Migrate to the Firebase AI Logic SDKs from the Preview version of Vertex AI in Firebase SDKs

Stay organized with collections Save and categorize content based on your preferences.

Swift

1. In Xcode, with your app project open, update your Firebase package to v11.13.0 or later using one of the following options:

   • Option 1: Update all packages: Navigate to File > Packages > Update to Latest Package Versions.

   • Option 2: Update Firebase individually: Navigate to the Firebase package in the section called Package Dependencies. Right-click on the Firebase package, and then select Update Package.

2. Make sure that the Firebase package now shows v11.13.0 or later. If it doesn't, verify that your specified Package Requirements allow updating to v11.13.0 or later.

3. Select your app's target in the Project Editor, and then navigate to the Frameworks, Libraries, and Embedded Content section.

4. Add the new library: Select the + button, and then add FirebaseAI from the Firebase package.

5. After you've finished migrating your app (see the remaining sections in this guide), make sure to remove the old library:
   Select FirebaseVertexAI-Preview, and then press the — button.

Kotlin

1. In your module (app-level) Gradle file (usually <project>/<app-module>/build.gradle.kts or <project>/<app-module>/build.gradle), replace old dependencies (as applicable) with the following.

   Note that it might be easier to migrate your app's codebase (see the remaining sections in this guide) before deleting the old dependency.

   // BEFORE
   dependencies {
     implementation("com.google.firebase:firebase-vertexai:16.0.0-betaXX")
   }
   
   
   // AFTER
   dependencies {
     // Import the BoM for the Firebase platform
     implementation(platform("com.google.firebase:firebase-bom:33.13.0"))
   
     // Add the dependency for the Firebase AI Logic library
     // When using the BoM, you don't specify versions in Firebase library dependencies
     implementation("com.google.firebase:firebase-ai")
   }

2. Sync your Android project with Gradle files.

Note that if you choose to not use the Firebase Android BoM, then just add the dependency for the firebase-ai library and accept the latest version that's suggested by Android Studio.

Java

1. In your module (app-level) Gradle file (usually <project>/<app-module>/build.gradle.kts or <project>/<app-module>/build.gradle), replace old dependencies (as applicable) with the following.

   Note that it might be easier to migrate your app's codebase (see the remaining sections in this guide) before deleting the old dependency.

   // BEFORE
   dependencies {
     implementation("com.google.firebase:firebase-vertexai:16.0.0-betaXX")
   }
   
   
   // AFTER
   dependencies {
     // Import the BoM for the Firebase platform
     implementation(platform("com.google.firebase:firebase-bom:33.13.0"))
   
     // Add the dependency for the Firebase AI Logic library
     // When using the BoM, you don't specify versions in Firebase library dependencies
     implementation("com.google.firebase:firebase-ai")
   }

2. Sync your Android project with Gradle files.

Note that if you choose to not use the Firebase Android BoM, then just add the dependency for the firebase-ai library and accept the latest version that's suggested by Android Studio.

Web

1. Get the latest version of the Firebase JS SDK for Web using npm:

   npm i firebase@latest

   OR

   yarn add firebase@latest

2. Wherever you've imported the library, update your import statements to use firebase/ai instead.

   Note that it might be easier to migrate your app's codebase (see the remaining sections in this guide) before deleting the old imports.

   // BEFORE
   import { initializeApp } from "firebase/app";
   import { getVertexAI, getGenerativeModel } from "firebase/vertexai-preview";
   
   
   // AFTER
   import { initializeApp } from "firebase/app";
   import { getAI, getGenerativeModel } from "firebase/ai";

Dart

1. Update to the use the firebase_ai package in your pubspec.yaml file by running the following command from your Flutter project directory:

   flutter pub add firebase_ai

2. Rebuild your Flutter project:

   flutter run

3. After you've finished migrating your app (see the remaining sections in this guide), make sure to delete the old package:

   flutter pub remove firebase_vertexai

Unity

Support for Unity wasn't available from "Vertex AI in Firebase".

Learn how to get started with the Firebase AI Logic SDK for Unity.

