koishi-plugin-chatluna-google-gemini-adapter 1.1.0-alpha.0 → 1.1.0-alpha.2

package/lib/index.cjs

@@ -23,14 +23,14 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 // src/locales/zh-CN.schema.yml
 var require_zh_CN_schema = __commonJS({
   "src/locales/zh-CN.schema.yml"(exports2, module2) {
-    module2.exports = { $inner: [{}, { $desc: "请求选项", apiKeys: { $inner: ["Gemini 的 API Key", "Gemini API 的请求地址"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxTokens: "输入的最大上下文 Token（16~2097000，必须是 16 的倍数）。注意：仅当您使用的模型最大 Token 为 8000 及以上时，才建议设置超过 2000 token。", temperature: "回复的随机性程度，数值越高，回复越随机（范围：0~2）。", googleSearch: "为模型启用谷歌搜索。", searchThreshold: "搜索的置信度阈值，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。" }] };
+    module2.exports = { $inner: [{}, { $desc: "请求选项", apiKeys: { $inner: ["Gemini 的 API Key", "Gemini API 的请求地址"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxTokens: "输入的最大上下文 Token（16~2097000，必须是 16 的倍数）。注意：仅当您使用的模型最大 Token 为 8000 及以上时，才建议设置超过 2000 token。", temperature: "回复的随机性程度，数值越高，回复越随机（范围：0~2）。", googleSearch: "为模型启用谷歌搜索。", groundingContentDisplay: "是否显示谷歌搜索结果。", searchThreshold: "搜索的置信度阈值，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。" }] };
   }
 });
 
 // src/locales/en-US.schema.yml
 var require_en_US_schema = __commonJS({
   "src/locales/en-US.schema.yml"(exports2, module2) {
-    module2.exports = { $inner: [{}, { $desc: "API Configuration", apiKeys: { $inner: ["Gemini API Key", "Gemini API Endpoint (optional)"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxTokens: "Max output tokens (16-2097000, multiple of 16). >2000 for 8k+ models", temperature: "Sampling temperature (0-2). Higher: more random, Lower: more deterministic", googleSearch: "Enable Google search", searchThreshold: "Search confidence [threshold](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval) (0-1). Lower: more likely to use Google search" }] };
+    module2.exports = { $inner: [{}, { $desc: "API Configuration", apiKeys: { $inner: ["Gemini API Key", "Gemini API Endpoint (optional)"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxTokens: "Max output tokens (16-2097000, multiple of 16). >2000 for 8k+ models", temperature: "Sampling temperature (0-2). Higher: more random, Lower: more deterministic", googleSearch: "Enable Google search", groundingContentDisplay: "Enable display of search results", searchThreshold: "Search confidence [threshold](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval) (0-1). Lower: more likely to use Google search" }] };
   }
 });
 
@@ -248,7 +248,7 @@ function formatToolsToGeminiAITools(tools, config, model) {
       functionDeclarations: functions
     });
   } else if (functions.length > 0 && config.googleSearch) {
-    logger.warn("Google search is enabled, function call will be disabled.");
+    logger.warn("Google search is enabled, tool calling will be disable.");
   }
   if (config.googleSearch) {
     if (model.includes("gemini-2")) {
@@ -270,9 +270,10 @@ function formatToolsToGeminiAITools(tools, config, model) {
 }
 __name(formatToolsToGeminiAITools, "formatToolsToGeminiAITools");
 function formatToolToGeminiAITool(tool) {
-  const parameters = (0, import_zod_to_json_schema.zodToJsonSchema)(tool.schema);
-  delete parameters["$schema"];
-  delete parameters["additionalProperties"];
+  const parameters = removeAdditionalProperties(
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    (0, import_zod_to_json_schema.zodToJsonSchema)(tool.schema)
+  );
   return {
     name: tool.name,
     description: tool.description,
@@ -281,6 +282,30 @@ function formatToolToGeminiAITool(tool) {
   };
 }
 __name(formatToolToGeminiAITool, "formatToolToGeminiAITool");
+function removeAdditionalProperties(schema) {
+  const updatedSchema = { ...schema };
+  if (Object.hasOwn(updatedSchema, "additionalProperties")) {
+    delete updatedSchema["additionalProperties"];
+  }
+  if (Object.hasOwn(updatedSchema, "$schema")) {
+    delete updatedSchema["$schema"];
+  }
+  if (updatedSchema["properties"]) {
+    const keys = Object.keys(updatedSchema["properties"]);
+    removeProperties(updatedSchema["properties"], keys, 0);
+  }
+  return updatedSchema;
+}
+__name(removeAdditionalProperties, "removeAdditionalProperties");
+function removeProperties(properties, keys, index) {
+  if (index >= keys.length) {
+    return;
+  }
+  const key = keys[index];
