firebase-ai

This module provides the Firebase AI Logic API, com.google.firebase.ai, on Android, iOS, macOS, tvOS, JS and Wasm: the Android SDK on Android, the Firebase JS SDK's firebase/ai on JS and Wasm (wasmJs) and the Firebase AI Logic iOS SDK (FirebaseAILogic, through an Objective-C wrapper) on Apple. The Gemini models (generateContent, streaming, chats, structured output with generateObject, automatic function calling), server prompt templates and the Live API are available from common code with the Android SDK's classes and signatures.

The Android SDK's API is already built on coroutines, Flow and kotlinx.serialization, so it is used as is from common code. On the platforms other than Android requests and responses cross to the platform SDK as the JSON of the AI Logic REST API, and they are checked, and their errors mapped to FirebaseAIExceptions, the way the Android SDK does.

Not available in common code, as their signatures need Android/JVM-only types: images as android.graphics.Bitmap (ImagePart's constructors, Content.Builder.image, the Bitmap overloads of generateContent, countTokens and sendMessage; send images as InlineDataParts), Citation.publicationDate (java.util.Calendar), the AudioRecord/AudioTrack initialization handler of LiveAudioConversationConfig, and the Java futures of com.google.firebase.ai.java. The platforms other than Android have no on-device model, as Android without the firebase-ai-ondevice library: GenerativeModel.onDeviceExtension is null there, the hybrid inference modes use the cloud model and ONLY_ON_DEVICE fails with FirebaseAIOnDeviceNotAvailableException. LiveSession.startAudioConversation records and plays audio on Android only; elsewhere it does nothing (stream audio with sendAudioRealtime instead). The JS SDK has no realtime input configuration for the Live API, so LiveGenerationConfig.realtimeInputConfig is not applied on JS and Wasm.

On Apple platforms the iOS SDK has no JSON Schema in its configuration, so the JSON schemas of generateObject and of automatic functions are sent as the equivalent OpenAPI schemas; it also has no displayName for inline data, no parts in function responses and no total request timeout (RequestOptions.timeout limits the time without data), and its Live API sends realtime audio as audio/pcm.

This module provides the Firebase AI Logic API, com.google.firebase.ai, on Android, iOS, macOS, tvOS, JS and Wasm: the Android SDK on Android, the Firebase JS SDK's firebase/ai on JS and Wasm (wasmJs) and the Firebase AI Logic iOS SDK (FirebaseAILogic, through an Objective-C wrapper) on Apple. The Gemini models (generateContent, streaming, chats, structured output with generateObject, automatic function calling), server prompt templates and the Live API are available from common code with the Android SDK's classes and signatures.

The Android SDK's API is already built on coroutines, Flow and kotlinx.serialization, so it is used as is from common code. On the platforms other than Android requests and responses cross to the platform SDK as the JSON of the AI Logic REST API, and they are checked, and their errors mapped to FirebaseAIExceptions, the way the Android SDK does.

Not available in common code, as their signatures need Android/JVM-only types: images as android.graphics.Bitmap (ImagePart's constructors, Content.Builder.image, the Bitmap overloads of generateContent, countTokens and sendMessage; send images as InlineDataParts), Citation.publicationDate (java.util.Calendar), the AudioRecord/AudioTrack initialization handler of LiveAudioConversationConfig, and the Java futures of com.google.firebase.ai.java. The platforms other than Android have no on-device model, as Android without the firebase-ai-ondevice library: GenerativeModel.onDeviceExtension is null there, the hybrid inference modes use the cloud model and ONLY_ON_DEVICE fails with FirebaseAIOnDeviceNotAvailableException. LiveSession.startAudioConversation records and plays audio on Android only; elsewhere it does nothing (stream audio with sendAudioRealtime instead). The JS SDK has no realtime input configuration for the Live API, so LiveGenerationConfig.realtimeInputConfig is not applied on JS and Wasm.

On Apple platforms the iOS SDK has no JSON Schema in its configuration, so the JSON schemas of generateObject and of automatic functions are sent as the equivalent OpenAPI schemas; it also has no displayName for inline data, no parts in function responses and no total request timeout (RequestOptions.timeout limits the time without data), and its Live API sends realtime audio as audio/pcm.