Swift

import FirebaseAI

// Initialize the Gemini Developer API backend service
let ai = FirebaseAI.firebaseAI(backend: .googleAI())

// Create a `GenerativeModel` instance with a model that supports your use case
let model = ai.generativeModel(modelName: "gemini-2.0-flash")

Kotlin

// Initialize the Gemini Developer API backend service
// Create a `GenerativeModel` instance with a model that supports your use case
val model = Firebase.ai(backend = GenerativeBackend.googleAI())
                        .generativeModel("gemini-2.0-flash")

Java

// Initialize the Gemini Developer API backend service
// Create a `GenerativeModel` instance with a model that supports your use case
GenerativeModel ai = FirebaseAI.getInstance(GenerativeBackend.googleAI())
        .generativeModel("gemini-2.0-flash");

// Use the GenerativeModelFutures Java compatibility layer which offers
// support for ListenableFuture and Publisher APIs
GenerativeModelFutures model = GenerativeModelFutures.from(ai);

Web

import { initializeApp } from "firebase/app";
import { getAI, getGenerativeModel, GoogleAIBackend } from "firebase/ai";

// TODO(developer) Replace the following with your app's Firebase configuration
// See: https://firebase.google.com/docs/web/learn-more#config-object
const firebaseConfig = {
  // ...
};

// Initialize FirebaseApp
const firebaseApp = initializeApp(firebaseConfig);

// Initialize the Gemini Developer API backend service
const ai = getAI(firebaseApp, { backend: new GoogleAIBackend() });

// Create a `GenerativeModel` instance with a model that supports your use case
const model = getGenerativeModel(ai, { model: "gemini-2.0-flash" });

Dart

import 'package:firebase_ai/firebase_ai.dart';
import 'package:firebase_core/firebase_core.dart';
import 'firebase_options.dart';

// Initialize FirebaseApp
await Firebase.initializeApp(
  options: DefaultFirebaseOptions.currentPlatform,
);

// Initialize the Gemini Developer API backend service
// Create a `GenerativeModel` instance with a model that supports your use case
final model =
      FirebaseAI.googleAI().generativeModel(model: 'gemini-2.0-flash');

Unity

Support for Unity wasn't available from "Vertex AI in Firebase".

Learn how to get started with the Firebase AI Logic SDK for Unity.

Swift

• Enumerations

  • Replaced most enum types with structs with static variables. This change allows more flexibility for evolving the API in a backward-compatible way. When using switch statements, you must now include a default: case to cover unknown or unhandled values, including new values that are added to the SDK in the future.

  • Renamed the BlockThreshold enumeration to HarmBlockThreshold; this type is now a struct.

  • Removed unknown and unspecified cases from the following enumerations (now structs): HarmCategory, HarmBlockThreshold, HarmProbability, BlockReason, and FinishReason.

  • Replaced the enumeration ModelContent.Part with a protocol named Part to allow new types to be added in a backward-compatible way. This change is described in greater detail in the Content parts section.

• Content parts

  • Removed the ThrowingPartsRepresentable protocol, and simplified the initializers for ModelContent to avoid occasional compiler errors. Images that don't encode properly will still throw errors when being used in generateContent.

  • Replaced the ModelContent.Part cases with the following struct types conforming to the Part protocol:

    • .text to TextPart
    • .data to InlineDataPart
    • .fileData to FileDataPart
    • .functionCall to FunctionCallPart
    • .functionResponse to FunctionResponsePart

• Harm category

  • Changed the HarmCategory to no longer be nested in the SafetySetting type. If you're referring to it as SafetySetting.HarmCategory, that can be replaced with HarmCategory.

• Safety feedback

  • Removed the SafetyFeedback type, since it wasn't used in any of the responses.

• Citation metadata

  • Renamed the citationSources property to citations in CitationMetadata.

• Total billable characters

  • Changed the totalBillableCharacters property in CountTokensResponse to be optional to reflect situations where no characters are sent.

• Candidate response

  • Renamed CandidateResponse to Candidate to match other platforms.

• Generation configuration

  • Changed the public properties of GenerationConfig to internal. They all remain configurable in the initializer.

Kotlin

• Enumerations

  • Replaced enum classes and sealed classes with regular classes. This change allows more flexibility for evolving the API in a backward compatible way.

  • Renamed the BlockThreshold enumeration to HarmBlockThreshold.

  • Removed values from the following enumerations: HarmBlockThreshold, HarmProbability, HarmSeverity, BlockReason, and FinishReason.

• Blob methods

  • Renamed all methods that included Blob as part of their name to use InlineData instead.

• Safety settings

  • Changed the field method to be nullable.

• Duration class

  • Removed all usages of Kotlin's Duration class, and replaced it with long. This change provides better interoperability with Java.

• Citation metadata

  • Wrapped all the fields previously declared in CitationMetadata into a new class called Citation. Citations can be found in the list called citations in CitationMetadata. This change allows better alignment of types across platforms.

• Count tokens

  • Changed the field totalBillableCharacters to be nullable.

