Firebase AI Logic in Flutter: Gemini without a custom backend

You do not always need a Node/Go proxy to call an LLM from a Flutter app. Firebase AI Logic (firebase_ai) gives you a typed client for Gemini with Firebase Auth, App Check, and quota controls already in the path.

Diagram: Firebase AI Logic

Why teams pick it

Client-side multimodal calls (photos → structured nutrition logs, for example) without standing up infra.
Server Prompt Templates keep system prompts and tool definitions out of the binary.
Works with the same Firebase project you already use for Crashlytics/Auth.

Minimal client shape

import 'package:firebase_ai/firebase_ai.dart';

final model = FirebaseAI.googleAI().generativeModel(model: 'gemini-2.5-flash');
final response = await model.generateContent([
  Content.text('Describe this UI screenshot for a bug report.'),
  Content.data('image/png', pngBytes),
]);
print(response.text);

(Exact API surface evolves — pin the package version and read the current docs.)

Security checklist

App Check on.
Auth required for expensive models.
Server templates for prompts that encode business rules.
Client never holds long-lived provider keys.

When to use Genkit instead

If flows, tools, and observability live server-side, prefer Genkit Dart. Firebase AI Logic shines when the product interaction is on-device and latency-sensitive.