This module provides the Firebase AI Logic API, com.google.firebase.ai, on Android, iOS, macOS, tvOS, JS and Wasm: the Android SDK on Android, the Firebase JS SDK's firebase/ai on JS and Wasm (wasmJs) and the Firebase AI Logic iOS SDK (FirebaseAILogic, through an Objective-C wrapper) on Apple. The Gemini models (generateContent, streaming, chats, structured output with generateObject, automatic function calling), server prompt templates and the Live API are available from common code with the Android SDK's classes and signatures.

The Android SDK's API is already built on coroutines, Flow and kotlinx.serialization, so it is used as is from common code. On the platforms other than Android requests and responses cross to the platform SDK as the JSON of the AI Logic REST API, and they are checked, and their errors mapped to FirebaseAIExceptions, the way the Android SDK does.

Not available in common code, as their signatures need Android/JVM-only types: images as android.graphics.Bitmap (ImagePart's constructors, Content.Builder.image, the Bitmap overloads of generateContent, countTokens and sendMessage; send images as InlineDataParts), Citation.publicationDate (java.util.Calendar), the AudioRecord/AudioTrack initialization handler of LiveAudioConversationConfig, and the Java futures of com.google.firebase.ai.java. The platforms other than Android have no on-device model, as Android without the firebase-ai-ondevice library: GenerativeModel.onDeviceExtension is null there, the hybrid inference modes use the cloud model and ONLY_ON_DEVICE fails with FirebaseAIOnDeviceNotAvailableException. LiveSession.startAudioConversation records and plays audio on Android only; elsewhere it does nothing (stream audio with sendAudioRealtime instead). The JS SDK has no realtime input configuration for the Live API, so LiveGenerationConfig.realtimeInputConfig is not applied on JS and Wasm.

On Apple platforms the iOS SDK has no JSON Schema in its configuration, so the JSON schemas of generateObject and of automatic functions are sent as the equivalent OpenAPI schemas; it also has no displayName for inline data, no parts in function responses and no total request timeout (RequestOptions.timeout limits the time without data), and its Live API sends realtime audio as audio/pcm.

This module provides the Firebase AI Logic API, com.google.firebase.ai, on Android, iOS, macOS, tvOS, JS and Wasm: the Android SDK on Android, the Firebase JS SDK's firebase/ai on JS and Wasm (wasmJs) and the Firebase AI Logic iOS SDK (FirebaseAILogic, through an Objective-C wrapper) on Apple. The Gemini models (generateContent, streaming, chats, structured output with generateObject, automatic function calling), server prompt templates and the Live API are available from common code with the Android SDK's classes and signatures.

The Android SDK's API is already built on coroutines, Flow and kotlinx.serialization, so it is used as is from common code. On the platforms other than Android requests and responses cross to the platform SDK as the JSON of the AI Logic REST API, and they are checked, and their errors mapped to FirebaseAIExceptions, the way the Android SDK does.

Not available in common code, as their signatures need Android/JVM-only types: images as android.graphics.Bitmap (ImagePart's constructors, Content.Builder.image, the Bitmap overloads of generateContent, countTokens and sendMessage; send images as InlineDataParts), Citation.publicationDate (java.util.Calendar), the AudioRecord/AudioTrack initialization handler of LiveAudioConversationConfig, and the Java futures of com.google.firebase.ai.java. The platforms other than Android have no on-device model, as Android without the firebase-ai-ondevice library: GenerativeModel.onDeviceExtension is null there, the hybrid inference modes use the cloud model and ONLY_ON_DEVICE fails with FirebaseAIOnDeviceNotAvailableException. LiveSession.startAudioConversation records and plays audio on Android only; elsewhere it does nothing (stream audio with sendAudioRealtime instead). The JS SDK has no realtime input configuration for the Live API, so LiveGenerationConfig.realtimeInputConfig is not applied on JS and Wasm.

On Apple platforms the iOS SDK has no JSON Schema in its configuration, so the JSON schemas of generateObject and of automatic functions are sent as the equivalent OpenAPI schemas; it also has no displayName for inline data, no parts in function responses and no total request timeout (RequestOptions.timeout limits the time without data), and its Live API sends realtime audio as audio/pcm.