deploy: current vibn theia state

Made-with: Cursor

packages/ai-google/.eslintrc.js

@@ -0,0 +1,10 @@
+/** @type {import('eslint').Linter.Config} */
+module.exports = {
+    extends: [
+        '../../configs/build.eslintrc.json'
+    ],
+    parserOptions: {
+        tsconfigRootDir: __dirname,
+        project: 'tsconfig.json'
+    }
+};

packages/ai-google/README.md

@@ -0,0 +1,33 @@
+<div align='center'>
+
+<br />
+
+<img src='https://raw.githubusercontent.com/eclipse-theia/theia/master/logo/theia.svg?sanitize=true' alt='theia-ext-logo' width='100px' />
+
+<h2>ECLIPSE THEIA - AI GOOGLE EXTENSION</h2>
+
+<hr />
+
+</div>
+
+## Description
+
+The `@theia/ai-google` integrates Google's models with Theia AI.
+The Google API key and the models to use can be configured via preferences.
+Alternatively the API key can also be handed in via the `GOOGLE_API_KEY` environment variable.
+
+## Additional Information
+
+- [API documentation for `@theia/ai-google`](https://eclipse-theia.github.io/theia/docs/next/modules/_theia_ai-google.html)
+- [Theia - GitHub](https://github.com/eclipse-theia/theia)
+- [Theia - Website](https://theia-ide.org/)
+
+## License
+
+- [Eclipse Public License 2.0](http://www.eclipse.org/legal/epl-2.0/)
+- [一 (Secondary) GNU General Public License, version 2 with the GNU Classpath Exception](https://projects.eclipse.org/license/secondary-gpl-2.0-cp)
+
+## Trademark
+
+"Theia" is a trademark of the Eclipse Foundation
+<https://www.eclipse.org/theia>

packages/ai-google/package.json

@@ -0,0 +1,52 @@
+{
+  "name": "@theia/ai-google",
+  "version": "1.68.0",
+  "description": "Theia - Google AI Integration",
+  "dependencies": {
+    "@google/genai": "^1.30.0",
+    "@theia/ai-core": "1.68.0",
+    "@theia/core": "1.68.0"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "theiaExtensions": [
+    {
+      "frontend": "lib/browser/google-frontend-module",
+      "backend": "lib/node/google-backend-module"
+    }
+  ],
+  "keywords": [
+    "theia-extension",
+    "ai",
+    "google",
+    "gemini"
+  ],
+  "license": "EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/eclipse-theia/theia.git"
+  },
+  "bugs": {
+    "url": "https://github.com/eclipse-theia/theia/issues"
+  },
+  "homepage": "https://github.com/eclipse-theia/theia",
+  "files": [
+    "lib",
+    "src"
+  ],
+  "scripts": {
+    "build": "theiaext build",
+    "clean": "theiaext clean",
+    "compile": "theiaext compile",
+    "lint": "theiaext lint",
+    "test": "theiaext test",
+    "watch": "theiaext watch"
+  },
+  "devDependencies": {
+    "@theia/ext-scripts": "1.68.0"
+  },
+  "nyc": {
+    "extends": "../../configs/nyc.json"
+  }
+}

