@firebase/ai 2.4.0-canary.b7e18d0ff → 2.4.0-canary.c8263c471

package/dist/ai-public.d.ts

@@ -558,6 +558,12 @@ export declare interface EnhancedGenerateContentResponse extends GenerateContent
      * set to `true`.
      */
     thoughtSummary: () => string | undefined;
+    /**
+     * Indicates whether inference happened on-device or in-cloud.
+     *
+     * @beta
+     */
+    inferenceSource?: InferenceSource;
 }
 
 /**
@@ -1833,6 +1839,23 @@ export declare const InferenceMode: {
  */
 export declare type InferenceMode = (typeof InferenceMode)[keyof typeof InferenceMode];
 
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export declare const InferenceSource: {
+    readonly ON_DEVICE: "on_device";
+    readonly IN_CLOUD: "in_cloud";
+};
+
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export declare type InferenceSource = (typeof InferenceSource)[keyof typeof InferenceSource];
+
 /**
  * Content part interface if the part represents an image.
  * @public

package/dist/ai.d.ts

@@ -604,6 +604,12 @@ export declare interface EnhancedGenerateContentResponse extends GenerateContent
      * set to `true`.
      */
     thoughtSummary: () => string | undefined;
+    /**
+     * Indicates whether inference happened on-device or in-cloud.
+     *
+     * @beta
+     */
+    inferenceSource?: InferenceSource;
 }
 
 /**
@@ -1945,6 +1951,23 @@ export declare const InferenceMode: {
  */
 export declare type InferenceMode = (typeof InferenceMode)[keyof typeof InferenceMode];
 
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export declare const InferenceSource: {
+    readonly ON_DEVICE: "on_device";
+    readonly IN_CLOUD: "in_cloud";
+};
+
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+export declare type InferenceSource = (typeof InferenceSource)[keyof typeof InferenceSource];
+
 /**
  * Content part interface if the part represents an image.
  * @public

package/dist/esm/index.esm.js

@@ -4,7 +4,7 @@ import { FirebaseError, Deferred, getModularInstance } from '@firebase/util';
 import { Logger } from '@firebase/logger';
 
 var name = "@firebase/ai";
-var version = "2.4.0-canary.b7e18d0ff";
+var version = "2.4.0-canary.c8263c471";
 
