@capgo/capacitor-llm 8.0.19 → 8.1.0

package/android/src/main/java/ee/forgr/capgo_llm/LLMPlugin.java

@@ -1,18 +1,15 @@
 package ee.forgr.capgo_llm;
 
-import com.getcapacitor.JSArray;
 import com.getcapacitor.JSObject;
 import com.getcapacitor.Plugin;
 import com.getcapacitor.PluginCall;
 import com.getcapacitor.PluginMethod;
 import com.getcapacitor.annotation.CapacitorPlugin;
-import java.util.ArrayList;
-import java.util.List;
 
 @CapacitorPlugin(name = "CapgoLLM")
 public class LLMPlugin extends Plugin {
 
-    private final String pluginVersion = "8.0.19";
+    private final String pluginVersion = "8.1.0";
 
     private LLM llm;
 
@@ -44,34 +41,41 @@ public class LLMPlugin extends Plugin {
             return;
         }
 
-        llm.sendMessage(
-            chatId,
-            message,
-            new LLM.MessageCallback() {
-                @Override
-                public void onTextReceived(String chatId, String text, boolean isChunk) {
-                    JSObject event = new JSObject();
-                    event.put("text", text);
-                    event.put("chatId", chatId);
-                    event.put("isChunk", isChunk);
-                    notifyListeners("textFromAi", event);
-                }
-
-                @Override
-                public void onComplete(String chatId) {
-                    JSObject event = new JSObject();
-                    event.put("chatId", chatId);
-                    notifyListeners("aiFinished", event);
-                }
+        try {
+            llm.sendMessage(
+                chatId,
+                message,
+                new LLM.MessageCallback() {
+                    @Override
+                    public void onTextReceived(String chatId, String text, boolean isChunk) {
+                        JSObject event = new JSObject();
+                        event.put("text", text);
+                        event.put("chatId", chatId);
+                        event.put("isChunk", isChunk);
+                        notifyListeners("textFromAi", event);
+                    }
+
+                    @Override
+                    public void onComplete(String chatId) {
+                        JSObject event = new JSObject();
+                        event.put("chatId", chatId);
+                        notifyListeners("aiFinished", event);
+                    }
 
-                @Override
-                public void onError(String error) {
-                    call.reject("Failed to send message", error);
+                    @Override
+                    public void onError(String error) {
+                        JSObject event = new JSObject();
+                        event.put("chatId", chatId);
+                        event.put("error", error);
+                        notifyListeners("generationError", event);
+                    }
                 }
-            }
-        );
+            );
 
-        call.resolve();
+            call.resolve();
+        } catch (IllegalStateException exception) {
+            call.reject(exception.getMessage());
+        }
     }
 
     @PluginMethod
@@ -97,35 +101,19 @@ public class LLMPlugin extends Plugin {
         }
 
         // Extract model parameters
-        String engine = call.getString("engine", "auto");
-        String modelType = call.getString("modelType");
-        String tokenizerPath = call.getString("tokenizerPath");
         Integer maxTokens = call.getInt("maxTokens", 2048);
-        Integer sequenceLength = call.getInt("sequenceLength", maxTokens);
         Integer topk = call.getInt("topk", 40);
         Float temperature = call.getFloat("temperature", 0.8f);
-        List<String> specialTokens = getStringList(call.getArray("specialTokens"));
-
-        if (sequenceLength <= 0) {
-            call.reject("sequenceLength must be > 0");
-            return;
-        }
-
-        if (!specialTokens.isEmpty()) {
-            call.reject("specialTokens are only supported by ExecuTorch on iOS");
-            return;
-        }
+        Integer randomSeed = call.getInt("randomSeed", 0);
+        String modelType = call.getString("modelType");
 
         llm.setModel(
             path,
-            engine,
             modelType,
-            tokenizerPath,
-            specialTokens,
             maxTokens,
-            sequenceLength,
             topk,
             temperature,
+            randomSeed,
             new LLM.ModelLoadCallback() {
                 @Override
                 public void onSuccess() {
@@ -204,21 +192,6 @@ public class LLMPlugin extends Plugin {
             .start();
     }
 
