关于 windows-ai
面向 Windows AI/ML 开发者的参考技能,按任务索引 Windows AI Foundry 文档,涵盖设备端 AI、模型推理、Agent 集成及 NPU 分析。
主要功能
- 01按开发任务导航参考文档
- 02涵盖内置 AI API 与本地推理
- 03整理 Foundry Local 与 MCP 集成
- 04提供 DirectML 和 NPU 分析资料
适用场景
- 01查找 Windows ML 的 ONNX 推理资料
- 02选用 Windows 内置 AI API
- 03实现 MCP 服务或 App Actions
- 04检查 NPU 要求并分析推理性能
MCP 相关
包含 MCP 服务实现、注册与测试,以及 Agent 和应用的集成参考。
安装
复制下面的命令,在终端中运行。
npx skills add https:/ / github.com/ fandhe-ai/ agent-reference-skills --skill windows-ai安装后在 Claude Code 中输入/windows-ai确认已加载。
包含的文件
除 SKILL.md 外,安装时一起下载的参考资料和脚本。
references/ai-apis/ai-dev-gallery.mdapp-content-search-tutorial.mdchoose-your-windows-ai-solution.mdcontent-moderation.mdcopilot-plus-developer-guide.mddevice-requirements.mdfaq-cloud-vs-local.mdfaq-using-ai.mdfoundry-toolkit-vscode.mdget-started.mdimage-description-generator.mdimage-foreground-extractor.mdimage-generator.mdimage-object-extractor.mdimage-object-remover.mdimage-scaler.mdimaging-overview.mdimaging-tutorial.mdlanguage-model-best-practices.mdlanguage-model-context.mdlanguage-model-options.mdlanguage-model.mdlocal-llms.mdphi-silica-lora.mdphi-silica-structured-output.mdphi-silica-transparency-note.mdphi-silica-tutorial-maui.mdphi-silica-tutorial-winui.mdplatform-card-click-to-do.mdplatform-card-image-ai-apis.md详细介绍
Windows AI 参考
围绕 Windows AI Foundry,汇总 Windows 上 AI/ML 开发功能的官方文档。 涵盖内置 AI API、本地 ONNX 推理(Windows ML)、本地模型托管(Foundry Local)、 Agent 集成(MCP / App Actions)、低级 GPU 推理(DirectML)、NPU 硬件性能分析(Copilot+ PC / WPR / WPA)。 根据用户的任务阅读相应的 README.md,并从中查找具体文件。
目标版本:MS Learn 获取日期 2026-10-16〜17、Windows ML 2.4.89 (ORT 1.27.1) / Windows ML Runtime API 2.7.2021-experimental (ORT 1.30.0)。
目录结构
skills/windows-ai/
SKILL.md
references/
ai-apis/
README.md
ai-dev-gallery.md
platform-card-click-to-do.md
platform-card-paint.md
platform-card-recall.md
platform-card-windows-studio-effects.md
faq-cloud-vs-local.md
choose-your-windows-ai-solution.md
content-moderation.md
copilot-plus-developer-guide.md
device-requirements.md
faq-using-ai.md
foundry-toolkit-vscode.md
phi-silica-structured-output.md
image-description-generator.md
image-foreground-extractor.md
image-generator.md
image-object-extractor.md
image-object-remover.md
image-scaler.md
language-model-best-practices.md
language-model.md
language-model-context.md
language-model-options.md
phi-silica-lora.md
platform-card-image-ai-apis.md
platform-card-phi-silica.md
platform-card-ocr.md
local-llms.md
recognized-text.md
responsible-ai.md
semantic-search.md
speech-recognition-model.md
text-intelligence-skills.md
text-recognizer.md
video-scaler.md
troubleshooting.md
get-started.md
phi-silica-tutorial-maui.md
phi-silica-tutorial-winui.md
text-recognition-tutorial.md
app-content-search-tutorial.md
imaging-tutorial.md
imaging-overview.md
phi-silica-transparency-note.md
windows-ml/
README.md
runtime-overview.md
runtime-get-started.md
runtime-concepts.md
runtime-tutorial.md
runtime-text-generation.md
runtime-speech-recognition.md
multiple-onnx-versions.md
efficiency-tips.md
