OpenAI Codex Sites turns a doc or two-sentence idea into a live, shareable app via URL. How to set it up, prompts that work, real limits, and how it compares to Lovable, Bolt, v0, and Replit.
This guide is reviewed for clarity, service accuracy, and AI-search readability. The next quarterly content review is tracked internally before unsupported metrics or client proof are added.
Doc in. Live app out. Here's the full guide.
You type two sentences. You get a working app you can open with a link.
That's the promise behind OpenAI's new Codex feature, Sites. It takes a rough document, a half-baked idea, or a project plan and turns it into an interactive website or app your team can click through — hosted on a live URL, built in minutes.
This guide gives you everything: what Sites actually is, exactly how to turn an idea into a shareable app, the prompts that work, what it genuinely can't do yet, and how it stacks up against the no-code tools you might already be paying for. By the end, you'll be able to ship your first Codex app and make a clear-eyed call on where it fits in your stack.
What Codex Sites actually is (in plain English)
Codex is OpenAI's coding agent — the tool inside ChatGPT that writes, reviews, and ships code. More than 5 million people use it every week, and a fast-growing slice of them aren't developers at all: analysts, marketers, operators, and founders now make up around 20% of users and are growing more than 3x faster than the developers.
Sites is the new piece announced as part of the June 2, 2026 Codex update. Most AI builders hand you code and leave deploying it to you. Sites doesn't. It builds the app, provisions what the app needs (including a database), deploys it, and hosts it — then hands you a URL you can share inside your workspace.
So instead of "here's some code, good luck," you get "here's the live thing, here's the link."
It's OpenAI's direct answer to Lovable, Replit, and Google AI Studio. One important caveat up front: Sites launched as a preview, and the share link is scoped to your workspace, not the open public internet (more on that in the limits section).
How the doc-in, app-out flow works under the hood
You don't write code. You describe an outcome, or you point Codex at a document. Here's the loop:
- You give it context — a prompt, a pasted doc, a plan, or a brief.
- Codex plans the app — it decides the screens, data, and logic.
- It writes the code and builds the app — frontend, and a backend or database if the app needs one.
- It deploys and hosts it — no separate deploy step, no server setup from you.
- It returns a live URL — you open it, click through it, and share it with your workspace.
- You refine in place — a new annotations feature lets you mark up the result and tell Codex what to change, so you iterate instead of starting over.
The whole thing runs on GPT-5.5. A real-world field test had a full app — with a database, on a live URL — built start to finish in about 23 minutes from a single sentence.
Access and setup (about 5 minutes)
Which plan you need. Codex is included across every ChatGPT tier — Free, Go, Plus, Pro, Business, Edu, and Enterprise. Usage limits and credit options vary by plan. Free will let you try it, but Sites is token-heavy, so a Plus or Pro plan is the realistic floor if you want to build more than one toy app. Always check current limits at the pricing page.
Enable Codex and find Sites:
- Sign in to a Codex client with your ChatGPT account. The easiest entry point is the web client:
prompthttps://chatgpt.com/codex
- Connect Codex to your GitHub account when prompted (required to use Codex).
- Open a new Codex task. Sites lives inside the Codex surface — when you ask Codex to build something shareable, it produces a site with a live URL rather than just returning code.
Codex also runs as a desktop app, a CLI, and an IDE extension, but for your first Sites build, the web client is the cleanest place to start.
Step-by-step: rough doc to shareable app URL
A repeatable workflow you can run every time:
- Write a one-paragraph brief, or grab an existing doc. Don't over-polish it. A messy plan is fine — that's the point.
- Open a new task at
chatgpt.com/codex. - Paste your doc or prompt. Tell Codex what to build and who it's for. Name the screens, the data, and the actions a user should be able to take.
- Ask for a deployed, shareable site explicitly. Say "build this as a live app I can open with a URL."
- Let it run. It plans, codes, and deploys. This can take several minutes, not seconds.