-    private List<String> getStringList(JSArray values) {
-        List<String> result = new ArrayList<>();
-        if (values == null) {
-            return result;
-        }
-
-        for (int i = 0; i < values.length(); i++) {
-            String value = values.optString(i, null);
-            if (value != null) {
-                result.add(value);
-            }
-        }
-        return result;
-    }
-
     private void downloadFile(String urlString, java.io.File outputFile, ProgressCallback callback) throws Exception {
         java.net.URL url = new java.net.URL(urlString);
         java.net.HttpURLConnection connection = (java.net.HttpURLConnection) url.openConnection();

package/dist/docs.json

@@ -81,7 +81,7 @@
             "text": "Promise that resolves when model is loaded"
           }
         ],
-        "docs": "Sets the model configuration\n- iOS: Use \"Apple Intelligence\" as path for system model, provide a MediaPipe model, or set engine to \"executorch\"\n- Android: Path to a MediaPipe or ExecuTorch model file (in assets or files directory)",
+        "docs": "Sets the model configuration\n- iOS: Use \"Apple Intelligence\" as path for the system model. Custom LiteRT-LM `.litertlm` models are supported on iOS only when this plugin is integrated through Swift Package Manager.\n- Android: Use \"Gemini Nano\" for the AICore system model on supported devices where it is already available. Prefer LiteRT-LM `.litertlm` bundles for custom models; legacy MediaPipe `.task` models are still supported\n- Web: Provide a web-ready model asset for `@mediapipe/tasks-genai` such as Gemma 4 `*-web.task`",
         "complexTypes": [
           "ModelOptions"
         ],
@@ -187,6 +187,42 @@
         ],
         "slug": "addlisteneraifinished-"
       },
+      {
+        "name": "addListener",
+        "signature": "(eventName: 'generationError', listenerFunc: (event: GenerationErrorEvent) => void) => Promise<{ remove: () => Promise<void>; }>",
+        "parameters": [
+          {
+            "name": "eventName",
+            "docs": "- Event name 'generationError'",
+            "type": "'generationError'"
+          },
+          {
+            "name": "listenerFunc",
+            "docs": "- Callback function for generation errors",
+            "type": "(event: GenerationErrorEvent) => void"
+          }
+        ],
+        "returns": "Promise<{ remove: () => Promise<void>; }>",
+        "tags": [
+          {
+            "name": "param",
+            "text": "eventName - Event name 'generationError'"
+          },
+          {
+            "name": "param",
+            "text": "listenerFunc - Callback function for generation errors"
+          },
+          {
+            "name": "returns",
+            "text": "Promise with remove function to unsubscribe"
+          }
+        ],
+        "docs": "Adds a listener for generation failures that happen after streaming starts",
+        "complexTypes": [
+          "GenerationErrorEvent"
+        ],
+        "slug": "addlistenergenerationerror-"
+      },
       {
         "name": "addListener",
         "signature": "(eventName: 'downloadProgress', listenerFunc: (event: DownloadProgressEvent) => void) => Promise<{ remove: () => Promise<void>; }>",
@@ -293,45 +329,24 @@
     {
       "name": "ModelOptions",
       "slug": "modeloptions",
-      "docs": "Model configuration options",
+      "docs": "Model configuration options\nOnly `path` is required. All other properties are optional overrides.",
       "tags": [],
       "methods": [],
       "properties": [
         {
           "name": "path",
           "tags": [],
-          "docs": "Model path or \"Apple Intelligence\" for iOS system model",
+          "docs": "Model path, \"Apple Intelligence\" for the Apple system model on iOS, or \"Gemini Nano\" for the Android AICore system model when already available on the device. On iOS, custom `.litertlm` models require the plugin to be integrated through Swift Package Manager. Gemma 4 examples use `.litertlm` on mobile and `*-web.task` on web.",
           "complexTypes": [],
           "type": "string"
         },
-        {
-          "name": "engine",
-          "tags": [],
-          "docs": "Runtime engine to use.\n- \"auto\": uses Apple Intelligence on iOS when path is \"Apple Intelligence\", ExecuTorch for `.pte` or tokenizerPath, otherwise MediaPipe\n- \"apple\": iOS Foundation Models / Apple Intelligence\n- \"mediapipe\": MediaPipe GenAI `.task` models\n- \"executorch\": ExecuTorch `.pte` models with a tokenizer file",
-          "complexTypes": [],
-          "type": "'auto' | 'apple' | 'mediapipe' | 'executorch' | undefined"
-        },
         {
           "name": "modelType",
           "tags": [],
-          "docs": "Model file type/extension (e.g., \"task\", \"bin\", \"litertlm\"). If not provided, will be extracted from path.",
+          "docs": "Optional. Model file type/extension (for example `task`, `bin`, `litertlm`, or `gemini-nano`). If not provided, it is extracted from the path. Use `litertlm` on iOS only in SwiftPM integrations.",
           "complexTypes": [],
           "type": "string | undefined"
         },
-        {
-          "name": "tokenizerPath",
-          "tags": [],
-          "docs": "Tokenizer path for ExecuTorch models. Required when engine is \"executorch\".",
-          "complexTypes": [],
-          "type": "string | undefined"
-        },
-        {
-          "name": "specialTokens",
-          "tags": [],
-          "docs": "Optional special tokens passed to iOS ExecuTorch tokenizers.",
-          "complexTypes": [],
-          "type": "string[] | undefined"
-        },
         {
           "name": "maxTokens",
           "tags": [],
@@ -339,13 +354,6 @@
           "complexTypes": [],
           "type": "number | undefined"
         },
-        {
-          "name": "sequenceLength",
-          "tags": [],
-          "docs": "Sequence length for ExecuTorch generation. Defaults to maxTokens when omitted.",
-          "complexTypes": [],
-          "type": "number | undefined"
-        },
         {
           "name": "topk",
           "tags": [],
@@ -363,7 +371,7 @@
         {
           "name": "randomSeed",
           "tags": [],
-          "docs": "Random seed for generation",
+          "docs": "Optional. Random seed for generation.",
           "complexTypes": [],
           "type": "number | undefined"
         }
@@ -395,7 +403,7 @@
     {
       "name": "DownloadModelOptions",
       "slug": "downloadmodeloptions",
-      "docs": "Options for downloading a model",
+      "docs": "Options for downloading a model\nOnly `url` is required. `companionUrl` and `filename` are optional.",
       "tags": [],
       "methods": [],
       "properties": [
@@ -409,7 +417,7 @@
         {
           "name": "companionUrl",
           "tags": [],
-          "docs": "Optional: URL of companion file (e.g., .litertlm for Android)",
+          "docs": "Optional: URL of a companion file for legacy model formats",
           "complexTypes": [],
           "type": "string | undefined"
         },
@@ -468,6 +476,29 @@
         }
       ]
     },
+    {
+      "name": "GenerationErrorEvent",
+      "slug": "generationerrorevent",
+      "docs": "Event data for generation failures\n`chatId` is optional and may be omitted when the failure is not tied to a specific chat.",
+      "tags": [],
+      "methods": [],
+      "properties": [
+        {
+          "name": "chatId",
+          "tags": [],
+          "docs": "Optional. The chat session ID that failed, when available.",
+          "complexTypes": [],
+          "type": "string | undefined"
+        },
+        {
+          "name": "error",
+          "tags": [],
+          "docs": "Error message describing the failure",
+          "complexTypes": [],
+          "type": "string"
+        }
+      ]
+    },
     {
       "name": "DownloadProgressEvent",
       "slug": "downloadprogressevent",

