ChatGPT gets a feature to build, host, and publish sites and apps right inside the chat
OpenAI has documented ChatGPT Sites, a feature that takes you from conversation to published site without leaving the chat, with access control, login, and a custom domain. What's missing is what runs under the hood.
From conversation to published site
OpenAI has documented ChatGPT Sites, a feature that lets any user describe an idea for a site, app, or game right inside ChatGPT and walk away with a published address. The screenshots illustrating the feature's official page carry metadata from September 2026, which indicates recently updated documentation.
The flow is conversational: you ask ChatGPT to build something, or mention @Sites, describe what you want to do, and refine the result by exchanging messages. You can attach files and visual references to the initial request, according to the feature's official page. When the result looks good, you just publish and share the link.
One detail matters to those who already use OpenAI's coding tools: you can start a Site from an existing Codex project. If you already have a codebase generated or maintained with Codex, OpenAI's coding agent, ChatGPT Sites can inherit that project instead of starting from scratch. It's OpenAI stitching together two ends of its own catalog: the agent that writes code and the layer that hosts and publishes the result.
What it's for in practice
The documentation itself sums up the feature as something to build for work, your own business, or personal use, and cites as examples:
- Team dashboards, with an example of an internal panel used as a demo on the page
- Client portals
- Personal portfolios
These are, in general, screens that today live as improvised spreadsheets, Notion pages, or slides, and that wouldn't justify opening a repository, setting up CI/CD, and paying for hosting just to exist.
Who can see and edit
By default, a newly created Site is only visible to its creator and to workspace admins. From there, access control has levels defined in the documentation:
| Access level | Who sees it | Who edits it |
|---|---|---|
| Private | Only the creator and workspace admins | Only the creator |
| Shared | Selected people or the whole workspace | Invitees can edit and publish updates |
| Public | Anyone with the link | Only those invited as editors |
For those building products in a team, this brings ChatGPT Sites closer to a Figma or a Notion: inviting collaborators to edit and publish comes built in, instead of depending on a separate Git repository and deploy pipeline.
Login, storage, and connected tools
The feature lets you add "Sign in with ChatGPT" to the published Site, so visitors can pick up where they left off. This uses storage built into the product itself, to save information and uploaded files, including documents and images, without the Site's creator needing to set up a database.
In eligible workspaces, whoever accesses the Site can approve read-only access to their own connected tools in their ChatGPT account. OpenAI is explicit about the scope: the Site only sees information the visitor already had permission to access. In practice, it's a way to give an app created through conversation access to real user data without the Site's developer writing a custom OAuth integration for each service.
Custom domain, someone else's hosting
Every Site ships with an address hosted on an OpenAI domain, which can be renamed. You can also connect your own domain. If the hosted address is renamed later, the old one automatically redirects to the new one, according to the feature's page.
There is no mention, in the published documentation, of exporting the generated code, running the Site outside OpenAI's infrastructure, or details such as the framework used under the hood, request limits, and a pricing policy specific to the feature.
Where this fits in the AI builder race
ChatGPT Sites enters a field that already has v0 (Vercel), bolt.new, Lovable, and Replit's Agent competing for the same audience: people who want to go from a descriptive sentence to a working site or app without writing a single line of code by hand. OpenAI's positioning difference is packaging this inside the product most people already use day to day, ChatGPT itself, instead of selling a separate tool.
For those who develop professionally, the feature tends to carry more weight for internal prototypes, team dashboards, and client portals than for products that need fine-grained infrastructure control. The proposal runs precisely into what it doesn't say: there is no confirmation, in the source, that a Site's code can be extracted, versioned in a repository of your own, or audited outside OpenAI's environment.
What's still open
In short: OpenAI showed what ChatGPT Sites does, but didn't detail how it does it. Left unanswered, at least in this documentation: what technical stack runs behind each Site, whether any kind of source code export exists, what the usage limits are, and whether the feature will be available on all ChatGPT plans or only on workspace accounts (Team, Business, Enterprise).
Before treating ChatGPT Sites as a substitute for a real deploy pipeline, it's worth waiting for OpenAI to publish these details, or testing the feature directly to see where it breaks in practice.
Translated from the Brazilian Portuguese original · Read the original
FLUX 3 Image brings bounding-box control and native 4K rendering
Black Forest Labs details FLUX 3 Image, an image generation model that accepts coordinate-based layout, renders natively in 4K, and allows editing element by element without regenerating the entire scene.
