Lectures

In this lecture, you will build a “world map” of the ecosystem: what a ChatGPT App is, what it consists of (UI widget + MCP tools + context), how it differs from Legacy Plugins, Custom GPTs, Assistants API, and “just” the OpenAI API. We will examine sandbox limitations, the role of the Apps SDK, its place in the Store, and a special focus on commerce via ACP, as well as fix common thinking mistakes that lead teams to choose the wrong tool for the job.

In this lecture we assemble the ChatGPT Apps “layer cake” into a single mental model: what exactly the ChatGPT UI does, where your widget lives on the Apps SDK (Next.js 16 + React 19), why you need an MCP server and the tool-call protocol ( call_tool), when the Agents SDK appears for orchestration, and how the bottom layer (ACP/commerce/payments) turns an app into a product. We will cover an end-to-end scenario, deployment variants (monolith vs split), and typical mistakes that break the architecture and complicate Store review.

We break down the mental model of ChatGPT Apps: how a user expresses intent in text, and the model itself decides whether to answer “on its own” or to connect an app, call its tools, and (if needed) display a widget. We’ll understand the difference between explicit and implicit launch, what Discovery is, why tool metadata (fields name, description, inputSchema) is the real API for the model, and how semantic routing picks the right tool (for example, suggest_gifts) instead of “magic” functions like run_func.

We examine the lifecycle of a ChatGPT App from the platform’s point of view: how connecting via Developer Mode (a personal sandbox with a dev-URL) differs from publishing to the Store (review, listing, stable prod-URL), what states exist ( Draft, Under review, Published, Paused), how to think about dev/ staging/ production environments, and why releases in the Store are closer to the App Store model than to “we just pushed a website.”

In this lecture we build an engineering mental model of ChatGPT Apps constraints: how the widget sandbox works (why you can’t rely on window like on the open web and why fetch isn’t “to anywhere”), how the permissions model works through securitySchemes declarations and just-in-time consent, and how usage/privacy policies shape UX and architecture (thin UI + MCP tools, minimizing PII, honest descriptions, and a transparent openExternal).

Step-by-step instructions for installing the JavaRush plugin and solving course tasks.