package/dist/esm/definitions.d.ts

@@ -28,8 +28,9 @@ export interface LLMPlugin {
     }>;
     /**
      * Sets the model configuration
-     * - iOS: Use "Apple Intelligence" as path for system model, provide a MediaPipe model, or set engine to "executorch"
-     * - Android: Path to a MediaPipe or ExecuTorch model file (in assets or files directory)
+     * - iOS: Use "Apple Intelligence" as path for the system model. Custom LiteRT-LM `.litertlm` models are supported on iOS only when this plugin is integrated through Swift Package Manager.
+     * - Android: Use "Gemini Nano" for the AICore system model on supported devices where it is already available. Prefer LiteRT-LM `.litertlm` bundles for custom models; legacy MediaPipe `.task` models are still supported
+     * - Web: Provide a web-ready model asset for `@mediapipe/tasks-genai` such as Gemma 4 `*-web.task`
      * @param options - The model configuration
      * @returns Promise that resolves when model is loaded
      */
@@ -60,6 +61,15 @@ export interface LLMPlugin {
     addListener(eventName: 'aiFinished', listenerFunc: (event: AiFinishedEvent) => void): Promise<{
         remove: () => Promise<void>;
     }>;
+    /**
+     * Adds a listener for generation failures that happen after streaming starts
+     * @param eventName - Event name 'generationError'
+     * @param listenerFunc - Callback function for generation errors
+     * @returns Promise with remove function to unsubscribe
+     */
+    addListener(eventName: 'generationError', listenerFunc: (event: GenerationErrorEvent) => void): Promise<{
+        remove: () => Promise<void>;
+    }>;
     /**
      * Adds a listener for model download progress events
      * @param eventName - Event name 'downloadProgress'
@@ -112,13 +122,24 @@ export interface AiFinishedEvent {
     /** The chat session ID that finished */
     chatId: string;
 }
+/**
+ * Event data for generation failures
+ * `chatId` is optional and may be omitted when the failure is not tied to a specific chat.
+ */
+export interface GenerationErrorEvent {
+    /** Optional. The chat session ID that failed, when available. */
+    chatId?: string;
+    /** Error message describing the failure */
+    error: string;
+}
 /**
  * Options for downloading a model
+ * Only `url` is required. `companionUrl` and `filename` are optional.
  */
 export interface DownloadModelOptions {
     /** URL of the model file to download */
     url: string;
-    /** Optional: URL of companion file (e.g., .litertlm for Android) */
+    /** Optional: URL of a companion file for legacy model formats */
     companionUrl?: string;
     /** Optional: Custom filename (defaults to filename from URL) */
     filename?: string;
@@ -152,32 +173,19 @@ export interface ReadinessChangeEvent {
 }
 /**
  * Model configuration options
+ * Only `path` is required. All other properties are optional overrides.
  */
 export interface ModelOptions {
-    /** Model path or "Apple Intelligence" for iOS system model */
+    /** Model path, "Apple Intelligence" for the Apple system model on iOS, or "Gemini Nano" for the Android AICore system model when already available on the device. On iOS, custom `.litertlm` models require the plugin to be integrated through Swift Package Manager. Gemma 4 examples use `.litertlm` on mobile and `*-web.task` on web. */
     path: string;
-    /**
-     * Runtime engine to use.
-     * - "auto": uses Apple Intelligence on iOS when path is "Apple Intelligence", ExecuTorch for `.pte` or tokenizerPath, otherwise MediaPipe
-     * - "apple": iOS Foundation Models / Apple Intelligence
-     * - "mediapipe": MediaPipe GenAI `.task` models
-     * - "executorch": ExecuTorch `.pte` models with a tokenizer file
-     */
-    engine?: 'auto' | 'apple' | 'mediapipe' | 'executorch';
-    /** Model file type/extension (e.g., "task", "bin", "litertlm"). If not provided, will be extracted from path. */
+    /** Optional. Model file type/extension (for example `task`, `bin`, `litertlm`, or `gemini-nano`). If not provided, it is extracted from the path. Use `litertlm` on iOS only in SwiftPM integrations. */
     modelType?: string;
-    /** Tokenizer path for ExecuTorch models. Required when engine is "executorch". */
-    tokenizerPath?: string;
-    /** Optional special tokens passed to iOS ExecuTorch tokenizers. */
-    specialTokens?: string[];
     /** Maximum number of tokens the model handles */
     maxTokens?: number;
-    /** Sequence length for ExecuTorch generation. Defaults to maxTokens when omitted. */
-    sequenceLength?: number;
     /** Number of tokens the model considers at each step */
     topk?: number;
     /** Amount of randomness in generation (0.0-1.0) */
     temperature?: number;
-    /** Random seed for generation */
+    /** Optional. Random seed for generation. */
     randomSeed?: number;
 }