packages/ai-google/src/browser/google-frontend-application-contribution.ts

@@ -0,0 +1,100 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+import { FrontendApplicationContribution } from '@theia/core/lib/browser';
+import { inject, injectable } from '@theia/core/shared/inversify';
+import { GoogleLanguageModelsManager, GoogleModelDescription } from '../common';
+import { API_KEY_PREF, MODELS_PREF, MAX_RETRIES, RETRY_DELAY_OTHER_ERRORS, RETRY_DELAY_RATE_LIMIT } from '../common/google-preferences';
+import { PreferenceService } from '@theia/core';
+
+const GOOGLE_PROVIDER_ID = 'google';
+
+@injectable()
+export class GoogleFrontendApplicationContribution implements FrontendApplicationContribution {
+
+    @inject(PreferenceService)
+    protected preferenceService: PreferenceService;
+
+    @inject(GoogleLanguageModelsManager)
+    protected manager: GoogleLanguageModelsManager;
+
+    protected prevModels: string[] = [];
+
+    onStart(): void {
+        this.preferenceService.ready.then(() => {
+            const apiKey = this.preferenceService.get<string>(API_KEY_PREF, undefined);
+            this.manager.setApiKey(apiKey);
+
+            this.manager.setMaxRetriesOnErrors(this.preferenceService.get<number>(MAX_RETRIES, 3));
+            this.manager.setRetryDelayOnRateLimitError(this.preferenceService.get<number>(RETRY_DELAY_RATE_LIMIT, 60));
+            this.manager.setRetryDelayOnOtherErrors(this.preferenceService.get<number>(RETRY_DELAY_OTHER_ERRORS, -1));
+
+            const models = this.preferenceService.get<string[]>(MODELS_PREF, []);
+            this.manager.createOrUpdateLanguageModels(...models.map(modelId => this.createGeminiModelDescription(modelId)));
+            this.prevModels = [...models];
+
+            this.preferenceService.onPreferenceChanged(event => {
+                if (event.preferenceName === API_KEY_PREF) {
+                    const newApiKey = this.preferenceService.get<string>(API_KEY_PREF, undefined);
+                    this.manager.setApiKey(newApiKey);
+                    this.handleKeyChange(newApiKey);
+                } else if (event.preferenceName === MAX_RETRIES) {
+                    this.manager.setMaxRetriesOnErrors(this.preferenceService.get<number>(MAX_RETRIES, 3));
+                } else if (event.preferenceName === RETRY_DELAY_RATE_LIMIT) {
+                    this.manager.setRetryDelayOnRateLimitError(this.preferenceService.get<number>(RETRY_DELAY_RATE_LIMIT, 60));
+                } else if (event.preferenceName === RETRY_DELAY_OTHER_ERRORS) {
+                    this.manager.setRetryDelayOnOtherErrors(this.preferenceService.get<number>(RETRY_DELAY_OTHER_ERRORS, -1));
+                } else if (event.preferenceName === MODELS_PREF) {
+                    this.handleModelChanges(this.preferenceService.get<string[]>(MODELS_PREF, []));
+                }
+            });
+        });
+    }
+
+    /**
+     * Called when the API key changes. Updates all Google models on the manager to ensure the new key is used.
+     */
+    protected handleKeyChange(newApiKey: string | undefined): void {
+        if (this.prevModels && this.prevModels.length > 0) {
+            this.manager.createOrUpdateLanguageModels(...this.prevModels.map(modelId => this.createGeminiModelDescription(modelId)));
+        }
+    }
+
+    protected handleModelChanges(newModels: string[]): void {
+        const oldModels = new Set(this.prevModels);
+        const updatedModels = new Set(newModels);
+
+        const modelsToRemove = [...oldModels].filter(model => !updatedModels.has(model));
+        const modelsToAdd = [...updatedModels].filter(model => !oldModels.has(model));
+
+        this.manager.removeLanguageModels(...modelsToRemove.map(model => `${GOOGLE_PROVIDER_ID}/${model}`));
+        this.manager.createOrUpdateLanguageModels(...modelsToAdd.map(modelId => this.createGeminiModelDescription(modelId)));
+        this.prevModels = newModels;
+    }
+
+    protected createGeminiModelDescription(modelId: string): GoogleModelDescription {
+        const id = `${GOOGLE_PROVIDER_ID}/${modelId}`;
+
+        const description: GoogleModelDescription = {
+            id: id,
+            model: modelId,
+            apiKey: true,
+            enableStreaming: true
+        };
+
+        return description;
+    }
+}

packages/ai-google/src/browser/google-frontend-module.ts

@@ -0,0 +1,32 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+import { ContainerModule } from '@theia/core/shared/inversify';
+import { GooglePreferencesSchema } from '../common/google-preferences';
+import { FrontendApplicationContribution, RemoteConnectionProvider, ServiceConnectionProvider } from '@theia/core/lib/browser';
+import { GoogleFrontendApplicationContribution } from './google-frontend-application-contribution';
+import { GOOGLE_LANGUAGE_MODELS_MANAGER_PATH, GoogleLanguageModelsManager } from '../common';
+import { PreferenceContribution } from '@theia/core';
+
+export default new ContainerModule(bind => {
+    bind(PreferenceContribution).toConstantValue({ schema: GooglePreferencesSchema });
+    bind(GoogleFrontendApplicationContribution).toSelf().inSingletonScope();
+    bind(FrontendApplicationContribution).toService(GoogleFrontendApplicationContribution);
+    bind(GoogleLanguageModelsManager).toDynamicValue(ctx => {
+        const provider = ctx.container.get<ServiceConnectionProvider>(RemoteConnectionProvider);
+        return provider.createProxy<GoogleLanguageModelsManager>(GOOGLE_LANGUAGE_MODELS_MANAGER_PATH);
+    }).inSingletonScope();
+});

packages/ai-google/src/common/google-language-models-manager.ts

@@ -0,0 +1,50 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+export const GOOGLE_LANGUAGE_MODELS_MANAGER_PATH = '/services/google/language-model-manager';
+export const GoogleLanguageModelsManager = Symbol('GoogleLanguageModelsManager');
+export interface GoogleModelDescription {
+    /**
+     * The identifier of the model which will be shown in the UI.
+     */
+    id: string;
+    /**
+     * The model ID as used by the Google Gemini API.
+     */
+    model: string;
+    /**
+     * The key for the model. If 'true' is provided the global Gemini API key will be used.
+     */
+    apiKey: string | true | undefined;
+    /**
+     * Indicate whether the streaming API shall be used.
+     */
+    enableStreaming: boolean;
+    /**
+     * Maximum number of tokens to generate. Default is 4096.
+     */
+    maxTokens?: number;
+
+}
+
+export interface GoogleLanguageModelsManager {
+    apiKey: string | undefined;
+    setApiKey(key: string | undefined): void;
+    setMaxRetriesOnErrors(maxRetries: number): void;
+    setRetryDelayOnRateLimitError(retryDelay: number): void;
+    setRetryDelayOnOtherErrors(retryDelay: number): void;
+    createOrUpdateLanguageModels(...models: GoogleModelDescription[]): Promise<void>;
+    removeLanguageModels(...modelIds: string[]): void
+}

