Dora NodaGoogle announced on March 19, 2026 that it's shutting down Firebase Studio — its AI-native, browser-based app builder — for good. New workspaces stopped on June 22, 2026. Every remaining workspace goes dark on March 22, 2027, less than two years after the product launched. If you built something there, the clock is already running.
That's a bad week for anyone who prompted an app into existence on Firebase Studio. But it's also, unexpectedly, the cleanest natural experiment the vibe-coding market has run so far: when a platform that generates and hosts your app disappears, what does "your app" actually turn out to be? Source code you can carry anywhere, or a pile of prompts that only ever ran on someone else's servers? Google's own migration docs answer that question in detail — and the answer generalizes past Firebase Studio to every "describe it, we'll build and host it" tool on the market.
What Firebase Studio's Shutdown Actually Requires You to Do
Google shipped an official migration path, not just a deprecation notice. A "Move now" button in each workspace routes you to one of three destinations, and Google's own docs spell out exactly what comes with you and what doesn't:
| Destination | What it's for | How it works |
|---|---|---|
| Google AI Studio | Web-based, multi-device prototyping | Code opens automatically in the new environment; publish via App Hosting (keeps your existing URL) or Cloud Run (new URL) |
| Google Antigravity | Desktop, code-first, agent-driven development | Download a ZIP, open it in the Antigravity IDE, run an @fbs-to-agy-export agent prompt, preview locally, then publish |
| Manual export | Anywhere else | npx firebase-tools@latest studio:export PATH, or the same ZIP download, deployed via Firebase CLI or any external host |
Google frames the shutdown as a focus decision, not a retreat from AI app-building: it's consolidating around AI Studio for browser-based prototyping and the newer Antigravity for agent-driven desktop development, while folding core Firebase services — Cloud Firestore, Authentication, App Hosting — directly into those tools instead of routing them through a standalone Studio product. Those underlying services aren't going anywhere; it's specifically the all-in-one workspace that generated your app and hosted it in one place that's being retired. That distinction matters for the deadline math: miss March 22, 2027 and the workspace itself — plus anything you never exported out of it — is gone permanently, even though the Firebase project it was built on top of keeps running.
What actually travels with you: source code, full git repository history, project structure.
What doesn't, and has to be rebuilt or moved by hand: node_modules (reinstall via npm install), .env files holding API keys (Google explicitly flags these as a security step to handle separately, not something the export script touches), and agent chat history (it's sitting in a local ~/.idx/ai directory you have to go find yourself). If you're on a shared workspace, only the original creator can run the migration tools at all — everyone else has to duplicate the project or export manually.
None of that is trivial, and the deadline is real. But read the table again: what exports is a real git repo with real source code. That's the finding worth sitting with, because it's not true of the rest of the category.
The Same Question, Asked of Every Other Vibe-Coding Platform
Firebase Studio's shutdown forces the export question for its own users right now. But the honest way to read it is as a stress test every vibe-coding platform should be run through before you build on it, not after the sunset notice arrives. Here's where the market's other major players land on the same question — can you leave, with a working app, and how much of it is actually yours:
| Platform | What exports | What stays locked to the vendor |
|---|---|---|
| Bolt.new | Full codebase — compiles client-side via StackBlitz's WebContainers, exports to GitHub, deploys to Vercel, Netlify, or your own infra | Nothing structural; portability is the architecture, not a bolted-on feature |
| Firebase Studio | Source code + git history (see above) | .env secrets, node_modules, agent chat history — all recoverable, none of it blocks the app from running elsewhere |
| Lovable | Frontend code synced to GitHub; as of July 2026, official Export/Pause/Remove buttons let you export project data and disconnect Lovable Cloud in favor of your own Supabase | The backend migration is real work even with the official tooling — community-documented migration guides run 33 steps to move tables, auth users with password hashes, storage files, edge functions, and cron jobs intact |
| Replit Agent | Code, via git | Replit DB — a proprietary key-value store that only exists inside Replit. If your data lives there, migrating means writing an export script, redesigning your data model for a relational database, and rewriting every query that touched it |
| Base44 | React frontend only | Backend logic, database, authentication, and third-party integrations stay server-side behind Base44's SDK — the exported code won't run as a working app on its own without a replacement backend built from scratch |
Lay those five side by side and a spectrum falls out, from "portability is the whole design" to "you get the part that was never the hard part." Bolt.new and Firebase Studio hand you something that runs elsewhere with modest cleanup. Lovable, as of mid-2026, finally ships an official door out, but walking through it is a multi-step data migration, not a button click. Replit's lock-in isn't in the code — it's in a proprietary database with no export format. And Base44's export is close to cosmetic: you get the part of the app a browser renders, not the part that makes it work.
The Checklist This Should Actually Change
Firebase Studio's own numbers make the case for checking this before you build, not after: the product launched in April 2025 and is fully dead by March 2027 — under two years of runway for anything built on it. That's not a uniquely Google problem. It's the base rate for a category still finding its shape, and the broader 2026 vibe-coding field is already watching for the next platform-level failure, not just individual app bugs. Ninety-two percent of US developers now use AI coding tools daily — but trust in the code they produce fell from roughly 40% to 29% over the same stretch, which tells you the market hasn't converged on "safe to build a real business on this" for either the code or the platform underneath it.
Before you pick a vibe-coding tool for anything you intend to keep running, four questions settle where it lands on the spectrum above:
- Does "export" produce a git repo with real source, or a read-only snapshot? Bolt.new and Firebase Studio pass this cleanly. Base44 fails it for anything past the frontend.
- Does the database use a portable connection string, or a proprietary store with no export format? A Postgres connection string in an environment variable survives a platform's death. Replit DB does not.
- Is backend logic exportable code, or calls into a vendor SDK that only resolves against that vendor's servers? This is exactly where Base44's export stops being useful — the SDK calls are the app, and they don't run anywhere else.
- Is there an official migration path, or only a community-reverse-engineered one? Lovable's official Export/Pause/Remove buttons (July 2026) are a materially different guarantee than a third-party script someone wrote by trial and error — official tooling means the vendor is committed to the export format working, not just that it happened to work for one person once.
A platform that fails all four isn't necessarily a bad place to prototype. But it's a bad place to park anything you'd be upset to lose on eighteen months' notice, and Firebase Studio's shutdown is the concrete proof that eighteen months' notice is the realistic floor, not a pessimistic edge case.
Where the Exported Code Actually Lands
Run the checklist and you'll notice it points at a second, quieter question: once the code does export cleanly, where does it go next? Bolt.new's own answer is "deploy to Vercel, Netlify, or your own infrastructure" — the generation tool doesn't try to own that decision, and neither should any platform standing between an agent-written repo and a running production URL.
That's the layer a git-push PaaS occupies, and it's worth being deliberate about which one, for the same reason this whole piece argues for checking a vibe-coding tool's exit before committing to it: a hosting platform that only accepts deploys from its own dashboard is a second lock-in point stacked on top of the first one you just checked for.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, whether that repo came from Bolt.new, a migrated Firebase Studio export, or a human typing in an editor. Star the repo on GitHub or deploy your first app today.