+  properties[key] = removeAdditionalProperties(properties[key]);
+  removeProperties(properties, keys, index + 1);
+}
+__name(removeProperties, "removeProperties");
 function messageTypeToGeminiRole(type) {
   switch (type) {
     case "system":
@@ -308,6 +333,136 @@ var GeminiRequester = class extends import_api.ModelRequester {
   static {
     __name(this, "GeminiRequester");
   }
+  async completion(params) {
+    try {
+      const response = await this._post(
+        `models/${params.model}`,
+        {
+          contents: await langchainMessageToGeminiMessage(
+            params.input,
+            params.model
+          ),
+          safetySettings: [
+            {
+              category: "HARM_CATEGORY_HARASSMENT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_HATE_SPEECH",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_DANGEROUS_CONTENT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_CIVIC_INTEGRITY",
+              threshold: "BLOCK_NONE"
+            }
+          ],
+          generationConfig: {
+            stopSequences: params.stop,
+            temperature: params.temperature,
+            maxOutputTokens: params.model.includes("vision") ? void 0 : params.maxTokens,
+            topP: params.topP
+            // thinkingConfig: { includeThoughts: true }
+          },
+          tools: params.tools != null || this._pluginConfig.googleSearch ? formatToolsToGeminiAITools(
+            params.tools ?? [],
+            this._pluginConfig,
+            params.model
+          ) : void 0
+        },
+        {
+          signal: params.signal
+        }
+      );
+      const parsedResponse = await response.json();
+      let groundingContent = "";
+      let currentGroudingIndex = 0;
+      let reasoningContent = "";
+      let content = "";
+      const functionCall = {
+        name: "",
+        args: "",
+        arguments: ""
+      };
+      const part = parsedResponse.candidates[0].content;
+      const messagePart = partAsType(part.parts[0]);
+      const chatFunctionCallingPart = partAsType(
+        part.parts[0]
+      );
+      if (messagePart.text) {
+        if (messagePart.thought) {
+          reasoningContent = messagePart.text;
+        }
+        content = messagePart.text;
+      }
+      const deltaFunctionCall = chatFunctionCallingPart.functionCall;
+      if (deltaFunctionCall) {
+        let args = deltaFunctionCall.args?.input ?? deltaFunctionCall.args;
+        try {
+          let parsedArgs = JSON.parse(args);
+          if (typeof parsedArgs !== "string") {
+            args = parsedArgs;
+          }
+          parsedArgs = JSON.parse(args);
+          if (typeof parsedArgs !== "string") {
+            args = parsedArgs;
+          }
+        } catch (e) {
+        }
+        functionCall.args = JSON.stringify(args);
+        functionCall.name = deltaFunctionCall.name;
+        functionCall.arguments = deltaFunctionCall.args;
+      }
+      for (const source of parsedResponse.candidates[0].groundingMetadata?.groundingChunks ?? []) {
+        groundingContent += `[^${currentGroudingIndex++}]: [${source.web.title}](${source.web.uri})
+`;
+      }
+      try {
+        const messageChunk = new import_messages2.AIMessageChunk(content);
+        messageChunk.additional_kwargs = {
+          function_call: functionCall.name.length > 0 ? {
+            name: functionCall.name,
+            arguments: functionCall.args,
+            args: functionCall.arguments
+          } : void 0
+          // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        };
+        messageChunk.content = content;
+        if (groundingContent.length > 0) {
+          logger.debug(`grounding content: ${groundingContent}`);
+          if (this._pluginConfig.groundingContentDisplay) {
+            messageChunk.content += `
+
+${groundingContent}`;
+          }
+        }
+        if (reasoningContent.length > 0) {
+          logger.debug(`reasoning content: ${reasoningContent}`);
+        }
+        const generationChunk = new import_outputs.ChatGenerationChunk({
+          message: messageChunk,
+          text: messageChunk.content
+        });
+        return generationChunk;
+      } catch (e) {
+        logger.error("error with", parsedResponse);
+        throw new import_error.ChatLunaError(import_error.ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