packages/ai-google/src/common/google-preferences.ts

@@ -0,0 +1,71 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+import { AI_CORE_PREFERENCES_TITLE } from '@theia/ai-core/lib/common/ai-core-preferences';
+import { nls, PreferenceSchema } from '@theia/core';
+
+export const API_KEY_PREF = 'ai-features.google.apiKey';
+export const MODELS_PREF = 'ai-features.google.models';
+export const MAX_RETRIES = 'ai-features.google.maxRetriesOnErrors';
+export const RETRY_DELAY_RATE_LIMIT = 'ai-features.google.retryDelayOnRateLimitError';
+export const RETRY_DELAY_OTHER_ERRORS = 'ai-features.google.retryDelayOnOtherErrors';
+
+export const GooglePreferencesSchema: PreferenceSchema = {
+    properties: {
+        [API_KEY_PREF]: {
+            type: 'string',
+            markdownDescription: nls.localize('theia/ai/google/apiKey/description',
+                'Enter an API Key of your official Google AI (Gemini) Account. **Please note:** By using this preference the GOOGLE AI API key will be stored in clear text\
+            on the machine running Theia. Use the environment variable `GOOGLE_API_KEY` to set the key securely.'),
+            title: AI_CORE_PREFERENCES_TITLE,
+        },
+        [MODELS_PREF]: {
+            type: 'array',
+            description: nls.localize('theia/ai/google/models/description', 'Official Google Gemini models to use'),
+            title: AI_CORE_PREFERENCES_TITLE,
+            default: ['gemini-3-pro-preview', 'gemini-3-flash-preview', 'gemini-2.5-pro', 'gemini-2.5-flash'],
+            items: {
+                type: 'string'
+            }
+        },
+        [MAX_RETRIES]: {
+            type: 'integer',
+            description: nls.localize('theia/ai/google/maxRetriesOnErrors/description',
+                'Maximum number of retries in case of errors. If smaller than 1, then the retry logic is disabled'),
+            title: AI_CORE_PREFERENCES_TITLE,
+            default: 3,
+            minimum: 0
+        },
+        [RETRY_DELAY_RATE_LIMIT]: {
+            type: 'number',
+            description: nls.localize('theia/ai/google/retryDelayOnRateLimitError/description',
+                'Delay in seconds between retries in case of rate limit errors. See https://ai.google.dev/gemini-api/docs/rate-limits'),
+            title: AI_CORE_PREFERENCES_TITLE,
+            default: 60,
+            minimum: 0
+        },
+        [RETRY_DELAY_OTHER_ERRORS]: {
+            type: 'number',
+            description: nls.localize('theia/ai/google/retryDelayOnOtherErrors/description',
+                'Delay in seconds between retries in case of other errors (sometimes the Google GenAI reports errors such as incomplete JSON syntax returned from the model \
+                or 500 Internal Server Error). Setting this to -1 prevents retries in these cases. Otherwise a retry happens either immediately (if set to 0) or after \
+                this delay in seconds (if set to a positive number).'),
+            title: AI_CORE_PREFERENCES_TITLE,
+            default: -1,
+            minimum: -1
+        }
+    }
+};

packages/ai-google/src/common/index.ts

@@ -0,0 +1,16 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+export * from './google-language-models-manager';

packages/ai-google/src/node/google-backend-module.ts

@@ -0,0 +1,36 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+import { ContainerModule } from '@theia/core/shared/inversify';
+import { GOOGLE_LANGUAGE_MODELS_MANAGER_PATH, GoogleLanguageModelsManager } from '../common/google-language-models-manager';
+import { ConnectionHandler, PreferenceContribution, RpcConnectionHandler } from '@theia/core';
+import { GoogleLanguageModelsManagerImpl } from './google-language-models-manager-impl';
+import { ConnectionContainerModule } from '@theia/core/lib/node/messaging/connection-container-module';
+import { GooglePreferencesSchema } from '../common/google-preferences';
+
+// We use a connection module to handle AI services separately for each frontend.
+const geminiConnectionModule = ConnectionContainerModule.create(({ bind, bindBackendService, bindFrontendService }) => {
+    bind(GoogleLanguageModelsManagerImpl).toSelf().inSingletonScope();
+    bind(GoogleLanguageModelsManager).toService(GoogleLanguageModelsManagerImpl);
+    bind(ConnectionHandler).toDynamicValue(ctx =>
+        new RpcConnectionHandler(GOOGLE_LANGUAGE_MODELS_MANAGER_PATH, () => ctx.container.get(GoogleLanguageModelsManager))
+    ).inSingletonScope();
+});
+
+export default new ContainerModule(bind => {
+    bind(PreferenceContribution).toConstantValue({ schema: GooglePreferencesSchema });
+    bind(ConnectionContainerModule).toConstantValue(geminiConnectionModule);
+});

