OpenAI’s annual conference for third-party developers, DevDay, kicked off with a bang today, as the co-founder of i CEO Sam Altman announced new “Apps SDK”, or a development kit that allows you to “build applications inside ChatGPT” including paid applications for which companies can charge to exploit OpenAI’s recently announced Agentic Commerce Protocol (ACP).
In other words, instead of launching apps individually on your phone, computer, or web, now you can do it all without whenever you leave ChatGPT.
This feature allows the user to log into their accounts on external apps and bring all their information back to ChatGPT and exploit the app in much the same way they would outside the chatbot, but now with the ability to ask ChatGPT to perform specific actions, analyze content, or go beyond what each app can offer on its own.
You can instruct Canva to create slides based on text description, ask Zillow for listings of homes in a specific area that meet specific requirements, or query Coursera for the content of a specific lesson while playing a video, all from within ChatGPT – many other apps also already offer their own connections (see below).
“This will create a new generation of applications that are interactive, adaptive, personalized, and that you can talk to,” Altman said.
Even though the Apps SDK is already in preview, Openai said it won’t start accepting fresh apps under ChatGPT or allow them to charge users “until the end of this year.”
Online access to the ChatGPT app is already available to ChatGPT Free, Plus, Go and Pro users – for now only outside the European Union – with Business, Enterprise and Education tiers expected to gain access to the app later this year.
Built on the common MCP standard
Built on the open-source Model Context Protocol (MCP) standard introduced by competitor Anthropic almost a year ago, the Apps SDK enables third-party developers working independently or on behalf of enterprises huge and petite to connect selected data, “trigger actions, and render a fully interactive user interface [user interface]Altman explained during his opening keynote address.
The Apps SDK includes a “talk to apps” feature that allows ChatGPT and the underlying GPT-5 or other “O-series” models to pilot it underneath to get updated context from a third-party app or service, so the model “always knows exactly what the user is interacting with,” according to fellow presenter and OpenAI engineer Alexi Christakis.
Developers can create applications that:
-
to appear built-in in chat as light cards or carousels
-
expand to full screen for engaging tasks such as maps, menus or slides
-
use picture in picture for live sessions such as videos, games or quizzes
Each mode is designed to maintain ChatGPT’s minimal conversational flow while enhancing interactivity and brand presence.
Early integrations with Coursera, Canva, Zillow and more…
Christakis showed early integration external applications built based on the application SDKincluding from an e-learning company Kursracloud design software company Canvaand search engine for real estate offers and agent connections, Zillow.
Altman also announced Apps SDK integration with additional partners not officially demonstrated during the keynote, including: Reservation.com, Expedia, Figma AND Spotify and in the documentation it says there are more upcoming partners on board: AllTrails, Peloton, OpenTable, Target, theFork and Uberrepresenting lifestyle, commerce and productivity categories.
The Coursera demo provides an example of how a user joins an external application, including a new application login screen (Coursera) that appears in the ChatGPT chat interface, activated simply by a text prompt from the user asking: “Coursera, can you teach me something about machine learning”?
Once logged in, the app launches in an “in-line” chat interface and can render anything from the Internet, including interactive elements such as video.
Christakis explained and demonstrated that the Apps SDK also supports picture-in-picture and full-screen views, allowing the user to choose how to interact with it.
Playing the displayed Coursera video showed that it automatically pinned the video to the top of the screen so that the user could watch it even while still viewing the text dialog from ChatGPT in the typical I/O prompts and responses below.
Users can then query ChatGPT about the content that appears in the video, without specifying exactly what was saidbecause the Agent SDK transmits information from the back-end, server-side, from the connected application to the underlying ChatGPT AI model. So asking “can you explain more about what they’re saying now” will automatically display the relevant part of the video and make it available to the underlying AI model so it can analyze and respond via text.
