Short answer: yes. OpenAI Codex can write a full app's code across many files, run its tests, and hand you a commit or pull request. What it gives you is code in a repository: getting that code running for other people (a server, a database, an address) is a separate step, covered by OpenAI's Sites beta, an app platform, or hosting you set up yourself.
We build VibeKit, one of the options below. Everything said about Codex here was checked against OpenAI's own documentation in November 2026.
What is OpenAI Codex today?
Codex is OpenAI's coding agent. You describe a task in plain language, and it reads your code, edits files and runs commands. It comes in several forms:
- Codex cloud. You start it from ChatGPT on the web, on your phone or in the desktop app. You pick GitHub repositories, Codex installs the project's dependencies and tests the setup, and each task runs in its own workspace that keeps going while your computer is asleep. You review the changed files and test results, then commit or open a pull request.
- Codex CLI. A terminal program that works on the repository on your own machine, using the tools you already have installed. The CLI is open source under the Apache 2.0 license.
- IDE extension. Codex inside your code editor. VS Code, Cursor and Windsurf use OpenAI's extension; Xcode and JetBrains IDEs have their own integrations.
Every one of these assumes you have a code project. That is the key to the question "can Codex build an app": it is a developer's tool, and it does a developer's work well.
Which ChatGPT plans include Codex?
As of November 2026, OpenAI's Codex pricing page lists Codex in every ChatGPT plan, in very different amounts:
- Free ($0) and Go ($8/month): Codex in the ChatGPT desktop app, subject to rollout, for quick or lightweight coding tasks.
- Plus ($20/month): Codex on the web, in the CLI, in the IDE extension and on iOS, plus cloud integrations such as automatic code review. OpenAI describes it as enough to "power a few focused coding sessions each week."
- Pro (from $100/month, with tiers at $100, $200 and $500): everything in Plus. OpenAI says Pro plans currently have no five-hour usage limit.
- Business ($20 per user per month billed annually for two or more users, or $25 billed monthly), Enterprise and Edu: shared workspaces with admin controls. Enterprise and Edu pricing is through OpenAI sales.
Plus and Pro users who hit their limit can buy extra credits instead of upgrading.
Can I use Codex with an API key instead?
Yes. The CLI, the IDE extension and the ChatGPT desktop app accept an OpenAI API key, and OpenAI bills that usage through your OpenAI Platform account at standard API rates. The trade-off is that API-key use gets no cloud features: Codex cloud requires signing in with ChatGPT, and the pricing page lists no GitHub code review or Slack integration for API keys.
What can Codex build?
Codex works on whatever is in your repository, so the limit is less the kind of app than how well each task is described. OpenAI's own framing for cloud tasks runs "from investigating bugs to building features." It also works in Xcode, through Apple's integration, so it can write iPhone app code too (what shipping a native app involves is its own topic).
Where it is strong:
- Real repositories. It works on your actual project and reads the AGENTS.md instruction files you leave in it, so it can follow your project's conventions.
- Multi-file changes. A request like "add user accounts" touches routes, database code, pages and tests at once. In OpenAI's words, the CLI lets Codex "explore unfamiliar code, plan a change, edit files, and run your local development tools" in one session.
- Tests and review. In the cloud it runs your setup and tests and shows you the results before you commit. Locally,
/reviewchecks uncommitted changes, a commit or a branch before you merge.
What it expects from you: a repository, a way to run the project, and the judgment to read a diff and decide whether to merge it. For a developer that is a normal day. For someone who has never opened a terminal, the code Codex writes is real, but it is not yet an app anyone can visit.
Does Codex host the app it builds?
Codex cloud runs the agent in the cloud. It does not keep your finished app online for visitors. A working app still needs somewhere to run, a database if it saves anything, and an address people can open.
OpenAI now covers that inside ChatGPT with Sites, in public beta. You describe a website in ChatGPT or Codex, and Sites builds, hosts and shares it. A site can have a relational database for saved records, file storage, Sign in with ChatGPT for visitors, and a custom domain you already own (Sites does not register domains). Hosting is free through December 31, 2026; from January 1, 2027, you pay only for usage above your plan's included hosting allowance.
The limits OpenAI lists: some frameworks, databases, background services and hosting patterns are not supported, and you create and manage sites in ChatGPT on the web or desktop app, not from the CLI or IDE extension. OpenAI's plan table lists Sites for Plus, Pro, Business and Enterprise, not for API-key use, and availability also depends on your region and workspace settings.
If you already pay for ChatGPT Plus and want a team dashboard, tracker or request form, Sites is probably your shortest path. Try it first.
The other two routes:
- Your own hosting. Developers often push Codex's code to GitHub and deploy it wherever they like. We compared the options in where to host AI apps.
- An app platform that runs the agent and the hosting together, covered next.
How do app platforms run Codex-class models for you?
Platforms like VibeKit, Lovable, Bolt and Replit put a coding agent behind a chat box: you describe the app, the agent writes the code, and the platform hosts the result at a real URL. Which model does the work, and who pays for it, differs between them. Our BYOK comparison shows which builders accept your own key.
In VibeKit:
- Each app gets its own coding agent that remembers the project, and the app is hosted on AWS at
yourname.vibekit.page. - A database is included on Starter ($9.99/month on the web) and is a $3/month add-on on Free. Paid plans include custom domains.
- Free apps sleep after 60 idle minutes and wake on the next visit. Starter includes one always-on app.
For OpenAI models there are two ways to pay, available on every plan including Free, both with zero VibeKit markup:
- Sign in with ChatGPT. The agent runs on Codex models, and its usage counts against your ChatGPT plan's Codex allowance instead of VibeKit credits.
- Your own OpenAI API key. Calls go straight to OpenAI and bill to your OpenAI account at OpenAI's rates.
Setup, token security and what each route covers are on our Codex BYOK page. With neither connected, the agent uses platform models paid from your VibeKit credits (OpenAI, Claude, Gemini, DeepSeek, Grok and GLM), or an always-free model tier. Plans are on the pricing page.
What this is not: VibeKit's agent is not OpenAI's Codex app, so you do not get Codex cloud's GitHub code review or Slack integration here. What you get is the part Codex leaves open: a running app with saved data and a URL, which you keep improving by chat on the web or in the iPhone app.
Which route should you pick?
| If you... | Start with |
|---|---|
| Are a developer with a repo and a host already | Codex CLI, IDE extension or cloud |
| Pay for ChatGPT Plus and want a simple team site | OpenAI Sites |
| Want the agent and the hosting together, with your own OpenAI access or none | An app platform such as VibeKit |
Whichever route you pick, the hard parts tend to be the same: vague requests, and services only you can sign up for, like Stripe for payments. We wrote about both in can AI build an app for me, and the Claude and ChatGPT versions of this question are answered in can Claude build an app and can ChatGPT build an app.
Try it: build a small app with Codex models
Start in VibeKit, or open the iOS app. To use your ChatGPT plan or an OpenAI key, connect OpenAI first: Settings > AI providers on the web, or the Profile tab on iPhone. Then choose a new web app and describe a first version you can test. For example:
Build a reading log where I can add a book with its title, author and a rating from 1 to 5, see my books sorted by rating, and mark one as currently reading. Keep my books saved between visits and give me the live URL.
Open the URL, add two books, reload the page, and check that both are still there with their ratings. Then ask for one change at a time, such as a search box, and check each one the same way.