packages/ai-google/src/node/google-language-model.ts

@@ -0,0 +1,524 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+import {
+    createToolCallError,
+    ImageContent,
+    LanguageModel,
+    LanguageModelMessage,
+    LanguageModelRequest,
+    LanguageModelResponse,
+    LanguageModelStatus,
+    LanguageModelStreamResponse,
+    LanguageModelStreamResponsePart,
+    LanguageModelTextResponse,
+    TokenUsageService,
+    ToolCallResult,
+    ToolInvocationContext,
+    UserRequest
+} from '@theia/ai-core';
+import { CancellationToken } from '@theia/core';
+import { GoogleGenAI, FunctionCallingConfigMode, FunctionDeclaration, Content, Schema, Part, Modality, FunctionResponse, ToolConfig } from '@google/genai';
+import { wait } from '@theia/core/lib/common/promise-util';
+import { GoogleLanguageModelRetrySettings } from './google-language-models-manager-impl';
+import { UUID } from '@theia/core/shared/@lumino/coreutils';
+
+interface ToolCallback {
+    readonly name: string;
+    readonly id: string;
+    args: string;
+}
+/**
+ * Converts a tool call result to the Gemini FunctionResponse format.
+ * Gemini requires response to be an object, not an array or primitive.
+ */
+function toFunctionResponse(content: ToolCallResult): FunctionResponse['response'] {
+    if (content === undefined) {
+        return {};
+    }
+    if (Array.isArray(content)) {
+        return { result: content };
+    }
+    if (typeof content === 'object') {
+        return content as FunctionResponse['response'];
+    }
+    return { result: content };
+}
+
+const convertMessageToPart = (message: LanguageModelMessage): Part[] | undefined => {
+    if (LanguageModelMessage.isTextMessage(message) && message.text.length > 0) {
+        return [{ text: message.text }];
+    } else if (LanguageModelMessage.isToolUseMessage(message)) {
+        return [{
+            functionCall: {
+                id: message.id, name: message.name, args: message.input as Record<string, unknown>
+            },
+            thoughtSignature: message.data?.thoughtSignature,
+        }];
+    } else if (LanguageModelMessage.isToolResultMessage(message)) {
+        return [{ functionResponse: { name: message.name, response: toFunctionResponse(message.content) } }];
+    } else if (LanguageModelMessage.isThinkingMessage(message)) {
+        return [{ thought: true, text: message.thinking }];
+    } else if (LanguageModelMessage.isImageMessage(message) && ImageContent.isBase64(message.image)) {
+        return [{ inlineData: { data: message.image.base64data, mimeType: message.image.mimeType } }];
+    }
+};
+/**
+ * Transforms Theia language model messages to Gemini API format
+ * @param messages Array of LanguageModelRequestMessage to transform
+ * @returns Object containing transformed messages and optional system message
+ */
+function transformToGeminiMessages(
+    messages: readonly LanguageModelMessage[]
+): { contents: Content[]; systemMessage?: string } {
+    // Extract the system message (if any), as it is a separate parameter in the Gemini API.
+    const systemMessageObj = messages.find(message => message.actor === 'system');
+    const systemMessage = systemMessageObj && LanguageModelMessage.isTextMessage(systemMessageObj) && systemMessageObj.text || undefined;
+
+    const contents: Content[] = [];
+
+    for (const message of messages) {
+        if (message.actor === 'system') {
+            continue; // Skip system messages as they're handled separately
+        }
+        const resultParts = convertMessageToPart(message);
+        if (resultParts === undefined) {
+            continue;
+        }
+
+        const role = toGoogleRole(message);
+        const lastContent = contents.pop();
+
+        if (!lastContent) {
+            contents.push({ role, parts: resultParts });
+        } else if (lastContent.role !== role) {
+            contents.push(lastContent);
+            contents.push({ role, parts: resultParts });
+        } else {
+            lastContent?.parts?.push(...resultParts);
+            contents.push(lastContent);
+        }
+
+    }
+
+    return {
+        contents: contents,
+        systemMessage,
+    };
+}
+
+export const GoogleModelIdentifier = Symbol('GoogleModelIdentifier');
+
+/**
+ * Converts Theia message actor to Gemini role
+ * @param message The message to convert
+ * @returns Gemini role ('user' or 'model')
+ */
+function toGoogleRole(message: LanguageModelMessage): 'user' | 'model' {
+    switch (message.actor) {
+        case 'ai':
+            return 'model';
+        default:
+            return 'user';
+    }
+}
+
+/**
+ * Implements the Gemini language model integration for Theia
+ */
+export class GoogleModel implements LanguageModel {
+
+    constructor(
+        public readonly id: string,
+        public model: string,
+        public status: LanguageModelStatus,
+        public enableStreaming: boolean,
+        public apiKey: () => string | undefined,
+        public retrySettings: () => GoogleLanguageModelRetrySettings,
+        protected readonly tokenUsageService?: TokenUsageService
+    ) { }
+
+    protected getSettings(request: LanguageModelRequest): Readonly<Record<string, unknown>> {
+        return request.settings ?? {};
+    }
+
+    async request(request: UserRequest, cancellationToken?: CancellationToken): Promise<LanguageModelResponse> {
+        if (!request.messages?.length) {
+            throw new Error('Request must contain at least one message');
+        }
+
+        const genAI = this.initializeGemini();
+
+        try {
+            if (this.enableStreaming) {
+                return this.handleStreamingRequest(genAI, request, cancellationToken);
+            }
+            return this.handleNonStreamingRequest(genAI, request);
+        } catch (error) {
+            const errorMessage = error instanceof Error ? error.message : 'Unknown error occurred';
+            throw new Error(`Gemini API request failed: ${errorMessage}`);
+        }
+    }
+    protected async handleStreamingRequest(