package/dist/esm/definitions.js.map

@@ -1 +1 @@
-{"version":3,"file":"definitions.js","sourceRoot":"","sources":["../../src/definitions.ts"],"names":[],"mappings":"","sourcesContent":["/**\n * LLM Plugin interface for interacting with on-device language models\n */\nexport interface LLMPlugin {\n  /**\n   * Creates a new chat session\n   * @returns Promise with chat id and optional instructions\n   */\n  createChat(): Promise<{ id: string; instructions?: string }>;\n\n  /**\n   * Sends a message to the AI in a specific chat session\n   * @param options - The chat id and message to send\n   * @returns Promise that resolves when message is sent\n   */\n  sendMessage(options: { chatId: string; message: string }): Promise<void>;\n\n  /**\n   * Gets the readiness status of the LLM\n   * @returns Promise with readiness status string\n   */\n  getReadiness(): Promise<{ readiness: string }>;\n\n  /**\n   * Sets the model configuration\n   * - iOS: Use \"Apple Intelligence\" as path for system model, provide a MediaPipe model, or set engine to \"executorch\"\n   * - Android: Path to a MediaPipe or ExecuTorch model file (in assets or files directory)\n   * @param options - The model configuration\n   * @returns Promise that resolves when model is loaded\n   */\n  setModel(options: ModelOptions): Promise<void>;\n\n  /**\n   * Downloads a model from a URL and saves it to the appropriate location\n   * - iOS: Downloads to the app's documents directory\n   * - Android: Downloads to the app's files directory\n   * @param options - The download configuration\n   * @returns Promise with download result including the saved path\n   */\n  downloadModel(options: DownloadModelOptions): Promise<DownloadModelResult>;\n\n  /**\n   * Adds a listener for text received from AI\n   * @param eventName - Event name 'textFromAi'\n   * @param listenerFunc - Callback function for text events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'textFromAi',\n    listenerFunc: (event: TextFromAiEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for AI completion events\n   * @param eventName - Event name 'aiFinished'\n   * @param listenerFunc - Callback function for finish events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'aiFinished',\n    listenerFunc: (event: AiFinishedEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for model download progress events\n   * @param eventName - Event name 'downloadProgress'\n   * @param listenerFunc - Callback function for progress events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'downloadProgress',\n    listenerFunc: (event: DownloadProgressEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for readiness status changes\n   * @param eventName - Event name 'readinessChange'\n   * @param listenerFunc - Callback function for readiness events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'readinessChange',\n    listenerFunc: (event: ReadinessChangeEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Get the native Capacitor plugin version.\n   *\n   * @returns Promise that resolves with the plugin version\n   * @throws Error if getting the version fails\n   * @since 1.0.0\n   * @example\n   * ```typescript\n   * const { version } = await CapgoLLM.getPluginVersion();\n   * console.log('Plugin version:', version);\n   * ```\n   */\n  getPluginVersion(): Promise<{ version: string }>;\n}\n\n/**\n * Event data for text received from AI\n */\nexport interface TextFromAiEvent {\n  /** The text content from AI - this is an incremental chunk, not the full text */\n  text: string;\n  /** The chat session ID */\n  chatId: string;\n  /** Whether this is a complete chunk (true) or partial streaming data (false) */\n  isChunk?: boolean;\n}\n\n/**\n * Event data for AI completion\n */\nexport interface AiFinishedEvent {\n  /** The chat session ID that finished */\n  chatId: string;\n}\n\n/**\n * Options for downloading a model\n */\nexport interface DownloadModelOptions {\n  /** URL of the model file to download */\n  url: string;\n  /** Optional: URL of companion file (e.g., .litertlm for Android) */\n  companionUrl?: string;\n  /** Optional: Custom filename (defaults to filename from URL) */\n  filename?: string;\n}\n\n/**\n * Result of model download\n */\nexport interface DownloadModelResult {\n  /** Path where the model was saved */\n  path: string;\n  /** Path where the companion file was saved (if applicable) */\n  companionPath?: string;\n}\n\n/**\n * Event data for download progress\n */\nexport interface DownloadProgressEvent {\n  /** Percentage of download completed (0-100) */\n  progress: number;\n  /** Total bytes to download */\n  totalBytes?: number;\n  /** Bytes downloaded so far */\n  downloadedBytes?: number;\n}\n\n/**\n * Event data for readiness status changes\n */\nexport interface ReadinessChangeEvent {\n  /** The readiness status */\n  readiness: string;\n}\n\n/**\n * Model configuration options\n */\nexport interface ModelOptions {\n  /** Model path or \"Apple Intelligence\" for iOS system model */\n  path: string;\n  /**\n   * Runtime engine to use.