model-distribution.md
api-reference.md
bring-your-own-eps.md
check-ep-versions.md
deployment-bootstrap.md
eps-vs-bring-your-own.md
execution-provider-catalog.md
execution-provider-errors.md
execution-providers-overview.md
get-started.md
install-execution-providers.md
learning-model-binding.md
learning-model-evaluation-result.md
learning-model-session.md
learning-model.md
legacy-windows-machine-learning.md
logs.md
migrate-to-windows-ml.md
model-catalog-source-schema.md
model-catalog.md
model-compilation.md
model-conversion.md
models.md
onnx-runtime-inference.md
onnx-versions.md
overview.md
register-execution-providers.md
run-genai-onnx-models.md
samples.md
select-execution-providers.md
supported-execution-providers.md
tutorial.md
update-eps.md
webgpu-ep-experimental.md
winml-cli.md
foundry-local/
README.md
architecture.md
cache-management.md
installation-and-cli.md
model-catalog.md
overview.md
rest-api.md
sdk.md
winml-package.md
get-started.md
winui-local-ai-tutorial.md
mcp-app-actions/
README.md
mcp-overview.md
mcp-server-overview.md
mcp-windows-identity.md
mcp-mcpb.md
mcp-manual.md
mcp-containment.md
test-mcp-server.md
quickstart-mcp-host.md
file-connector.md
odr-tool.md
app-actions-overview.md
actions-get-started.md
actions-uri-launch.md
actions-iactionprovider-manual.md
actions-filter-caller.md
actions-streaming-text.md
actions-remote-files.md
actions-availability.md
actions-display-ui.md
actions-json.md
actions-provider-manifest.md
actions-test-tool.md
actions-consume.md
actions-rai-safety.md
agent-launchers-overview.md
agents-get-started.md
agents-json.md
agent-workspace.md
security-consent-model.md
directml/
README.md
api-reference-constants.md
api-reference-enumerations.md
api-reference-functions.md
api-reference-interfaces.md
api-reference-structures.md
cuda-on-wsl2.md
debug-layer.md
dmlcreatedevice.md
directml-overview.md
directml-tools.md
directmlx.md
errors-and-device-removal.md
feature-level-history.md
fused-operators.md
gpu-accelerated-training.md
graphs.md
helper-functions.md
idmlbindingtable.md
idmlcommandrecorder.md
idmlcompiledoperator.md
idmloperator.md
idmloperatorinitializer.md
onnxruntime-directml.md
operators.md
programming-guide.md
pytorch-directml.md
resource-lifetime-sync.md
sample-applications.md
strides-padding-layout.md
tensors.md
tensorflow-directml.md
tensorflow-directml-faq.md
uav-barriers.md
version-history.md
webnn-overview.md
webnn-tutorial.md
npu-devices/
README.md
byom-model-sources.md
gpuview-npu.md
npu-hardware-overview.md
onnxruntime-etw-tracing.md
task-manager-npu.md
wpr-wpa-npu-profiling.md
samples/
README.md
first-windows-ai-api-call.md
foundry-local-quickstart.md
compile-onnx-model-on-first-run.md
windows-ml-runtime-api-run-model.md
scripts/
README.md
install.md
diagnostics.md
探索步骤
根据任务查找类别,并在类别的 README.md 中确定目标页面:
- 在下方映射表中查找与任务对应的类别
- 参阅该类别的
references/<category>/README.md,确定目标页面 - Read 相应的
.md文件,查看详细信息
任务 → 类别映射
| 任务 | 类别 | 参考 README |
|---|---|---|