+        genAI: GoogleGenAI,
+        request: UserRequest,
+        cancellationToken?: CancellationToken,
+        toolMessages?: Content[]
+    ): Promise<LanguageModelStreamResponse> {
+        const settings = this.getSettings(request);
+        const { contents: parts, systemMessage } = transformToGeminiMessages(request.messages);
+        const functionDeclarations = this.createFunctionDeclarations(request);
+
+        const toolConfig: ToolConfig = {};
+
+        if (functionDeclarations.length > 0) {
+            toolConfig.functionCallingConfig = {
+                mode: FunctionCallingConfigMode.AUTO,
+            };
+        }
+
+        // Wrap the API call in the retry mechanism
+        const stream = await this.withRetry(async () =>
+            genAI.models.generateContentStream({
+                model: this.model,
+                config: {
+                    systemInstruction: systemMessage,
+                    toolConfig,
+                    responseModalities: [Modality.TEXT],
+                    ...(functionDeclarations.length > 0 && {
+                        tools: [{
+                            functionDeclarations
+                        }]
+                    }),
+                    thinkingConfig: {
+                        // https://ai.google.dev/gemini-api/docs/thinking#summaries
+                        includeThoughts: true,
+                    },
+                    temperature: 1,
+                    ...settings
+                },
+                contents: [...parts, ...(toolMessages ?? [])]
+            }));
+
+        const that = this;
+
+        const asyncIterator = {
+            async *[Symbol.asyncIterator](): AsyncIterator<LanguageModelStreamResponsePart> {
+                const toolCallMap: { [key: string]: ToolCallback } = {};
+                const collectedParts: Part[] = [];
+                try {
+                    for await (const chunk of stream) {
+                        if (cancellationToken?.isCancellationRequested) {
+                            break;
+                        }
+                        const finishReason = chunk.candidates?.[0].finishReason;
+                        if (finishReason) {
+                            switch (finishReason) {
+                                // 'STOP' is the only valid (non-error) finishReason
+                                // "Natural stop point of the model or provided stop sequence."
+                                case 'STOP':
+                                    break;
+                                // MALFORMED_FUNCTION_CALL: The model produced a malformed function call.
+                                // Log warning but continue - there might still be usable text content.
+                                case 'MALFORMED_FUNCTION_CALL':
+                                    console.warn('Gemini returned MALFORMED_FUNCTION_CALL finish reason.', {
+                                        finishReason,
+                                        candidate: chunk.candidates?.[0],
+                                        content: chunk.candidates?.[0]?.content,
+                                        parts: chunk.candidates?.[0]?.content?.parts,
+                                        text: chunk.text,
+                                        usageMetadata: chunk.usageMetadata
+                                    });
+                                    break;
+                                // All other reasons are error-cases. Throw an Error.
+                                // e.g. SAFETY, MAX_TOKENS, RECITATION, LANGUAGE, ...
+                                // https://ai.google.dev/api/generate-content#FinishReason
+                                default:
+                                    console.error('Gemini streaming ended with unexpected finish reason:', {
+                                        finishReason,
+                                        candidate: chunk.candidates?.[0],
+                                        content: chunk.candidates?.[0]?.content,
+                                        parts: chunk.candidates?.[0]?.content?.parts,
+                                        safetyRatings: chunk.candidates?.[0]?.safetyRatings,
+                                        text: chunk.text,
+                                        usageMetadata: chunk.usageMetadata
+                                    });
+                                    throw new Error(`Unexpected finish reason: ${finishReason}`);
+                            }
+                        }
+                        // Handle thinking, text content, and function calls from parts
+                        if (chunk.candidates?.[0]?.content?.parts) {
+                            for (const part of chunk.candidates[0].content.parts) {
+                                collectedParts.push(part);
+                                if (part.text) {
+                                    if (part.thought) {
+                                        yield { thought: part.text, signature: part.thoughtSignature ?? '' };
+                                    } else {
+                                        yield { content: part.text };
+                                    }
+                                } else if (part.functionCall) {
+                                    const functionCall = part.functionCall;
+                                    // Gemini does not always provide a function call ID (unlike Anthropic/OpenAI).
+                                    // We need a stable ID to track calls in toolCallMap and correlate results.
+                                    const callId = functionCall.id ?? UUID.uuid4().replace(/-/g, '');
+                                    let toolCall = toolCallMap[callId];
+                                    if (toolCall === undefined) {
+                                        toolCall = {
+                                            name: functionCall.name ?? '',