\n   * - \"auto\": uses Apple Intelligence on iOS when path is \"Apple Intelligence\", ExecuTorch for `.pte` or tokenizerPath, otherwise MediaPipe\n   * - \"apple\": iOS Foundation Models / Apple Intelligence\n   * - \"mediapipe\": MediaPipe GenAI `.task` models\n   * - \"executorch\": ExecuTorch `.pte` models with a tokenizer file\n   */\n  engine?: 'auto' | 'apple' | 'mediapipe' | 'executorch';\n  /** Model file type/extension (e.g., \"task\", \"bin\", \"litertlm\"). If not provided, will be extracted from path. */\n  modelType?: string;\n  /** Tokenizer path for ExecuTorch models. Required when engine is \"executorch\". */\n  tokenizerPath?: string;\n  /** Optional special tokens passed to iOS ExecuTorch tokenizers. */\n  specialTokens?: string[];\n  /** Maximum number of tokens the model handles */\n  maxTokens?: number;\n  /** Sequence length for ExecuTorch generation. Defaults to maxTokens when omitted. */\n  sequenceLength?: number;\n  /** Number of tokens the model considers at each step */\n  topk?: number;\n  /** Amount of randomness in generation (0.0-1.0) */\n  temperature?: number;\n  /** Random seed for generation */\n  randomSeed?: number;\n}\n"]}
+{"version":3,"file":"definitions.js","sourceRoot":"","sources":["../../src/definitions.ts"],"names":[],"mappings":"","sourcesContent":["/**\n * LLM Plugin interface for interacting with on-device language models\n */\nexport interface LLMPlugin {\n  /**\n   * Creates a new chat session\n   * @returns Promise with chat id and optional instructions\n   */\n  createChat(): Promise<{ id: string; instructions?: string }>;\n\n  /**\n   * Sends a message to the AI in a specific chat session\n   * @param options - The chat id and message to send\n   * @returns Promise that resolves when message is sent\n   */\n  sendMessage(options: { chatId: string; message: string }): Promise<void>;\n\n  /**\n   * Gets the readiness status of the LLM\n   * @returns Promise with readiness status string\n   */\n  getReadiness(): Promise<{ readiness: string }>;\n\n  /**\n   * Sets the model configuration\n   * - iOS: Use \"Apple Intelligence\" as path for the system model. Custom LiteRT-LM `.litertlm` models are supported on iOS only when this plugin is integrated through Swift Package Manager.\n   * - Android: Use \"Gemini Nano\" for the AICore system model on supported devices where it is already available. Prefer LiteRT-LM `.litertlm` bundles for custom models; legacy MediaPipe `.task` models are still supported\n   * - Web: Provide a web-ready model asset for `@mediapipe/tasks-genai` such as Gemma 4 `*-web.task`\n   * @param options - The model configuration\n   * @returns Promise that resolves when model is loaded\n   */\n  setModel(options: ModelOptions): Promise<void>;\n\n  /**\n   * Downloads a model from a URL and saves it to the appropriate location\n   * - iOS: Downloads to the app's documents directory\n   * - Android: Downloads to the app's files directory\n   * @param options - The download configuration\n   * @returns Promise with download result including the saved path\n   */\n  downloadModel(options: DownloadModelOptions): Promise<DownloadModelResult>;\n\n  /**\n   * Adds a listener for text received from AI\n   * @param eventName - Event name 'textFromAi'\n   * @param listenerFunc - Callback function for text events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'textFromAi',\n    listenerFunc: (event: TextFromAiEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for AI completion events\n   * @param eventName - Event name 'aiFinished'\n   * @param listenerFunc - Callback function for finish events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'aiFinished',\n    listenerFunc: (event: AiFinishedEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for