By Google I/O 2026 opened today with a Firebase story that deserves more attention than it will get. Firebase AI Logic is now generally available — meaning you can call Gemini models directly from iOS, Android, web, and Flutter apps in production, without running a server, without embedding API keys in client code, and without building your own security layer. For developers who’ve been waiting to ship AI features without a backend project attached, that wait is over.
The Security Architecture Nobody Talks About
Most “AI in your app” tutorials skip past the part where your API key lives in client code that anyone can decompile. Firebase AI Logic doesn’t.
The naive approach: embed your Gemini API key in the app bundle, make direct API calls, hope nobody notices. They will notice. Your key gets extracted, your quota gets burned, your bill explodes. Firebase AI Logic works differently. Every SDK request goes to Firebase’s proxy first. The proxy runs Firebase App Check — validating that the request came from your legitimate app, not a bot or a scraper. The proxy then forwards the request to Gemini with your server-side credentials. Your API key never touches the client.
As of this month, Google added replay attack protection to App Check: tokens are now strictly single-use. Even if an attacker intercepts a valid token in transit, they cannot replay it — a meaningful security upgrade that slipped under most I/O coverage.
What the SDK Actually Looks Like
Two lines to get a Gemini model instance in an Android app:
val model = Firebase.ai(backend = GenerativeBackend.googleAI())
.generativeModel("gemini-2.5-flash")That’s it. The SDK handles the proxy routing, App Check, credential management, and model instantiation. You write prompts, not infrastructure. Full SDK coverage at GA:
- Swift (iOS and macOS)
- Kotlin and Java (Android)
- JavaScript (web)
- Dart (Flutter)
- Unity
Google is also shipping hybrid inference for Android and iOS — the SDK uses an on-device model when available and falls back to the cloud-hosted version when it isn’t. That matters for latency-sensitive features and for keeping costs down when local inference is sufficient.
Coming From Vertex AI in Firebase?
Firebase AI Logic is the renamed, expanded version of “Vertex AI in Firebase,” rebranded in May 2025. If you were building on the old SDK, you need to migrate. The official migration guide covers all platforms. The core change: the new SDK lets you choose between the Gemini Developer API (recommended for most apps) and the Vertex AI Gemini API (for data residency requirements or existing Google Cloud environments).
One practical difference: the Gemini Developer API does not support Cloud Storage URLs for file attachments. If your app uses Cloud Storage for user-uploaded files that you pass to Gemini, use the Vertex AI backend instead.
Firebase Studio Is Going Away. Here’s What Isn’t.
Google also confirmed today that Firebase Studio — the cloud IDE that absorbed Project IDX — is shutting down in 2027. New workspace creation closes June 22, 2026. Existing workspaces are accessible until March 22, 2027. The migration path is Google Antigravity, the VS Code-based agentic IDE now integrated directly into AI Studio.
Worth stating clearly: core Firebase services are not affected. Firestore, Authentication, App Hosting, Firebase AI Logic, and Firebase SQL Connect are all continuing. Google is consolidating its IDE story, not its platform story.
What Else Dropped for Firebase at I/O 2026
Firebase AI Logic was not the only developer-focused announcement today:
- Firestore native full-text search (preview): Keyword and phrase search built into Firestore, without syncing to Algolia or Elasticsearch. Backed by Google’s search technology.
- Firebase SQL Connect: Fully-managed PostgreSQL for Firebase apps, with real-time syncing, offline cache, and native full-text search.
- Security Rules parity: The same authorization model you use for Firestore now applies to Gemini API access — consistent security across your entire Firebase project.
Taken together, today’s Firebase announcements close three gaps that have pushed developers toward third-party services: AI integration security (Firebase AI Logic), relational data (SQL Connect), and search (Firestore full-text). It’s a more complete platform than it was last week.
Where to Start
The Firebase AI Logic documentation has platform-specific getting started guides for all five SDKs. Fastest path: Firebase console → AI Services → AI Logic → Get started. A guided workflow sets up the required APIs and App Check configuration for your project.
If you’re migrating from Vertex AI in Firebase, start with the migration guide. The full Firebase AI Logic announcement — including auto function calling and JSON schema mapping details — is on the Firebase blog.