 /**
  * @license
@@ -379,6 +379,15 @@ const InferenceMode = {
     'ONLY_IN_CLOUD': 'only_in_cloud',
     'PREFER_IN_CLOUD': 'prefer_in_cloud'
 };
+/**
+ * Indicates whether inference happened on-device or in-cloud.
+ *
+ * @beta
+ */
+const InferenceSource = {
+    'ON_DEVICE': 'on_device',
+    'IN_CLOUD': 'in_cloud'
+};
 /**
  * Represents the result of the code execution.
  *
@@ -1592,7 +1601,7 @@ function hasValidCandidates(response) {
  * Creates an EnhancedGenerateContentResponse object that has helper functions and
  * other modifications that improve usability.
  */
-function createEnhancedContentResponse(response) {
+function createEnhancedContentResponse(response, inferenceSource = InferenceSource.IN_CLOUD) {
     /**
      * The Vertex AI backend omits default values.
      * This causes the `index` property to be omitted from the first candidate in the
@@ -1603,6 +1612,7 @@ function createEnhancedContentResponse(response) {
         response.candidates[0].index = 0;
     }
     const responseWithHelpers = addHelpers(response);
+    responseWithHelpers.inferenceSource = inferenceSource;
     return responseWithHelpers;
 }
 /**
@@ -1979,16 +1989,16 @@ const responseLineRE = /^data\: (.*)(?:\n\n|\r\r|\r\n\r\n)/;
  *
  * @param response - Response from a fetch call
  */
-function processStream(response, apiSettings) {
+function processStream(response, apiSettings, inferenceSource) {
     const inputStream = response.body.pipeThrough(new TextDecoderStream('utf8', { fatal: true }));
     const responseStream = getResponseStream(inputStream);
     const [stream1, stream2] = responseStream.tee();
     return {
-        stream: generateResponseSequence(stream1, apiSettings),
-        response: getResponsePromise(stream2, apiSettings)
+        stream: generateResponseSequence(stream1, apiSettings, inferenceSource),
+        response: getResponsePromise(stream2, apiSettings, inferenceSource)
     };
 }
-async function getResponsePromise(stream, apiSettings) {
+async function getResponsePromise(stream, apiSettings, inferenceSource) {
     const allResponses = [];
     const reader = stream.getReader();
     while (true) {
@@ -1998,12 +2008,12 @@ async function getResponsePromise(stream, apiSettings) {
             if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
                 generateContentResponse = mapGenerateContentResponse(generateContentResponse);
             }
-            return createEnhancedContentResponse(generateContentResponse);
+            return createEnhancedContentResponse(generateContentResponse, inferenceSource);
         }
         allResponses.push(value);
     }
 }
-async function* generateResponseSequence(stream, apiSettings) {
+async function* generateResponseSequence(stream, apiSettings, inferenceSource) {
     const reader = stream.getReader();
     while (true) {
         const { value, done } = await reader.read();
@@ -2012,10 +2022,10 @@ async function* generateResponseSequence(stream, apiSettings) {
         }
         let enhancedResponse;
         if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
-            enhancedResponse = createEnhancedContentResponse(mapGenerateContentResponse(value));
+            enhancedResponse = createEnhancedContentResponse(mapGenerateContentResponse(value), inferenceSource);
         }
         else {
-            enhancedResponse = createEnhancedContentResponse(value);
+            enhancedResponse = createEnhancedContentResponse(value, inferenceSource);
         }
         const firstCandidate = enhancedResponse.candidates?.[0];
         // Don't yield a response with no useful data for the developer.
@@ -2185,31 +2195,52 @@ const errorsCausingFallback = [
  */
 async function callCloudOrDevice(request, chromeAdapter, onDeviceCall, inCloudCall) {
     if (!chromeAdapter) {
-        return inCloudCall();
+        return {
+            response: await inCloudCall(),
+            inferenceSource: InferenceSource.IN_CLOUD
+        };
     }
     switch (chromeAdapter.mode) {
         case InferenceMode.ONLY_ON_DEVICE:
             if (await chromeAdapter.isAvailable(request)) {
-                return onDeviceCall();
+                return {
+                    response: await onDeviceCall(),
+                    inferenceSource: InferenceSource.ON_DEVICE
+                };
             }
             throw new AIError(AIErrorCode.UNSUPPORTED, 'Inference mode is ONLY_ON_DEVICE, but an on-device model is not available.');
         case InferenceMode.ONLY_IN_CLOUD:
-            return inCloudCall();
+            return {
+                response: await inCloudCall(),
+                inferenceSource: InferenceSource.IN_CLOUD
+            };
         case InferenceMode.PREFER_IN_CLOUD:
             try {
-                return await inCloudCall();
+                return {
+                    response: await inCloudCall(),
+                    inferenceSource: InferenceSource.IN_CLOUD
+                };
             }
             catch (e) {
                 if (e instanceof AIError && errorsCausingFallback.includes(e.code)) {
-                    return onDeviceCall();
+                    return {
+                        response: await onDeviceCall(),
+                        inferenceSource: InferenceSource.ON_DEVICE
+                    };
                 }
                 throw e;
             }
         case InferenceMode.PREFER_ON_DEVICE:
             if (await chromeAdapter.isAvailable(request)) {
-                return onDeviceCall();
+                return {
+                    response: await onDeviceCall(),
+                    inferenceSource: InferenceSource.ON_DEVICE
+                };
             }
-            return inCloudCall();
+            return {
+                response: await inCloudCall(),
+                inferenceSource: InferenceSource.IN_CLOUD
+            };
         default:
             throw new AIError(AIErrorCode.ERROR, `Unexpected infererence mode: ${chromeAdapter.mode}`);
     }
@@ -2239,8 +2270,8 @@ async function generateContentStreamOnCloud(apiSettings, model, params, requestO
     /* stream */ true, JSON.stringify(params), requestOptions);
 }
 async function generateContentStream(apiSettings, model, params, chromeAdapter, requestOptions) {
-    const response = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContentStream(params), () => generateContentStreamOnCloud(apiSettings, model, params, requestOptions));
-    return processStream(response, apiSettings); // TODO: Map streaming responses
+    const callResult = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContentStream(params), () => generateContentStreamOnCloud(apiSettings, model, params, requestOptions));
+    return processStream(callResult.response, apiSettings); // TODO: Map streaming responses
 }
 async function generateContentOnCloud(apiSettings, model, params, requestOptions) {
     if (apiSettings.backend.backendType === BackendType.GOOGLE_AI) {
@@ -2250,9 +2281,9 @@ async function generateContentOnCloud(apiSettings, model, params, requestOptions
     /* stream */ false, JSON.stringify(params), requestOptions);
 }
 async function generateContent(apiSettings, model, params, chromeAdapter, requestOptions) {
-    const response = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContent(params), () => generateContentOnCloud(apiSettings, model, params, requestOptions));
-    const generateContentResponse = await processGenerateContentResponse(response, apiSettings);
-    const enhancedResponse = createEnhancedContentResponse(generateContentResponse);
+    const callResult = await callCloudOrDevice(params, chromeAdapter, () => chromeAdapter.generateContent(params), () => generateContentOnCloud(apiSettings, model, params, requestOptions));
+    const generateContentResponse = await processGenerateContentResponse(callResult.response, apiSettings);
+    const enhancedResponse = createEnhancedContentResponse(generateContentResponse, callResult.inferenceSource);
     return {
         response: enhancedResponse
     };
@@ -4109,5 +4140,5 @@ function registerAI() {
 }
 registerAI();
 