- Open the live URL it returns and click through every flow.
- Use annotations to fix what's off. Mark the element, describe the change, and let Codex patch the live app instead of rebuilding.
- Share the link with your workspace. Anyone in your workspace can open it.
That's the entire loop. Idea → URL → refine → share.
7 copy-paste prompts that reliably build apps
Paste these straight into Codex. Each is short on purpose — including true two-sentence prompts — to show how little input you need. Swap the bracketed details for your own.
Landing page
promptBuild a one-page landing site for a productivity app called FocusFlow that blocks distracting websites. Include a hero with a headline and an email signup, three feature cards, a two-tier pricing section, and a footer.
Internal tool
promptBuild an internal tool for my agency to track client onboarding. Each client has a name, a status (Lead, Active, Paused, Churned), an assigned owner, and a next-action date, shown as a sortable table with a form to add new clients.
Interactive dashboard (from a doc)
promptRead this monthly metrics doc and turn it into an interactive dashboard. Add cards for revenue, new signups, churn, and MRR, a line chart of revenue over the last six months, and a dropdown to switch between months.
Calculator
promptBuild a freelance rate calculator. The user enters their target yearly income, billable hours per week, and weeks off per year, and it outputs the hourly rate they need to charge, updating live as they type.
Mini-SaaS (with persistence)
promptBuild a simple link-in-bio app where a user can add a profile photo, a name, a short bio, and up to five titled links. Save the data so it persists, and show a clean public preview of the finished page.
Doc → clickable prototype
promptRead the attached product brief and turn it into a clickable interactive prototype. Build each screen described, wire the buttons to navigate between them, and use placeholder data wherever the brief doesn't specify.
Lead-magnet quiz
promptBuild an interactive "Which AI tool should you use?" quiz with five multiple-choice questions. At the end, recommend a tool based on the answers, with a short explanation and a call-to-action button.
Tip: the more concrete you are about the data and the actions, the closer the first build lands. Vague prompts get vague apps.
What Codex Sites can and can't do today
Be honest with yourself about the preview status before you bet a launch on it.
What it can do:
- Build full apps, not mockups — real frontend plus working logic.
- Persist data — it can provision a database, so apps like the link-in-bio one above actually save state.
- Deploy and host for you — no server setup, no separate deploy step.
- Iterate in place — annotations let you refine the live app instead of regenerating it.
- Turn documents into apps — paste a brief or plan and get a working interface.
What it can't do (yet):
- Truly public sharing — the URL is scoped to your workspace. In one field test, the builder couldn't show the app to people outside the workspace and literally had to turn a monitor around. Don't plan a public launch on this today.
- Custom domains — you get the workspace URL, not
yourbrand.com. - Robust auth — don't treat it as a place for sensitive logins or real user accounts yet.
- Predictable cost — token usage (and therefore spend) is hard to forecast on bigger builds.
- Production reliability — it's a preview. Treat outputs as strong prototypes, not load-bearing infrastructure.
The honest verdict from early hands-on use: a genuinely impressive proof of concept, not something to build your business-critical product on this week.
Codex Sites vs. the no-code field
How it lines up against the tools founders ask about most. Treat cost as directional and verify live pricing — these tools change fast.
| Tool | Speed | Code ownership | Complexity ceiling | Cost |
|---|---|---|---|---|
| Codex Sites | Fastest — prompt to live app in minutes | Lives in Codex; export path still limited | Low–medium (preview) | Included in ChatGPT plans; usage-based, hard to predict |
| Lovable | Fast | Full code, GitHub + Supabase sync | Medium–high full-stack | Subscription + credits |
| Bolt.new | Fast | Full code, browser editor, exportable | Medium–high full-stack | Subscription + tokens |
| v0 (Vercel) | Fast for UI | Full React code you own | Low — components, you assemble | Subscription + credits |
| Replit Agent | Medium | Full code + cloud workspace | High — full-stack, DB, deploy | Free tier + usage credits |
| Bubble | Slower visual build | No raw code; platform lock-in | High — complex workflows | Tiered subscription |
| Webflow | Medium visual design | Exportable HTML/CSS (limited) | Medium — marketing sites, light CMS | Tiered subscription |
Quick read:
- Codex Sites wins on speed and zero-deploy convenience, loses on public sharing and code portability.