+                                            args: functionCall.args ? JSON.stringify(functionCall.args) : '{}',
+                                            id: callId,
+                                        };
+                                        toolCallMap[callId] = toolCall;
+
+                                        yield {
+                                            tool_calls: [{
+                                                finished: false,
+                                                id: toolCall.id,
+                                                function: {
+                                                    name: toolCall.name,
+                                                    arguments: toolCall.args
+                                                },
+                                                data: part.thoughtSignature ? { thoughtSignature: part.thoughtSignature } : undefined
+                                            }]
+                                        };
+                                    } else {
+                                        // Update to existing tool call
+                                        toolCall.args = functionCall.args ? JSON.stringify(functionCall.args) : '{}';
+                                        yield {
+                                            tool_calls: [{
+                                                function: {
+                                                    arguments: toolCall.args
+                                                },
+                                                data: part.thoughtSignature ? { thoughtSignature: part.thoughtSignature } : undefined
+                                            }]
+                                        };
+                                    }
+                                }
+                            }
+                        } else if (chunk.text) {
+                            yield { content: chunk.text };
+                        }
+
+                        // Report token usage if available
+                        if (chunk.usageMetadata && that.tokenUsageService && that.id) {
+                            const promptTokens = chunk.usageMetadata.promptTokenCount;
+                            const completionTokens = chunk.usageMetadata.candidatesTokenCount;
+                            if (promptTokens && completionTokens) {
+                                that.tokenUsageService.recordTokenUsage(that.id, {
+                                    inputTokens: promptTokens,
+                                    outputTokens: completionTokens,
+                                    requestId: request.requestId
+                                }).catch(error => console.error('Error recording token usage:', error));
+                            }
+                        }
+                    }
+
+                    // Process tool calls if any exist
+                    const toolCalls = Object.values(toolCallMap);
+                    if (toolCalls.length > 0) {
+                        // Collect tool results
+                        const toolResult = await Promise.all(toolCalls.map(async tc => {
+                            const tool = request.tools?.find(t => t.name === tc.name);
+                            let result;
+                            if (!tool) {
+                                result = createToolCallError(`Tool '${tc.name}' not found in the available tools for this request.`, 'tool-not-available');
+                            } else {
+                                try {
+                                    result = await tool.handler(tc.args, ToolInvocationContext.create(tc.id));
+                                } catch (e) {
+                                    console.error(`Error executing tool ${tc.name}:`, e);
+                                    result = createToolCallError(e.message || 'Tool execution failed');
+                                }
+                            }
+                            return {
+                                name: tc.name,
+                                result: result,
+                                id: tc.id,
+                                arguments: tc.args,
+                            };
+                        }));
+
+                        // Generate tool call responses
+                        const calls = toolResult.map(tr => ({
+                            finished: true,
+                            id: tr.id,
+                            result: tr.result,
+                            function: { name: tr.name, arguments: tr.arguments },
+                        }));
+                        yield { tool_calls: calls };
+
+                        // Format tool responses for Gemini
+                        // According to Gemini docs, functionResponse needs name and response
+                        const toolResponses: Part[] = toolResult.map(call => ({
+                            functionResponse: {
+                                name: call.name,
+                                response: toFunctionResponse(call.result)
+                            }
+                        }));
+                        const responseMessage: Content = { role: 'user', parts: toolResponses };
+
+                        // Build the model's response content from collected parts
+                        // Exclude thinking parts as they should not be included in the conversation history sent back to the model
+                        const modelResponseParts = collectedParts.filter(p => !p.thought);
+                        const modelContent: Content = { role: 'model', parts: modelResponseParts };
+
+                        const messages = [...(toolMessages ?? []), modelContent, responseMessage];
+
+                        // Continue the conversation with tool results
+                        const continuedResponse = await that.handleStreamingRequest(
+                            genAI,
+                            request,