-export { AIError, AIErrorCode, AIModel, AnyOfSchema, ArraySchema, Backend, BackendType, BlockReason, BooleanSchema, ChatSession, FinishReason, FunctionCallingMode, GenerativeModel, GoogleAIBackend, HarmBlockMethod, HarmBlockThreshold, HarmCategory, HarmProbability, HarmSeverity, ImagenAspectRatio, ImagenImageFormat, ImagenModel, ImagenPersonFilterLevel, ImagenSafetyFilterLevel, InferenceMode, IntegerSchema, Language, LiveGenerativeModel, LiveResponseType, LiveSession, Modality, NumberSchema, ObjectSchema, Outcome, POSSIBLE_ROLES, ResponseModality, Schema, SchemaType, StringSchema, URLRetrievalStatus, VertexAIBackend, getAI, getGenerativeModel, getImagenModel, getLiveGenerativeModel, startAudioConversation };
+export { AIError, AIErrorCode, AIModel, AnyOfSchema, ArraySchema, Backend, BackendType, BlockReason, BooleanSchema, ChatSession, FinishReason, FunctionCallingMode, GenerativeModel, GoogleAIBackend, HarmBlockMethod, HarmBlockThreshold, HarmCategory, HarmProbability, HarmSeverity, ImagenAspectRatio, ImagenImageFormat, ImagenModel, ImagenPersonFilterLevel, ImagenSafetyFilterLevel, InferenceMode, InferenceSource, IntegerSchema, Language, LiveGenerativeModel, LiveResponseType, LiveSession, Modality, NumberSchema, ObjectSchema, Outcome, POSSIBLE_ROLES, ResponseModality, Schema, SchemaType, StringSchema, URLRetrievalStatus, VertexAIBackend, getAI, getGenerativeModel, getImagenModel, getLiveGenerativeModel, startAudioConversation };
 //# sourceMappingURL=index.esm.js.map