platform not only provides out-of-the-box AI APIs backed by Microsoft’s native AI models but also equips developers with tools to customize these models. Furthermore, it enables the integration of open-source models sourced from Azure AI Foundry, alongside an inference runtime that allows developers to incorporate their own models.
Flexible AI Model Integration
The introduction of Windows AI Foundry reflects Microsoft’s commitment to versatility in AI model usage. Developers have access to a diverse range of models through Azure Foundry Local and other prominent catalogs such as Ollama and NVIDIA NIMs. Microsoft’s proprietary Foundry Local catalog ensures that AI models are optimized for performance across various hardware configurations, including CPUs, GPUs, and NPUs. A simple winget install Microsoft. FoundryLocal command allows developers to explore, download, and assess models compatible with their devices. Following model selection, integrating Foundry Local into applications is achieved effortlessly via the Foundry Local SDK.
Efficient Model Deployment with Windows ML
At the heart of this local AI initiative is Windows ML, a built-in inferencing runtime that streamlines the deployment of machine learning models across various hardware platforms. Built on DirectML, Windows ML supports chipset architectures from major providers, including AMD, Intel, NVIDIA, and Qualcomm. This infrastructure allows developers to focus on creating cutting-edge applications without the hassle of future silicon updates, as Windows ML is designed to manage dependencies effectively and adjust to new hardware requirements automatically.
Enhanced Fine-Tuning with LoRA Support
In a notable advancement, Microsoft also highlighted its new support for LoRA (Low-Rank Adaptation) within the Phi Silica model framework. This feature permits developers to fine-tune a minimal subset of model parameters using custom datasets, thereby enhancing performance on specific tasks. Currently in public preview alongside Windows App SDK 1.8 Experimental 2, LoRA will soon be accessible on both Intel and AMD Copilot+ PCs, enhancing its utility across platforms.
Revolutionizing Search with Semantic Search APIs
To further enrich the developer experience, Microsoft announced new Semantic Search APIs that empower the creation of advanced AI-driven search functionalities within applications. These APIs facilitate local execution and support the RAG (Retrieval-Augmented Generation) framework, making it easier for developers to enhance their application’s search capabilities. Currently, these APIs are in private preview and available for all Copilot+ PCs.
For further details and updates on this development, visit the official source.
Related Articles:
Microsoft Weekly: OneDrive challenges, Start menu insights, and additional updates
9:56
Microsoft releases Defender antivirus update for latest Windows 11 and 10 ISOs
21:51
Microsoft Copilot Testing Features That Could Provide Free Access to ChatGPT’s o4-mini-high
18:30
Leave a Reply ▼