At OpenAI’s Developer Day in San Francisco on Monday, CEO Sam Altman revealed a way to run third-party applications directly inside ChatGPT’s chat window, an effort to shift the product toward a platform that can host other services and tools.
Altman described the move as one that would "allow a new generation of apps that are adaptive, interactive, and personalized, that you can chat with." Beginning immediately, some developers can access a preview of a new apps software development kit (SDK) that lets them build apps inside ChatGPT using open standards. Distribution of those apps will be limited at first to a small set of large partners.
The company demonstrated how the feature works with integrations from Spotify, Canva, and Zillow appearing inside a chat and responding to typed requests. In a staged demo, OpenAI software engineer Alexi Christakis opened a conversation aimed at the Canva integration and asked it to design posters for a dog-walking business. He then requested a pitch deck to help raise capital for the venture.
Christakis followed by asking the chat for a city that would be a good place to expand the business; the assistant suggested Pittsburgh. He then invoked a Zillow app inside the same chat and asked it to display homes for sale in that city. An interactive map popped up, showing listings; Christakis narrowed the search to three-bedroom homes with yards and the results updated inside the chat thread.
Altman said OpenAI plans to give developers new ways to generate revenue from ChatGPT apps, including built-in purchasing flows. He added, "Soon we’ll offer an agentic commerce protocol, with instant checkout from right within ChatGPT."
OpenAI has experimented with ways to extend ChatGPT before. Two years ago the company offered tools for building custom widgets and GPTs, and when the GPT Store launched in January 2024 the company said developers had created more than 3 million custom GPTs. Those earlier widgets and GPTs drew developer interest but did not become a mainstream user habit.
The company did not disclose any revenue-sharing arrangements with the partners it showcased onstage. The new SDK announcement signals a deeper push to work with established enterprises and app makers while keeping users inside ChatGPT instead of sending them to separate web pages or mobile apps. In other words, OpenAI is attempting to combine aspects of web browsing and app-based experiences into a chat-first interface that acts like an operating layer.
Nick Turley, OpenAI’s head of product for ChatGPT, told reporters after the keynote that the team always intended to build more than a simple conversational bot. He said the company "never meant to build a chatbot; we meant to build a super assistant, and we got a little sidetracked." Turley added, "Will people spend all of their time in ChatGPT? I don’t think so," and suggested a typical day might begin in ChatGPT before the assistant steers users toward other apps and websites.
Beyond transforming how existing applications appear to users, OpenAI wants to sit at the center of work on autonomous agents that can complete tasks on people’s behalf. The company launched several agent-building tools, including AgentKit, a drag-and-drop interface designed to help creators assemble more advanced AI-driven capabilities.
Capturing developer mind-share also means providing coding tools. At Monday’s event OpenAI said Codex, a model optimized to write code, will leave research preview and become generally available. New Codex features include the ability to ask questions about code and apply edits via Slack messages, an SDK for the Codex model, and analytics tools that let organizations track internal Codex usage.
OpenAI faces competition for developer attention from Google, Anthropic, and Amazon. Competition has intensified as players such as Meta and DeepSeek release capable open source models that let developers fine-tune top-tier systems without paying for access to OpenAI’s APIs.
This year the company released an open source model, GPT-OSS, and it made its flagship model, GPT-5, available through the regular API. In September OpenAI upgraded its developer-facing coding model, a step likely aimed at countering strong coding models from Anthropic and Google. The company introduced agent-building toolsets in March.
At last year’s Developer Day OpenAI announced a low-latency voice interaction API, new options for fine-tuning vision models, and a tool that lets creators fine-tune small models using larger, more capable ones from OpenAI.
OpenAI is expanding its physical infrastructure while questions swirl about the long-term economics of generative AI. On Monday the company revealed a major agreement to acquire 6 gigawatts’ worth of AMD chips, a deal that could include taking roughly a 10 percent stake in the chipmaker.
During the keynote Altman argued that the company is already helping developers move faster and increase productivity. He said, "This is the best time in history to be a builder," and added, "It has never been faster to go from idea to product. You can really feel the acceleration at this point."