AI ML and AI APIs

AI-first index for ML, model inference, and app AI integration APIs.

ML / AI Discovery Map

Index map for two related but separate areas:

computer-vision ML: YOLO, model inference, datasets, and model-backed image search;
app AI integration: chat sessions, model profiles, tool plugins, retrieval, artifact store, MCP hosting, Responses API transport, and AI chat UI.

ML search is useful when target appearance varies or classes matter.
CV capture still provides frames.
Model inference returns predictions/rectangles/classes/scores.
Screen overlay can display model detections.
General AI chat/agent services are separate from CV model inference.
AI chat sessions are host/runtime features. A script, component, plugin, or service must obtain dependencies from its current host context; these APIs are not ambient C# globals.
AI tool plugins are normal .NET objects derived from AiKernelPlugin. [KernelFunction] methods become callable tools after registration.

MLSearchTrigger - aura trigger for model-backed search.
IComputerVisionExperimentalScriptingApi / ICvAccessor - script CV/ML entry points where available.
YOLO/model classes - model loading and inference anchors.
AiEngine - creates and tracks chat, MCP, and coding-agent sessions.
IAiChatSession / AiChatSession - chat transcript, active profile, plugin host, tool policy, retrieval, artifact store, and conversation engine.
IAiProfileRepository / AiModelProfile - configured model profiles, providers, endpoints, API-key references, function-calling flags, reasoning effort, and context-window metadata.
AiChatViewModel / AiChatComponent - Blazor-facing chat state and UI.
AiDocsKnowledgeBasePlugin / IBm25KnowledgeBase - local markdown docs indexing and doc_* tools.
AiKernelPlugin - base class for tool plugins.
AiMcpSession / AiMcpServerHost - expose registered plugins as an MCP server.

Prefer deterministic CV first when a template/color/OCR task is stable.
Prefer ML when object class or appearance variability is central.
Prefer osd/screen-overlay.md for prediction boxes.
Prefer ml/ai-runtime.md for chat/session/profile/UI integration.
Prefer ml/ai-kernel-plugins.md when adding tools callable by an AI model.
Prefer ml/script-ai-clients.md when deciding between the built-in EyeAuras.AISdk / EyeAuras.AI path and an intentional direct provider NuGet escape hatch.
Prefer a recipe when the task is an architecture question rather than a single API lookup.

Avoid model inference for simple fixed-template searches.
Avoid mixing chat/agent APIs with CV model inference unless the task needs it.
Avoid assuming AI runtime services are available in plain helper classes.
Avoid registering tools without clear descriptions, cancellation behavior, and duplicate-name checks.

MLSearchTrigger, ImageSearchTrigger, ICvAccessor, Yolo, AiEngine, IAiSession, IAiChatSession, AiChatSession, IAiChatSessionHandle, AiChatViewModel, AiChatComponent, IAiProfileRepository, AiModelProfile, AiKernelPlugin, AiDocsKnowledgeBasePlugin, IBm25KnowledgeBase, AiMcpSession, AiMcpServerHost.