| 想通过 LanguageModel 调用设备端生成式 AI(Phi Silica) | ai-apis | references/ai-apis/README.md |
| 想使用 TextRecognizer / ImageScaler / ImageObjectExtractor / ImageDescriptionGenerator 实现图像处理和 OCR | ai-apis | references/ai-apis/README.md |
| 想应用 SemanticSearch / ContentModeration / 负责任 AI 指南 | ai-apis | references/ai-apis/README.md |
| 想通过 Copilot+ PC 开发指南、AI Dev Gallery 和 Foundry Toolkit 选择并试制 AI 功能 | ai-apis | references/ai-apis/README.md |
| 想使用 Windows ML 在本地运行 ONNX 模型推理 | windows-ml | references/windows-ml/README.md |
| 想选择、注册、安装 execution providers 并进行故障排查 | windows-ml | references/windows-ml/README.md |
| 想使用 LearningModel / LearningModelSession 绑定并评估模型 | windows-ml | references/windows-ml/README.md |
| 想使用 Windows ML 的 CLI、诊断日志和 WebGPU EP | windows-ml | references/windows-ml/README.md |
| 想搭建 Windows AI API 开发环境、进行首次调用,并完成 Phi Silica / OCR / Imaging / Semantic Search 教程 | ai-apis | references/ai-apis/README.md |
| 想通过 Windows ML Runtime API(experimental,C++/Python)显式控制管道、阶段、张量和 GGUF (llama.cpp) | windows-ml | references/windows-ml/README.md |
| 想使用自有 LM / Whisper 模型进行 Text Generation / Speech Recognition (experimental) | windows-ml | references/windows-ml/README.md |
| 想设计模型分发(随应用打包 / 单独下载 / 应用间共享)、推理优化以及多个 ORT 版本共存 | windows-ml | references/windows-ml/README.md |
| 想将 Foundry Local 集成到 WinUI / C# 应用中 | foundry-local | references/foundry-local/README.md |
| 想使用 Foundry Local 在本地主机运行 LLM | foundry-local | references/foundry-local/README.md |
| 想使用 Foundry Local 的 CLI / REST API / SDK / model catalog / 缓存管理 | foundry-local | references/foundry-local/README.md |
| 想实现、注册和测试 MCP 服务 | mcp-app-actions | references/mcp-app-actions/README.md |
| 想通过 App Actions / IActionProvider 向外部公开应用功能 | mcp-app-actions | references/mcp-app-actions/README.md |
| 想实现 Agent Launchers / Agent Workspace / 安全同意模型 | mcp-app-actions | references/mcp-app-actions/README.md |
| 想通过 DirectML 直接控制 GPU 推理设备、运算符和张量 | directml | references/directml/README.md |
| 想从 ONNX Runtime / PyTorch / TensorFlow 使用 DirectML 后端 | directml | references/directml/README.md |
| 想通过浏览器或应用调用 WebNN 的 GPU/NPU 推理 | directml | references/directml/README.md |
| 想确认 Copilot+ PC / NPU 硬件要求 | npu-devices | references/npu-devices/README.md |
| 想通过 Task Manager / WPR / WPA / GPUView / ETW 测量 NPU 使用率和推理情况 | npu-devices | references/npu-devices/README.md |
| 想为 NPU 获取并量化自有 ONNX 模型 (BYOM) | npu-devices | references/npu-devices/README.md |
| 想了解典型用法(首次 API 调用、Foundry Local、首次启动时编译模型、运行 Runtime API) | samples | samples/README.md |
| 想了解安装和诊断命令 | scripts | scripts/README.md |
分工说明:Windows ML Runtime 的 Speech Recognition / Text Generation(导入自有模型,experimental)与 Windows AI APIs 的 SpeechRecognitionModel / LanguageModel(OS 提供的模型)是不同的功能。 このスキルは Windows AI Foundry 配下の AI/ML 機能(組み込み AI API、ローカル推論、エージェント連携、DirectML、NPU ハードウェア・プロファイリング)のみを扱う。 WinUI コントロールやレイアウト・データバインディングなど UI 実装は windows-winui-controls, windows-winui-ui, windows-app-sdk が担当し、パッケージング・配布は windows-packaging-publish、グラフィックス・メディア描画は windows-graphics-media が担当する。