# Windows Copilot Runtime Microsoft's on-device AI substrate for Windows 11 — an OS-level set of APIs and built-in models that apps can use without managing inference infrastructure. The Windows analog of [[Apple Intelligence]]. Tightly coupled to "Copilot+ PC" hardware certification. Announced May 2024 at Build, shipping progressively since. ## Components - **Phi Silica** — Microsoft's small on-device language model (variant of [[Phi LLMs]]), designed for [[Neural Processing Unit (NPU)]] inference - **Direct ML** — low-level ML runtime for Windows; one execution provider for [[ONNX Runtime Web]] - **Windows Studio Effects** — camera ML (background blur, eye contact, framing) - **Recall** — semantic search across screen history (controversial, opt-in) - **Live Captions, Translation, Voice Clarity** — system-level audio AI - **Cocreator, Image Creator** — generative image tools ## Copilot+ PC Hardware Threshold Microsoft requires: - 40+ TOPS NPU - 16+ GB RAM - 256+ GB storage This drove the industry to standardize NPU performance levels across Qualcomm Snapdragon X, Intel Core Ultra, and AMD Ryzen AI silicon. ## Developer APIs - **Windows Copilot Library** — high-level APIs for common AI tasks (text generation, summarization, image effects) - **Windows ML / DirectML** — lower-level model execution - **AI Toolkit for Visual Studio** — model selection, fine-tuning, deployment ## Relationship to Web Standards Edge ships with experimental [[Prompt API]] and [[Writing Assistance APIs]] support that can route to Phi Silica via the Copilot Runtime — a parallel to how Chrome routes to [[Gemini Nano]]. This is how the W3C standards meet Microsoft's on-device model strategy on Windows. ## Trade-offs **Strengths:** OS-level integration, NPU optimization, free for developers, privacy by locality **Limitations:** Windows-only, requires Copilot+ hardware tier, Microsoft-controlled model lifecycle ## References - https://learn.microsoft.com/en-us/windows/ai/ ## Related - [[Apple Intelligence]] - [[Neural Processing Unit (NPU)]] - [[On-Device Machine Learning]] - [[Edge AI]] - [[Browser-Provided Language Models]] - [[Prompt API]] - [[Writing Assistance APIs]] - [[Gemini Nano]] - [[Phi LLMs]] - [[ONNX Runtime Web]] - [[Small Language Models (SLMs)]]