+                            cancellationToken,
+                            messages
+                        );
+
+                        // Stream the continued response
+                        for await (const nestedEvent of continuedResponse.stream) {
+                            yield nestedEvent;
+                        }
+                    }
+                } catch (e) {
+                    console.error('Error in Gemini streaming:', e);
+                    throw e;
+                }
+            },
+        };
+
+        return { stream: asyncIterator };
+    }
+
+    private createFunctionDeclarations(request: LanguageModelRequest): FunctionDeclaration[] {
+        if (!request.tools || request.tools.length === 0) {
+            return [];
+        }
+
+        return request.tools.map(tool => ({
+            name: tool.name,
+            description: tool.description,
+            parameters: (tool.parameters && Object.keys(tool.parameters.properties).length !== 0) ? tool.parameters as Schema : undefined
+        }));
+    }
+
+    protected async handleNonStreamingRequest(
+        genAI: GoogleGenAI,
+        request: UserRequest
+    ): Promise<LanguageModelTextResponse> {
+        const settings = this.getSettings(request);
+        const { contents: parts, systemMessage } = transformToGeminiMessages(request.messages);
+        const functionDeclarations = this.createFunctionDeclarations(request);
+
+        // Wrap the API call in the retry mechanism
+        const model = await this.withRetry(async () => genAI.models.generateContent({
+            model: this.model,
+            config: {
+                systemInstruction: systemMessage,
+                toolConfig: {
+                    functionCallingConfig: {
+                        mode: FunctionCallingConfigMode.AUTO,
+                    }
+                },
+                ...(functionDeclarations.length > 0 && {
+                    tools: [{ functionDeclarations }]
+                }),
+                ...settings
+            },
+            contents: parts
+        }));
+
+        try {
+            let responseText = '';
+            // For non streaming requests we are always only interested in text parts
+            if (model.candidates?.[0]?.content?.parts) {
+                for (const part of model.candidates[0].content.parts) {
+                    if (part.text) {
+                        responseText += part.text;
+                    }
+                }
+            } else {
+                responseText = model.text ?? '';
+            }
+
+            // Record token usage if available
+            if (model.usageMetadata && this.tokenUsageService) {
+                const promptTokens = model.usageMetadata.promptTokenCount;
+                const completionTokens = model.usageMetadata.candidatesTokenCount;
+                if (promptTokens && completionTokens) {
+                    await this.tokenUsageService.recordTokenUsage(this.id, {
+                        inputTokens: promptTokens,
+                        outputTokens: completionTokens,
+                        requestId: request.requestId
+                    });
+                }
+            }
+            return { text: responseText };
+        } catch (error) {
+            throw new Error(`Failed to get response from Gemini API: ${error instanceof Error ? error.message : 'Unknown error'}`);
+        }
+    }
+
+    protected initializeGemini(): GoogleGenAI {
+        const apiKey = this.apiKey();
+        if (!apiKey) {
+            throw new Error('Please provide GOOGLE_API_KEY in preferences or via environment variable');
+        }
+
+        // TODO test vertexai
+        return new GoogleGenAI({ apiKey, vertexai: false });
+    }
+
+    /**
+     * Implements a retry mechanism for the handle(non)Streaming request functions.
+     * @param fn the wrapped function to which the retry logic should be applied.
+     * @param retrySettings the configuration settings for the retry mechanism.
+     * @returns the result of the wrapped function.
+     */
+    private async withRetry<T>(fn: () => Promise<T>): Promise<T> {
+        const { maxRetriesOnErrors, retryDelayOnRateLimitError, retryDelayOnOtherErrors } = this.retrySettings();
+
+        for (let i = 0; i <= maxRetriesOnErrors; i++) {
+            try {
+                return await fn();
+            } catch (error) {
+                if (i === maxRetriesOnErrors) {
+                    // no retries left - throw the original error
+                    throw error;
+                }
+
+                const message = (error as Error).message;
+                // Check for rate limit exhaustion (usually, there is a rate limit per minute, so we can retry after a delay...)
+                if (message && message.includes('429 Too Many Requests')) {
+                    if (retryDelayOnRateLimitError < 0) {
+                        // rate limit error should not retried because of the setting
+                        throw error;
+                    }
+
+                    const delayMs = retryDelayOnRateLimitError * 1000;
+                    console.warn(`Received 429 (Too Many Requests). Retrying in ${retryDelayOnRateLimitError}s. Attempt ${i + 1} of ${maxRetriesOnErrors}.`);
+                    await wait(delayMs);
+                } else if (retryDelayOnOtherErrors < 0) {
+                    // Other errors should not retried because of the setting
+                    throw error;
+                } else {
+                    const delayMs = retryDelayOnOtherErrors * 1000;
+                    console.warn(`Request failed: ${message}. Retrying in ${retryDelayOnOtherErrors}s. Attempt ${i + 1} of ${maxRetriesOnErrors}.`);
+                    await wait(delayMs);
+                }
+                // -> reiterate the loop for the next attempt
+            }
+        }
+        // This should not be reached
+        throw new Error('Retry mechanism failed unexpectedly.');
+    }
+}