+    } catch (e) {
+      if (e instanceof import_error.ChatLunaError) {
+        throw e;
+      } else {
+        throw new import_error.ChatLunaError(import_error.ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
+    }
+  }
   async *completionStream(params) {
     try {
       const response = await this._post(
@@ -363,6 +518,8 @@ var GeminiRequester = class extends import_api.ModelRequester {
       );
       const jsonParser = new import_json.JSONParser();
       const writable = stream.writable.getWriter();
+      let groundingContent = "";
+      let currentGroudingIndex = 0;
       jsonParser.onEnd = async () => {
         await writable.close();
       };
@@ -379,9 +536,13 @@ var GeminiRequester = class extends import_api.ModelRequester {
           for (const part of parts) {
             await writable.write(part);
           }
+          for (const source of candidate.groundingMetadata?.groundingChunks ?? []) {
+            groundingContent += `[^${currentGroudingIndex++}]: [${source.web.title}](${source.web.uri})
+`;
+          }
         }
       };
-      await (0, import_sse.sse)(
+      (0, import_sse.sse)(
         response,
         async (rawData) => {
           jsonParser.write(rawData);
@@ -389,9 +550,8 @@ var GeminiRequester = class extends import_api.ModelRequester {
         },
         0
       );
-      let content = "";
       let reasoningContent = "";
-      let isOldVisionModel = params.model.includes("vision");
+      let content = "";
       const functionCall = {
         name: "",
         args: "",
@@ -405,15 +565,7 @@ var GeminiRequester = class extends import_api.ModelRequester {
             reasoningContent += messagePart.text;
             continue;
           }
-          if (params.tools != null) {
-            content = messagePart.text;
-          } else {
-            content += messagePart.text;
-          }
-          if (isOldVisionModel && /\s*model:\s*/.test(content)) {
-            isOldVisionModel = false;
-            content = messagePart.text.replace(/\s*model:\s*/, "");
-          }
+          content = messagePart.text;
         }
         const deltaFunctionCall = chatFunctionCallingPart.functionCall;
         if (deltaFunctionCall) {
@@ -466,6 +618,20 @@ var GeminiRequester = class extends import_api.ModelRequester {
       if (reasoningContent.length > 0) {
         logger.debug(`reasoning content: ${reasoningContent}`);
       }
+      if (groundingContent.length > 0) {
+        logger.debug(`grounding content: ${groundingContent}`);
+        if (this._pluginConfig.groundingContentDisplay) {
+          const groundingMessage = new import_messages2.AIMessageChunk(
+            `
+${groundingContent}`
+          );
+          const generationChunk = new import_outputs.ChatGenerationChunk({
+            message: groundingMessage,
+            text: "\n" + groundingContent
+          });
+          yield generationChunk;
+        }
+      }
     } catch (e) {
       if (e instanceof import_error.ChatLunaError) {
         throw e;
@@ -618,13 +784,13 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
             if (model2.includes("gemini-1.5-pro")) {
               return 1048576;
             }
-            if (model2.includes("gemini-1.5-flash")) {
+            if (model2.includes("gemini-1.5-flash") || model2.includes("gemini-2.0-pro")) {
               return 2097152;
             }
             if (model2.includes("gemini-1.0-pro")) {
               return 30720;
             }
-            return 30720;
+            return 1048576;
           })(model),
           type: model.includes("embedding") ? import_types.ModelType.embeddings : import_types.ModelType.llm,
           functionCall: !model.includes("vision"),
@@ -715,6 +881,7 @@ var Config3 = import_koishi.Schema.intersect([
     maxTokens: import_koishi.Schema.number().min(16).max(2097e3).step(16).default(8064),
     temperature: import_koishi.Schema.percent().min(0).max(2).step(0.1).default(0.8),
     googleSearch: import_koishi.Schema.boolean().default(false),
+    groundingContentDisplay: import_koishi.Schema.boolean().default(false),
     searchThreshold: import_koishi.Schema.number().min(0).max(1).step(0.1).default(0.5)
   })
 ]).i18n({

package/lib/index.d.ts

@@ -2,7 +2,7 @@ import { ChatLunaPlugin } from 'koishi-plugin-chatluna/services/chat';
