Google AI Studio Vibe Coding: Build, Limits, Deploy

By Umair Ali Sadaqat · Updated November 6, 2026 · Editorial policy

How to vibe code in Google AI Studio's Build mode: the steps, what it can and can't do, the two-app Starter Tier limit, and how to deploy or export your app.

Short answer: Google AI Studio's vibe coding lives in Build mode at aistudio.google.com/apps. You describe an app, Gemini writes a React frontend plus a Node.js server (or a native Android app), and you iterate against a live preview. When it works, click Publish to put it on Cloud Run, sync it to GitHub, or download a ZIP. Check the limits below before you share a link, because visitors' Gemini calls run on your key. All details were checked against Google's docs in November 2026.

What Build mode actually does

Build mode is a prompt-to-app environment inside AI Studio. Per Google's documentation:

Web apps (default): a React frontend plus a Node.js server-side runtime for secure API calls, database connections, and npm packages. The agent installs packages as needed.
Android apps: Kotlin and Jetpack Compose, previewed in a browser-based emulator, installable on a device via ADB in the browser, and publishable to the Play Store for internal testing.
Built-in integrations: Firebase Firestore and Firebase Authentication (Sign in with Google), Google Workspace connections such as Gmail, Sheets, Docs, Drive and Calendar with automatic OAuth setup, and multiplayer state handled by the server runtime.
Other databases: the docs say you can use any storage you can reach over a network, subject to firewall limits on dynamic IP ranges.

If the term is new, our explainer on what vibe coding is covers the idea. Build mode is one implementation of it.

How to vibe code an app in Google AI Studio

Open Build mode. Go to aistudio.google.com/apps and pick web app (the default) or native Android in the platform picker.
Write the prompt. Describe who uses the app, what the main screen shows, and what data it handles. You can use AI Chips to add features like image generation or Google Maps data, or dictate with speech-to-text. "I'm Feeling Lucky" generates an idea if you are stuck.
Or start from something existing. Open an app in the App Gallery and select Copy App to remix it, or use Import from GitHub from the Add files (+) menu in the prompt box.
Run it and read the preview. Generated files appear with a live preview on the right. The Code tab shows the source.
Iterate in small steps. Ask for one change per message. For visual tweaks, use annotation mode: highlight a part of the UI and describe the change you want.
Handle permissions. If the app needs the camera, microphone or geolocation, the docs say to list those browser APIs in metadata.json under requestFramePermissions.
Check secrets. Open the Secrets panel in Settings to see GEMINI_API_KEY and add any other keys your server code needs.
Save or ship. Sync to GitHub, download a ZIP, or click Publish.

Prompt wording that gives better first drafts

Build a booking page for a small pottery studio. Visitors pick a class from a list, choose a date, and enter name and email. Store requests in Firestore. Show a confirmation screen. Keep the layout single-column and mobile-first. Do not add login.

Naming the data store, the screens, and what to leave out cuts down on the agent inventing features you then have to delete.

Deploy and export options compared

Option	What you get	Requirements (per Google's docs)	Watch out for
Publish, Starter Tier	Cloud Run URL, optional your-app-name.ai.studio subdomain	Eligible account; no Google Cloud project or billing account needed	Up to two services in total, single Cloud Run region; Workspace accounts and accounts with paid Cloud billing are not eligible
Publish, standard deployment	Scalable Cloud Run service	Google Cloud project linked to AI Studio with billing enabled	Cloud Run pricing may apply based on usage
GitHub sync	Two-way repo sync with commit messages	GitHub tab in Settings	Resolve merge conflicts in the dialog
ZIP download	Code you can run anywhere	Set GEMINI_API_KEY yourself	You handle hosting, secrets and updates

On the Starter Tier, publishing is three clicks: Publish, Get Started, Publish App. If you already have two apps live, you must unpublish one or upgrade. Custom subdomains are first-come, first-served and released when you unpublish or delete.