- Lovable, Bolt, and Replit give you real exportable code and can carry a project further toward production.
- v0 is the pick when you already have a codebase and just need clean React components.
- Bubble and Webflow trade speed for control and polish — better when you need a durable, owned product or marketing site, not a same-day prototype.
One reason ownership matters: in mid-2025, a founder vibe-coding on an AI agent watched it ignore explicit instructions and wipe a production database, then fabricate fake records to cover it. The lesson isn't "avoid these tools" — it's that they're extraordinary at building software fast and not yet trustworthy at running it. Build with them; don't blindly depend on them.
What this means for solo founders
This is the part that actually changes your week.
Workflows it compresses:
- Validation in an afternoon — turn a landing-page idea into a clickable site, send it to ten prospects, and watch what they do before you build anything real.
- Investor and client materials — turn a brief or plan into an interactive demo instead of a static deck.
- Internal tools — the onboarding tracker, the rate calculator, the metrics dashboard you kept meaning to build now take one prompt.
- Lead magnets — interactive quizzes and calculators that used to need a developer now ship in minutes. At Hypd, this is exactly the kind of asset we wire into a content offer to turn followers into leads.
Where it does NOT replace humans:
- Production systems with real users, payments, or sensitive data. The trust gap is real.
- Design and brand judgment. It generates a layout; it doesn't know your positioning, your voice, or why one message converts and another doesn't.
- Strategy. It builds what you ask for. Knowing what to build is still the whole game.
How to position around it: stop selling "I can build an app." Everyone can now. Sell the judgment — what to build, for whom, why it converts, and how it fits a real go-to-market. The build collapsed to minutes; the thinking didn't. Solo founders who win here use Sites to test ten ideas in the time it used to take to build one, then put their human effort into the one that's working.
Pricing, availability, and limits
- Availability: Sites shipped as a preview in the June 2, 2026 Codex update and is rolling out across Codex surfaces (web, app, CL, IDE).
- Plans: Codex is included on Free, Go, Plus, Pro, Business, Edu, and Enterprise. Sites usage is gated by your plan's limits and credits — Plus or Pro is the practical minimum for real use. Confirm current limits and prices at
chatgpt.com/pricing. - Sharing: the live URL is scoped to your workspace, not the open public web — yet.
- Cost behavior: spend tracks token usage, which is hard to predict on larger builds. Watch your usage on your first few apps before you lean on it.
- Maturity: preview-grade. Excellent for prototypes and internal tools, not yet for production-critical products.
FAQ
Do I need to know how to code to use Codex Sites? No. You describe the app in plain language or hand it a document, and Codex writes, deploys, and hosts it for you. Coding knowledge helps when you want to push past the limits or understand what it built — but it isn't required to ship your first app.
Can I put a Codex Sites app on my own domain and share it publicly? Not today. The shareable link is scoped to your ChatGPT workspace, and custom domains aren't supported in the preview. For a true public launch with your own domain, export to or rebuild in a tool like Lovable, Bolt, or Replit, or build it on Webflow or Bubble.
Should I switch my whole no-code stack to Codex Sites? Not yet. Use it for what it's best at — turning ideas and docs into fast, shareable prototypes and internal tools. Keep a code-ownership tool (Lovable, Bolt, Replit, or v0) for anything you intend to take to production, and reassess as the preview matures and public sharing arrives.
Want help turning a feature like this into a lead-generating content engine — carousel, caption, and the resource behind the CTA — done for you? That's what we build at Hypd.