 import { Context, Logger, Schema } from 'koishi';
 import { AIMessageChunk, BaseMessage, ChatMessageChunk, HumanMessageChunk, MessageType, SystemMessageChunk } from '@langchain/core/messages';
 import { StructuredTool } from '@langchain/core/tools';
-import { ChatGenerationChunk } from '@langchain/core/outputs';
+import { ChatGeneration, ChatGenerationChunk } from '@langchain/core/outputs';
 import { EmbeddingsRequester, EmbeddingsRequestParams, ModelRequester, ModelRequestParams } from 'koishi-plugin-chatluna/llm-core/platform/api';
 import { ClientConfig } from 'koishi-plugin-chatluna/llm-core/platform/config';
 import { PlatformModelAndEmbeddingsClient } from 'koishi-plugin-chatluna/llm-core/platform/client';
@@ -16,6 +16,7 @@ export interface Config extends ChatLunaPlugin.Config {
     temperature: number;
     googleSearch: boolean;
     searchThreshold: number;
+    groundingContentDisplay: boolean;
 }
 export const Config: Schema<Config>;
 export const inject: string[];
@@ -50,6 +51,26 @@ export type ChatFunctionResponsePart = {
 export interface ChatResponse {
     candidates: {
         content: ChatCompletionResponseMessage;
+        groundingMetadata: {
+            searchEntryPoint: {
+                renderedContent: string;
+            };
+            groundingChunks: {
+                web: {
+                    uri: string;
+                    title: string;
+                };
+            }[];
+            groundingSupports: {
+                segment: {
+                    endIndex: number;
+                    text: string;
+                };
+                groundingChunkIndices: number[];
+                confidenceScores: number[];
+            }[];
+            webSearchQueries: string[];
+        };
         finishReason: string;
         index: number;
         safetyRatings: {
@@ -92,6 +113,7 @@ export class GeminiRequester extends ModelRequester implements EmbeddingsRequest
     private _plugin;
     private _pluginConfig;
     constructor(_config: ClientConfig, _plugin: ChatLunaPlugin, _pluginConfig: Config);
+    completion(params: ModelRequestParams): Promise<ChatGeneration>;
     completionStream(params: ModelRequestParams): AsyncGenerator<ChatGenerationChunk>;
     embeddings(params: EmbeddingsRequestParams): Promise<number[] | number[][]>;
     getModels(): Promise<string[]>;

package/lib/index.mjs

@@ -8,14 +8,14 @@ var __commonJS = (cb, mod) => function __require() {
 // src/locales/zh-CN.schema.yml
 var require_zh_CN_schema = __commonJS({
   "src/locales/zh-CN.schema.yml"(exports, module) {
-    module.exports = { $inner: [{}, { $desc: "请求选项", apiKeys: { $inner: ["Gemini 的 API Key", "Gemini API 的请求地址"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxTokens: "输入的最大上下文 Token（16~2097000，必须是 16 的倍数）。注意：仅当您使用的模型最大 Token 为 8000 及以上时，才建议设置超过 2000 token。", temperature: "回复的随机性程度，数值越高，回复越随机（范围：0~2）。", googleSearch: "为模型启用谷歌搜索。", searchThreshold: "搜索的置信度阈值，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。" }] };
+    module.exports = { $inner: [{}, { $desc: "请求选项", apiKeys: { $inner: ["Gemini 的 API Key", "Gemini API 的请求地址"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxTokens: "输入的最大上下文 Token（16~2097000，必须是 16 的倍数）。注意：仅当您使用的模型最大 Token 为 8000 及以上时，才建议设置超过 2000 token。", temperature: "回复的随机性程度，数值越高，回复越随机（范围：0~2）。", googleSearch: "为模型启用谷歌搜索。", groundingContentDisplay: "是否显示谷歌搜索结果。", searchThreshold: "搜索的置信度阈值，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。" }] };
   }
 });
 
 // src/locales/en-US.schema.yml
 var require_en_US_schema = __commonJS({
   "src/locales/en-US.schema.yml"(exports, module) {
-    module.exports = { $inner: [{}, { $desc: "API Configuration", apiKeys: { $inner: ["Gemini API Key", "Gemini API Endpoint (optional)"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxTokens: "Max output tokens (16-2097000, multiple of 16). >2000 for 8k+ models", temperature: "Sampling temperature (0-2). Higher: more random, Lower: more deterministic", googleSearch: "Enable Google search", searchThreshold: "Search confidence [threshold](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval) (0-1). Lower: more likely to use Google search" }] };