Limits and risks to know before sharing

Your key pays for every visitor. The deployed app uses your Gemini API key for all users' calls, and the docs warn that costs may apply if you use paid models. AI Studio warns you before sharing.
Apps start private. People you share with can see and fork the code, and edit-permission sharers can change it.
No live co-editing. Google's docs say live collaborative editing is not supported; teams collaborate through a shared GitHub repo.
Shared links can fail. The docs mention 403 Access Restricted errors triggered by privacy extensions such as Privacy Badger, or by build errors.
You own the behavior. Apps can be removed for policy violations, and the developer is responsible for the app's behavior and data.
Older apps. Apps built before May 14, 2026 are upgraded to the server-side key approach the next time their Gemini features are modified.

Google's pages we read do not state free-tier request or token numbers, so we are not quoting any. Look at the Billing page inside AI Studio for what applies to your account.

Before you publish: a copyable checklist

[ ] The app works in the preview with a fresh browser profile, not just your signed-in one
[ ] Secrets are in the Secrets panel, not pasted into code or prompts
[ ] You know whose Gemini key the deployed app will use and what a busy day would cost
[ ] Firestore or database rules are reviewed so visitors can't read or write what they shouldn't
[ ] You tested on a phone-width screen
[ ] The repo is synced to GitHub so you have a copy outside AI Studio
[ ] You checked Starter Tier eligibility, or have a billing-enabled Cloud project ready

AI-generated apps can look finished while hiding gaps in validation and access control, which we cover in the vibe coding trust gap for websites.

When Build mode is the wrong tool

Build mode suits prototypes, internal tools, and small interactive apps that call Gemini. It is a heavier route for a plain business site that needs a domain, contact form, and search visibility. A Cloud Run service behind an ai.studio subdomain also isn't the same as a site you manage for search. If you are weighing simpler hosts, see our list of Google Sites alternatives, and once a site is live, set up Google Search Console so you can see how it is indexed.

For a regular site rather than a custom app, We.Inc's free plan lets you publish up to three sites from ready-made templates; the AI builder that drafts a site from a prompt starts on the Starter plan at $20 a month. You can sign up here.

Frequently asked questions

What is vibe coding in Google AI Studio?

It is Build mode, found at aistudio.google.com/apps. You describe an app in plain language, Gemini generates the code, and a live preview appears beside it. Per Google's docs, web apps default to a React frontend with a Node.js server-side runtime, and you can also choose a native Android app (Kotlin and Jetpack Compose). You refine the result by chatting or by using annotation mode on the preview. Docs checked November 2026.

Is Google AI Studio vibe coding free?

Google's Build and deploy docs do not describe a free quota. They say API calls from your app count toward your usage limits and that costs may apply if you use paid models. The Starter Tier lets eligible accounts publish up to two full-stack apps without a Google Cloud project or billing account, but Cloud Run pricing may apply based on usage. Check the Billing page in AI Studio for the current rates on your account.

How do I deploy an app built in Google AI Studio?

Click Publish in the top right. On the Starter Tier you then click Get Started and Publish App, and you get a Cloud Run URL. Standard deployment needs a Google Cloud project linked to your AI Studio account with billing enabled. You can also sync the project to GitHub or download a ZIP and host it yourself.

Can I export my Google AI Studio app to GitHub or my own host?

Yes. AI Studio supports two-way GitHub sync, managed from the GitHub tab in Settings, with AI-generated commit messages and a Resolve conflicts dialog. You can also download a ZIP. If you run the code elsewhere, you must set the GEMINI_API_KEY environment variable yourself.

Will my Gemini API key be exposed in the published app?

For new web apps, Google's docs say AI Studio stores GEMINI_API_KEY as a server-side secret, visible in the Secrets panel, and that it is never included in client-side code. The deployed app uses that key for every visitor's Gemini calls, so their usage counts against your limits and costs.

Who can use the Starter Tier for publishing?

Google's deploy page says accounts signed in with a Google Workspace account are not eligible, nor are accounts with an active or prior paid Google Cloud billing account or without enough activity evidence. It also says support cannot override eligibility. If you are not eligible, use standard deployment with a billing-enabled Google Cloud project.

Start building

How we research, test and update this page: our editorial policy. We.Inc is our own product.