OpenAI opens ChatGPT app submissions and ships a chat-native UI library

What OpenAI actually opened up

OpenAI is now accepting app submissions for review and publication inside ChatGPT, following the apps preview introduced at DevDay earlier in the year. Developers submit through the OpenAI Developer Platform, where they can track approval status. The first approved apps roll out gradually in the new year.

Submissions carry a specific payload: MCP connectivity details, testing guidelines, directory metadata, and country availability settings. That list tells you the shape of the platform — apps connect over MCP, get reviewed against quality and safety bars, and land in a directory that users reach from the tools menu or at chatgpt.com/apps.

Once connected, an app can be triggered by @ mention or picked from the tools menu. OpenAI says it is also experimenting with surfacing relevant apps mid-conversation using signals like conversational context, usage patterns, and user preferences — meaning discovery isn't only user-initiated.

The frontend is a chat-native interface, not a page

The most concrete signal for frontend teams is what OpenAI shipped to help build apps: best-practice guidance on what makes a great ChatGPT app, open-source example apps, a step-by-step quickstart, and — notably — an open-sourced UI library for chat-native interfaces.

A dedicated UI library matters because building inside a conversation is not the same as building a web app. There is no full viewport to own, no navigation chrome, no persistent layout. The interface has to render as a coherent unit inside a message stream and stay legible when the user's attention is on the dialogue, not your component.

OpenAI's own framing reinforces this. It says the strongest apps are 'tightly scoped, intuitive in chat,' and either complete a real-world workflow that starts in conversation or enable a fully AI-native experience inside ChatGPT. That's a design constraint, not marketing: a sprawling multi-screen app doesn't fit the surface.

Transactions still leave the conversation

Monetization is deliberately limited at this stage. Developers can link out from a ChatGPT app to their own website or native app to complete transactions for physical goods. OpenAI says it is exploring additional options over time, including digital goods.

So the chat-native interface handles intent and context, but the checkout still happens on your own frontend. For teams, that means the ChatGPT app and the existing web or mobn surface both have to work — and the handoff between them, via deep links into the app directory or out to a transaction page, becomes its own design problem.

The connection prompt is now part of the UX you own

OpenAI puts data disclosure and control directly in the connection flow. When a user connects to a new app, ChatGPT discloses what types of data may be shared with the third party and surfaces the app's privacy policy for review. Users can disconnect at any time, at which point the app immediately loses access.

We require developers to only request the information needed to make their apps work.Montana Labs

The specific implication for frontend teams: the moment a user decides to trust your app happens inside ChatGPT's disclosure screen, not on a page you designed. The scope of permissions you request, and the privacy policy you attach, are now the first interface a user sees — before your chat-native UI ever renders. Requesting less isn't just a compliance line; it's the conversion step that decides whether anyone connects at all.