generation failures that happen after streaming starts\n   * @param eventName - Event name 'generationError'\n   * @param listenerFunc - Callback function for generation errors\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'generationError',\n    listenerFunc: (event: GenerationErrorEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for model download progress events\n   * @param eventName - Event name 'downloadProgress'\n   * @param listenerFunc - Callback function for progress events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'downloadProgress',\n    listenerFunc: (event: DownloadProgressEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Adds a listener for readiness status changes\n   * @param eventName - Event name 'readinessChange'\n   * @param listenerFunc - Callback function for readiness events\n   * @returns Promise with remove function to unsubscribe\n   */\n  addListener(\n    eventName: 'readinessChange',\n    listenerFunc: (event: ReadinessChangeEvent) => void,\n  ): Promise<{ remove: () => Promise<void> }>;\n\n  /**\n   * Get the native Capacitor plugin version.\n   *\n   * @returns Promise that resolves with the plugin version\n   * @throws Error if getting the version fails\n   * @since 1.0.0\n   * @example\n   * ```typescript\n   * const { version } = await CapgoLLM.getPluginVersion();\n   * console.log('Plugin version:', version);\n   * ```\n   */\n  getPluginVersion(): Promise<{ version: string }>;\n}\n\n/**\n * Event data for text received from AI\n */\nexport interface TextFromAiEvent {\n  /** The text content from AI - this is an incremental chunk, not the full text */\n  text: string;\n  /** The chat session ID */\n  chatId: string;\n  /** Whether this is a complete chunk (true) or partial streaming data (false) */\n  isChunk?: boolean;\n}\n\n/**\n * Event data for AI completion\n */\nexport interface AiFinishedEvent {\n  /** The chat session ID that finished */\n  chatId: string;\n}\n\n/**\n * Event data for generation failures\n * `chatId` is optional and may be omitted when the failure is not tied to a specific chat.\n */\nexport interface GenerationErrorEvent {\n  /** Optional. The chat session ID that failed, when available. */\n  chatId?: string;\n  /** Error message describing the failure */\n  error: string;\n}\n\n/**\n * Options for downloading a model\n * Only `url` is required. `companionUrl` and `filename` are optional.\n */\nexport interface DownloadModelOptions {\n  /** URL of the model file to download */\n  url: string;\n  /** Optional: URL of a companion file for legacy model formats */\n  companionUrl?: string;\n  /** Optional: Custom filename (defaults to filename from URL) */\n  filename?: string;\n}\n\n/**\n * Result of model download\n */\nexport interface DownloadModelResult {\n  /** Path where the model was saved */\n  path: string;\n  /** Path where the companion file was saved (if applicable) */\n  companionPath?: string;\n}\n\n/**\n * Event data for download progress\n */\nexport interface DownloadProgressEvent {\n  /** Percentage of download completed (0-100) */\n  progress: number;\n  /** Total bytes to download */\n  totalBytes?: number;\n  /** Bytes downloaded so far */\n  downloadedBytes?: number;\n}\n\n/**\n * Event data for readiness status changes\n */\nexport interface ReadinessChangeEvent {\n  /** The readiness status */\n  readiness: string;\n}\n\n/**\n * Model configuration options\n * Only `path` is required. All other properties are optional overrides.\n */\nexport interface ModelOptions {\n  /** Model path, \"Apple Intelligence\" for the Apple system model on iOS, or \"Gemini Nano\" for the Android AICore system model when already available on the device. On iOS, custom `.litertlm` models require the plugin to be integrated through Swift Package Manager. Gemma 4 examples use `.litertlm` on mobile and `*-web.task` on web. */\n  path: string;\n  /** Optional. Model file type/extension (for example `task`, `bin`, `litertlm`, or `gemini-nano`). If not provided, it is extracted from the path. Use `litertlm` on iOS only in SwiftPM integrations. */\n  modelType?: string;\n  /** Maximum number of tokens the model handles */\n  maxTokens?: number;\n  /** Number of tokens the model considers at each step */\n  topk?: number;\n  /** Amount of randomness in generation (0.0-1.0) */\n  temperature?: number;\n  /** Optional. Random seed for generation. */\n  randomSeed?: number;\n}\n"]}