+    module.exports = { $inner: [{}, { $desc: "API Configuration", apiKeys: { $inner: ["Gemini API Key", "Gemini API Endpoint (optional)"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxTokens: "Max output tokens (16-2097000, multiple of 16). >2000 for 8k+ models", temperature: "Sampling temperature (0-2). Higher: more random, Lower: more deterministic", googleSearch: "Enable Google search", groundingContentDisplay: "Enable display of search results", searchThreshold: "Search confidence [threshold](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval) (0-1). Lower: more likely to use Google search" }] };
   }
 });
 
@@ -242,7 +242,7 @@ function formatToolsToGeminiAITools(tools, config, model) {
       functionDeclarations: functions
     });
   } else if (functions.length > 0 && config.googleSearch) {
-    logger.warn("Google search is enabled, function call will be disabled.");
+    logger.warn("Google search is enabled, tool calling will be disable.");
   }
   if (config.googleSearch) {
     if (model.includes("gemini-2")) {
@@ -264,9 +264,10 @@ function formatToolsToGeminiAITools(tools, config, model) {
 }
 __name(formatToolsToGeminiAITools, "formatToolsToGeminiAITools");
 function formatToolToGeminiAITool(tool) {
-  const parameters = zodToJsonSchema(tool.schema);
-  delete parameters["$schema"];
-  delete parameters["additionalProperties"];
+  const parameters = removeAdditionalProperties(
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    zodToJsonSchema(tool.schema)
+  );
   return {
     name: tool.name,
     description: tool.description,
@@ -275,6 +276,30 @@ function formatToolToGeminiAITool(tool) {
   };
 }
 __name(formatToolToGeminiAITool, "formatToolToGeminiAITool");
+function removeAdditionalProperties(schema) {
+  const updatedSchema = { ...schema };
+  if (Object.hasOwn(updatedSchema, "additionalProperties")) {
+    delete updatedSchema["additionalProperties"];
+  }
+  if (Object.hasOwn(updatedSchema, "$schema")) {
+    delete updatedSchema["$schema"];
+  }
+  if (updatedSchema["properties"]) {
+    const keys = Object.keys(updatedSchema["properties"]);
+    removeProperties(updatedSchema["properties"], keys, 0);
+  }
+  return updatedSchema;
+}
+__name(removeAdditionalProperties, "removeAdditionalProperties");
+function removeProperties(properties, keys, index) {
+  if (index >= keys.length) {
+    return;
+  }
+  const key = keys[index];
+  properties[key] = removeAdditionalProperties(properties[key]);
+  removeProperties(properties, keys, index + 1);
+}
+__name(removeProperties, "removeProperties");
 function messageTypeToGeminiRole(type) {
   switch (type) {
     case "system":
@@ -302,6 +327,136 @@ var GeminiRequester = class extends ModelRequester {
   static {
     __name(this, "GeminiRequester");
   }
+  async completion(params) {
+    try {
+      const response = await this._post(
+        `models/${params.model}`,
+        {
+          contents: await langchainMessageToGeminiMessage(
+            params.input,
+            params.model
+          ),
+          safetySettings: [
+            {
+              category: "HARM_CATEGORY_HARASSMENT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_HATE_SPEECH",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_DANGEROUS_CONTENT",
+              threshold: "BLOCK_NONE"
+            },
+            {
+              category: "HARM_CATEGORY_CIVIC_INTEGRITY",
+              threshold: "BLOCK_NONE"
+            }
+          ],
+          generationConfig: {
+            stopSequences: params.stop,
+            temperature: params.temperature,
+            maxOutputTokens: params.model.includes("vision") ? void 0 : params.maxTokens,
+            topP: params.topP
+            // thinkingConfig: { includeThoughts: true }
+          },
+          tools: params.tools != null || this._pluginConfig.googleSearch ? formatToolsToGeminiAITools(
+            params.tools ?? [],
+            this._pluginConfig,
+            params.model
+          ) : void 0
+        },
+        {
+          signal: params.signal
+        }
+      );
+      const parsedResponse = await response.json();
+      let groundingContent = "";
+      let currentGroudingIndex = 0;
+      let reasoningContent = "";
+      let content = "";
+      const functionCall = {
+        name: "",
+        args: "",
+        arguments: ""
+      };
+      const part = parsedResponse.candidates[0].content;
+      const messagePart = partAsType(part.parts[0]);
+      const chatFunctionCallingPart = partAsType(
+        part.parts[0]
+      );
+      if (messagePart.text) {