• Total billable characters

  • Changed the totalBillableCharacters property in CountTokensResponse to be optional to reflect situations where no characters are sent.

• Instantiating a model

  • Moved the requestOptions parameter to the end of the parameter list to align with other platforms.

• Live API

  • Removed UNSPECIFIED value for enum class ResponseModality. Instead use null.

  • Renamed LiveGenerationConfig.setResponseModalities to LiveGenerationConfig.setResponseModality.

  • Removed the LiveContentResponse.Status class, and instead have nested the status fields as properties of LiveContentResponse.

  • Removed the LiveContentResponse class, and instead have provided subclasses of LiveServerMessage that match the responses from the model.

  • Changed LiveModelFutures.connect to return ListenableFuture<LiveSessionFutures> instead of ListenableFuture<LiveSession>.

Java

• Enumerations

  • Replaced enum classes and sealed classes with regular classes. This change allows more flexibility for evolving the API in a backward compatible way.

  • Renamed the BlockThreshold enumeration to HarmBlockThreshold.

  • Removed values from the following enumerations: HarmBlockThreshold, HarmProbability, HarmSeverity, BlockReason, and FinishReason.

• Blob methods

  • Renamed all methods that included Blob as part of their name to use InlineData instead.

• Safety settings

  • Changed the field method to be nullable.

• Duration class

  • Removed all usages of Kotlin's Duration class, and replaced it with long. This change provides better interoperability with Java.

• Citation metadata

  • Wrapped all the fields previously declared in CitationMetadata into a new class called Citation. Citations can be found in the list called citations in CitationMetadata. This change allows better alignment of types across platforms.

• Count tokens

  • Changed the field totalBillableCharacters to be nullable.

• Total billable characters

  • Changed the totalBillableCharacters property in CountTokensResponse to be optional to reflect situations where no characters are sent.

• Instantiating a model

  • Moved the requestOptions parameter to the end of the parameter list to align with other platforms.

• Live API

  • Removed UNSPECIFIED value for enum class ResponseModality. Instead use null.

  • Renamed LiveGenerationConfig.setResponseModalities to LiveGenerationConfig.setResponseModality.

  • Removed the LiveContentResponse.Status class, and instead have nested the status fields as properties of LiveContentResponse.

  • Removed the LiveContentResponse class, and instead have provided subclasses of LiveServerMessage that match the responses from the model.

  • Changed LiveModelFutures.connect to return ListenableFuture<LiveSessionFutures> instead of ListenableFuture<LiveSession>.

• Changed various Java builder methods to now correctly return the instance of their class, instead of void.

Web

• Enumerations

  • Removed values from the following enumerations: HarmCategory, BlockThreshold, HarmProbability, HarmSeverity, BlockReason, and FinishReason.

• Block reason

  • Changed blockReason in PromptFeedback to be optional.

Changes required only if you're starting to use the Gemini Developer API (instead of the Vertex AI Gemini API):

• Safety settings

  • Removed usages of the unsupported SafetySetting.method.

• Inline data

  • Removed usages of the unsupported InlineDataPart.videoMetadata.

Dart

• Enumerations

  • Removed values from the following enumerations: HarmCategory, HarmProbability, BlockReason, and FinishReason.

• Data part

  • Renamed DataPart to InlineDataPart, and the static data function to inlineData to align with other platforms.

• Request options

  • Removed RequestOptions since timeout wasn't functional. It will be re-added in the near future, but it will be moved to the GenerativeModel type to match other platforms.

• Stop sequences

  • Changed the stopSequences parameter in GenerationConfig to be optional and to default to null instead of an empty array.

• Citations

  • Renamed the citationSources property to citations in CitationMetadata. The CitationSource type was renamed to Citation to match other platforms.

• Unnecessary public types, methods, and properties

  • Removed the following types, methods, and properties which were unintentionally exposed: defaultTimeout, CountTokensResponseFields, parseCountTokensResponse, parseEmbedContentResponse, parseGenerateContentResponse, parseContent, BatchEmbedContentsResponse, ContentEmbedding, EmbedContentRequest, and EmbedContentResponse.

• Count tokens

  • Removed extra fields from the countTokens function that are no longer necessary. Only contents is needed.

• Instantiating a model

  • Moved the systemInstruction parameter to the end of the parameter list to align with other platforms.

• Embedding functionality

  • Removed unsupported embedding functionality (embedContent and batchEmbedContents) from the model.

Unity

Support for Unity wasn't available from "Vertex AI in Firebase".

Learn how to get started with the Firebase AI Logic SDK for Unity.