package/dist/esm/web.d.ts

@@ -4,6 +4,8 @@ export declare class CapgoLLMWeb extends WebPlugin implements LLMPlugin {
     private llm;
     private chatSessions;
     private readiness;
+    private modelPath;
+    private modelType;
     getReadiness(): Promise<{
         readiness: string;
     }>;
@@ -20,4 +22,8 @@ export declare class CapgoLLMWeb extends WebPlugin implements LLMPlugin {
     getPluginVersion(): Promise<{
         version: string;
     }>;
+    private closeCurrentModel;
+    private resolveModelType;
+    private buildPrompt;
+    private usesGemmaChatTemplate;
 }

package/dist/esm/web.js

@@ -1,10 +1,3 @@
-var __asyncValues = (this && this.__asyncValues) || function (o) {
-    if (!Symbol.asyncIterator) throw new TypeError("Symbol.asyncIterator is not defined.");
-    var m = o[Symbol.asyncIterator], i;
-    return m ? m.call(o) : (o = typeof __values === "function" ? __values(o) : o[Symbol.iterator](), i = {}, verb("next"), verb("throw"), verb("return"), i[Symbol.asyncIterator] = function () { return this; }, i);
-    function verb(n) { i[n] = o[n] && function (v) { return new Promise(function (resolve, reject) { v = o[n](v), settle(resolve, reject, v.done, v.value); }); }; }
-    function settle(resolve, reject, d, v) { Promise.resolve(v).then(function(v) { resolve({ value: v, done: d }); }, reject); }
-};
 import { WebPlugin } from '@capacitor/core';
 import { FilesetResolver, LlmInference } from '@mediapipe/tasks-genai';
 export class CapgoLLMWeb extends WebPlugin {
@@ -13,6 +6,8 @@ export class CapgoLLMWeb extends WebPlugin {
         this.llm = null;
         this.chatSessions = new Map();
         this.readiness = 'not-loaded';
+        this.modelPath = '';
+        this.modelType = 'task';
     }
     async getReadiness() {
         return { readiness: this.readiness };
@@ -26,11 +21,13 @@ export class CapgoLLMWeb extends WebPlugin {
             id: chatId,
             llm: this.llm,
             isActive: true,
+            modelPath: this.modelPath,
+            modelType: this.modelType,
+            history: [],
         });
         return { id: chatId };
     }
     async sendMessage(options) {
-        var _a, e_1, _b, _c;
         const session = this.chatSessions.get(options.chatId);
         if (!session) {
             throw new Error(`Chat session ${options.chatId} not found`);
@@ -38,29 +35,31 @@ export class CapgoLLMWeb extends WebPlugin {
         if (!session.isActive) {
             throw new Error(`Chat session ${options.chatId} is not active`);
         }
+        let hasStreamed = false;
+        let responseText = '';
         try {
-            // Generate response using MediaPipe GenAI streaming API
-            const response = session.llm.generateResponseStream(options.message);
-            try {
-                for (var _d = true, response_1 = __asyncValues(response), response_1_1; response_1_1 = await response_1.next(), _a = response_1_1.done, !_a; _d = true) {
-                    _c = response_1_1.value;
-                    _d = false;
-                    const partialResponse = _c;
-                    // Send incremental text
-                    this.notifyListeners('textFromAi', {
-                        text: partialResponse,
-                        chatId: options.chatId,
-                        isChunk: true,
-                    });
+            const prompt = this.buildPrompt(session, options.message);
+            const finalResponse = await session.llm.generateResponse(prompt, (partialResponse, done) => {
+                if (done || !partialResponse) {
+                    return;
                 }
+                hasStreamed = true;
+                responseText += partialResponse;
+                this.notifyListeners('textFromAi', {
+                    text: partialResponse,
+                    chatId: options.chatId,
+                    isChunk: true,
+                });
+            });
+            if (!hasStreamed && finalResponse) {
+                responseText = finalResponse;
+                this.notifyListeners('textFromAi', {
+                    text: finalResponse,
+                    chatId: options.chatId,