packages/ai-google/src/node/google-language-models-manager-impl.ts

@@ -0,0 +1,121 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+import { LanguageModelRegistry, LanguageModelStatus, TokenUsageService } from '@theia/ai-core';
+import { inject, injectable } from '@theia/core/shared/inversify';
+import { GoogleModel } from './google-language-model';
+import { GoogleLanguageModelsManager, GoogleModelDescription } from '../common';
+
+export interface GoogleLanguageModelRetrySettings {
+    maxRetriesOnErrors: number;
+    retryDelayOnRateLimitError: number;
+    retryDelayOnOtherErrors: number;
+}
+
+@injectable()
+export class GoogleLanguageModelsManagerImpl implements GoogleLanguageModelsManager {
+    protected _apiKey: string | undefined;
+    protected retrySettings: GoogleLanguageModelRetrySettings = {
+        maxRetriesOnErrors: 3,
+        retryDelayOnRateLimitError: 60,
+        retryDelayOnOtherErrors: -1
+    };
+
+    @inject(LanguageModelRegistry)
+    protected readonly languageModelRegistry: LanguageModelRegistry;
+
+    @inject(TokenUsageService)
+    protected readonly tokenUsageService: TokenUsageService;
+
+    get apiKey(): string | undefined {
+        return this._apiKey ?? process.env.GOOGLE_API_KEY ?? process.env.GEMINI_API_KEY;
+    }
+
+    protected calculateStatus(effectiveApiKey: string | undefined): LanguageModelStatus {
+        return effectiveApiKey
+            ? { status: 'ready' }
+            : { status: 'unavailable', message: 'No Google API key set' };
+    }
+
+    async createOrUpdateLanguageModels(...modelDescriptions: GoogleModelDescription[]): Promise<void> {
+        for (const modelDescription of modelDescriptions) {
+            const model = await this.languageModelRegistry.getLanguageModel(modelDescription.id);
+            const apiKeyProvider = () => {
+                if (modelDescription.apiKey === true) {
+                    return this.apiKey;
+                }
+                if (modelDescription.apiKey) {
+                    return modelDescription.apiKey;
+                }
+                return undefined;
+            };
+            const retrySettingsProvider = () => this.retrySettings;
+
+            // Determine the effective API key for status
+            const status = this.calculateStatus(apiKeyProvider());
+
+            if (model) {
+                if (!(model instanceof GoogleModel)) {
+                    console.warn(`Gemini: model ${modelDescription.id} is not a Gemini model`);
+                    continue;
+                }
+                await this.languageModelRegistry.patchLanguageModel<GoogleModel>(modelDescription.id, {
+                    model: modelDescription.model,
+                    enableStreaming: modelDescription.enableStreaming,
+                    apiKey: apiKeyProvider,
+                    retrySettings: retrySettingsProvider,
+                    status
+                });
+            } else {
+                this.languageModelRegistry.addLanguageModels([
+                    new GoogleModel(
+                        modelDescription.id,
+                        modelDescription.model,
+                        status,
+                        modelDescription.enableStreaming,
+                        apiKeyProvider,
+                        retrySettingsProvider,
+                        this.tokenUsageService
+                    )
+                ]);
+            }
+        }
+    }
+
+    removeLanguageModels(...modelIds: string[]): void {
+        this.languageModelRegistry.removeLanguageModels(modelIds);
+    }
+
+    setApiKey(apiKey: string | undefined): void {
+        if (apiKey) {
+            this._apiKey = apiKey;
+        } else {
+            this._apiKey = undefined;
+        }
+    }
+
+    setMaxRetriesOnErrors(maxRetries: number): void {
+        this.retrySettings.maxRetriesOnErrors = maxRetries;
+    }
+
+    setRetryDelayOnRateLimitError(retryDelay: number): void {
+        this.retrySettings.retryDelayOnRateLimitError = retryDelay;
+    }
+
+    setRetryDelayOnOtherErrors(retryDelay: number): void {
+        this.retrySettings.retryDelayOnOtherErrors = retryDelay;
+    }
+}

packages/ai-google/src/package.spec.ts

@@ -0,0 +1,28 @@
+// *****************************************************************************
+// Copyright (C) 2025 EclipseSource GmbH and others.
+//
+// This program and the accompanying materials are made available under the
+// terms of the Eclipse Public License v. 2.0 which is available at
+// http://www.eclipse.org/legal/epl-2.0.
+//
+// This Source Code may also be made available under the following Secondary
+// Licenses when the conditions for such availability set forth in the Eclipse
+// Public License v. 2.0 are satisfied: GNU General Public License, version 2
+// with the GNU Classpath Exception which is available at
+// https://www.gnu.org/software/classpath/license.html.
+//
+// SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-only WITH Classpath-exception-2.0
+// *****************************************************************************
+
+/* note: this bogus test file is required so that
+   we are able to run mocha unit tests on this
+   package, without having any actual unit tests in it.
+   This way a coverage report will be generated,
+   showing 0% coverage, instead of no report.
+   This file can be removed once we have real unit
+   tests in place. */
+
+describe('ai-google package', () => {
+
+    it('support code coverage statistics', () => true);
+});

packages/ai-google/tsconfig.json

@@ -0,0 +1,19 @@
+{
+  "extends": "../../configs/base.tsconfig",
+  "compilerOptions": {
+    "composite": true,
+    "rootDir": "src",
+    "outDir": "lib"
+  },
+  "include": [
+    "src"
+  ],
+  "references": [
+    {
+      "path": "../ai-core"
+    },
+    {
+      "path": "../core"
+    }
+  ]
+}