In another example, Christakis opened an older, existing ChatGPT conversation he was having about his sibling’s dog walking business and resumed the conversation by asking another third-party app: Canvato generate a poster using one of ChatGPT’s recommended “Walk This Wag” business names, along with detailed instructions on font selection (“sans serif”) and overall color scheme and style (“brilliant and colorful”).
Instead of the user having to manually add all of these specific elements to the Canva template, ChatGPT issued commands and performed background actions on the user’s behalf.
After a few minutes, ChatGPT responded with several poster designs generated directly in Canva, but displayed them all in the user’s ChatGPT chat session, where they could view, preview, enlarge, and provide feedback or request corrections.
Christakis then asked ChatGPT to turn one of the slides into an entire deck of slides so that the founders of the dog walking company could present it to investors, which it did in the background for several minutes while he presented the final integrated application, Zillow.
He started a new chat session and asked a simple question: “Based on our conversations, what city would be a good place to expand my dog walking business?”
Using ChatGPT’s optional memory feature, he referenced a conversation while walking the dog and suggested Pittsburgh, which Christakis took as an opportunity to type “Zillow” and “show me some homes for sale there,” which brought up an interactive Zillow map of homes for sale and prices and hover animations, all within ChatGPT.
Clicking on a specific home also opened a full-screen view with “most Zillow features” entirely without leaving ChatGPT, including the ability to request home tours and contact agents and filter by bedrooms and other features such as outdoor space. ChatGPT retrieves the requested Zillow filtered search and also provides an in-line text response explaining what it did and why.
The user can then ask additional questions about the specific property, such as “how close is the dog park?” – or compare them to other properties, all within ChatGPT.
You can also use the app in conjunction with the search feature, searching the Internet to compare information about the app (in this case Zillow) with other sources.
Security, privacy and programming standards
OpenAI highlighted this Apps must meet stringent privacy, security and content requirements standards be listed in the ChatGPT directory. Applications must:
-
serve a a clear and valuable goal
-
To be predictable and reliable in behavior
-
To be safe for the general publicincluding youth aged 13–17
-
respect users’ privacy and limit data collection to only what is necessary
Each application must also contain a file clear, published privacy policyobtain user consent before connecting and identify any actions that modify external data (e.g. posting, uploading, uploading).
Apps that violate OpenAI’s terms of use, frequently crash or misrepresent their capabilities may be removed at any time. Developers must submit from verified accountsprovide contacts with customer serviceand maintain your applications for stability and compatibility.
OpenAI has also been published developer design guidelinesdescribing how applications should look, sound and behave. They must follow ChatGPT’s visual system – including consistent color palettes, typography, spacing, and iconography – and adhere to accessibility standards such as alternative text and readable contrast ratios.
Partners can display brand logos and accent colors, but cannot change the basic ChatGPT interface or use promotional language. Apps should stay “conversational, wise, basic, responsive and accessible” in accordance with the documentation.
A new ecosystem of conversational apps
By opening ChatGPT to third-party apps and payments, OpenAI is taking an important step towards transforming ChatGPT from a chatbot to a full-fledged AI operating system — one that combines conversational intelligence, rich interfaces, and built-in commerce.
For developers, this means direct access to over 800 million ChatGPT userswho can discover apps “at the right time” through natural conversation – whether they’re planning trips, studying or shopping.
This is what it means for users a fresh generation of apps you can talk to — where a single interface helps you book a flight, design a slide presentation, or learn a fresh skill without leaving ChatGPT.
As OpenAI put it: “This is just the beginning of applications in ChatGPT, providing new tools for users and new opportunities for developers.”
Some essential questions remain, namely: 1. What happens to all the data from third-party applications when they connect to ChatGPT and its users… Does OpenAI access it and train on it? 2. What is happening to the once very popular OpenAI GPT Store, which has historically been promoted as a way for third-party developers and developers to create custom, task-specific versions of ChatGPT and monetize them through a usage-based revenue sharing model?
We have asked the company about both issues and will update you when we hear back.