+                    isChunk: true,
+                });
             }
-            catch (e_1_1) { e_1 = { error: e_1_1 }; }
-            finally {
-                try {
-                    if (!_d && !_a && (_b = response_1.return)) await _b.call(response_1);
-                }
-                finally { if (e_1) throw e_1.error; }
-            }
+            session.history.push({ role: 'user', content: options.message }, { role: 'model', content: responseText || finalResponse });
             // Notify completion
             this.notifyListeners('aiFinished', {
                 chatId: options.chatId,
@@ -68,22 +67,20 @@ export class CapgoLLMWeb extends WebPlugin {
         }
         catch (error) {
             console.error('Error generating response:', error);
+            const message = error instanceof Error ? error.message : String(error);
+            if (hasStreamed) {
+                this.notifyListeners('generationError', {
+                    chatId: options.chatId,
+                    error: message,
+                });
+                return;
+            }
             throw error;
         }
     }
     async setModel(options) {
-        var _a;
-        const engine = (_a = options.engine) !== null && _a !== void 0 ? _a : 'auto';
-        const normalizedPath = options.path.toLowerCase();
-        const wantsApple = engine === 'apple' || (engine === 'auto' && normalizedPath === 'apple intelligence');
-        const wantsExecuTorch = engine === 'executorch' || (engine === 'auto' && (options.tokenizerPath || normalizedPath.endsWith('.pte')));
-        if (wantsApple) {
-            throw new Error('Apple Intelligence is only available on native iOS.');
-        }
-        if (wantsExecuTorch) {
-            throw new Error('ExecuTorch is only available on native iOS and Android.');
-        }
         try {
+            this.closeCurrentModel();
             // Update readiness
             this.readiness = 'loading';
             this.notifyListeners('readinessChange', { readiness: this.readiness });
@@ -100,11 +97,15 @@ export class CapgoLLMWeb extends WebPlugin {
             const genai = await FilesetResolver.forGenAiTasks('https://cdn.jsdelivr.net/npm/@mediapipe/tasks-genai@latest/wasm');
             // Create LLM instance
             this.llm = await LlmInference.createFromOptions(genai, config);
+            this.modelPath = options.path;
+            this.modelType = this.resolveModelType(options);
+            this.chatSessions.clear();
             // Update readiness
             this.readiness = 'ready';
             this.notifyListeners('readinessChange', { readiness: this.readiness });
         }
         catch (error) {
+            this.closeCurrentModel();
             this.readiness = 'error';
             this.notifyListeners('readinessChange', { readiness: this.readiness });
             throw error;
@@ -168,5 +169,33 @@ export class CapgoLLMWeb extends WebPlugin {
     async getPluginVersion() {
         return { version: 'web' };
     }
+    closeCurrentModel() {
+        var _a;
+        this.chatSessions.clear();
+        (_a = this.llm) === null || _a === void 0 ? void 0 : _a.close();
+        this.llm = null;
+        this.modelPath = '';
+        this.modelType = 'task';
+    }
+    resolveModelType(options) {
+        var _a;
+        if ((_a = options.modelType) === null || _a === void 0 ? void 0 : _a.trim()) {
+            return options.modelType.trim().toLowerCase();
+        }
+        const extension = options.path.split('.').pop();
+        return extension ? extension.toLowerCase() : 'task';
+    }
+    buildPrompt(session, message) {
+        if (!this.usesGemmaChatTemplate(session)) {
+            return message;
+        }
+        const history = [...session.history, { role: 'user', content: message }];
+        return `${history
+            .map((turn) => `<start_of_turn>${turn.role}\n${turn.content}<end_of_turn>`)
+            .join('\n')}\n<start_of_turn>model\n`;
+    }
+    usesGemmaChatTemplate(session) {
+        return /gemma/i.test(session.modelPath) || session.modelType === 'litertlm';
+    }
 }
 //# sourceMappingURL=web.js.map