+        if (messagePart.thought) {
+          reasoningContent = messagePart.text;
+        }
+        content = messagePart.text;
+      }
+      const deltaFunctionCall = chatFunctionCallingPart.functionCall;
+      if (deltaFunctionCall) {
+        let args = deltaFunctionCall.args?.input ?? deltaFunctionCall.args;
+        try {
+          let parsedArgs = JSON.parse(args);
+          if (typeof parsedArgs !== "string") {
+            args = parsedArgs;
+          }
+          parsedArgs = JSON.parse(args);
+          if (typeof parsedArgs !== "string") {
+            args = parsedArgs;
+          }
+        } catch (e) {
+        }
+        functionCall.args = JSON.stringify(args);
+        functionCall.name = deltaFunctionCall.name;
+        functionCall.arguments = deltaFunctionCall.args;
+      }
+      for (const source of parsedResponse.candidates[0].groundingMetadata?.groundingChunks ?? []) {
+        groundingContent += `[^${currentGroudingIndex++}]: [${source.web.title}](${source.web.uri})
+`;
+      }
+      try {
+        const messageChunk = new AIMessageChunk2(content);
+        messageChunk.additional_kwargs = {
+          function_call: functionCall.name.length > 0 ? {
+            name: functionCall.name,
+            arguments: functionCall.args,
+            args: functionCall.arguments
+          } : void 0
+          // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        };
+        messageChunk.content = content;
+        if (groundingContent.length > 0) {
+          logger.debug(`grounding content: ${groundingContent}`);
+          if (this._pluginConfig.groundingContentDisplay) {
+            messageChunk.content += `
+
+${groundingContent}`;
+          }
+        }
+        if (reasoningContent.length > 0) {
+          logger.debug(`reasoning content: ${reasoningContent}`);
+        }
+        const generationChunk = new ChatGenerationChunk({
+          message: messageChunk,
+          text: messageChunk.content
+        });
+        return generationChunk;
+      } catch (e) {
+        logger.error("error with", parsedResponse);
+        throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
+    } catch (e) {
+      if (e instanceof ChatLunaError) {
+        throw e;
+      } else {
+        throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
+      }
+    }
+  }
   async *completionStream(params) {
     try {
       const response = await this._post(
@@ -357,6 +512,8 @@ var GeminiRequester = class extends ModelRequester {
       );
       const jsonParser = new JSONParser();
       const writable = stream.writable.getWriter();
+      let groundingContent = "";
+      let currentGroudingIndex = 0;
       jsonParser.onEnd = async () => {
         await writable.close();
       };
@@ -373,9 +530,13 @@ var GeminiRequester = class extends ModelRequester {
           for (const part of parts) {
             await writable.write(part);
           }
+          for (const source of candidate.groundingMetadata?.groundingChunks ?? []) {
+            groundingContent += `[^${currentGroudingIndex++}]: [${source.web.title}](${source.web.uri})
+`;
+          }
         }
       };
-      await sse(
+      sse(
         response,
         async (rawData) => {
           jsonParser.write(rawData);
@@ -383,9 +544,8 @@ var GeminiRequester = class extends ModelRequester {
         },
         0
       );
-      let content = "";
       let reasoningContent = "";
-      let isOldVisionModel = params.model.includes("vision");
+      let content = "";
       const functionCall = {
         name: "",
         args: "",
@@ -399,15 +559,7 @@ var GeminiRequester = class extends ModelRequester {
             reasoningContent += messagePart.text;
             continue;
           }
-          if (params.tools != null) {
-            content = messagePart.text;
-          } else {
-            content += messagePart.text;
-          }
-          if (isOldVisionModel && /\s*model:\s*/.test(content)) {
-            isOldVisionModel = false;
-            content = messagePart.text.replace(/\s*model:\s*/, "");
-          }
+          content = messagePart.text;
         }
         const deltaFunctionCall = chatFunctionCallingPart.functionCall;
         if (deltaFunctionCall) {
@@ -460,6 +612,20 @@ var GeminiRequester = class extends ModelRequester {
       if (reasoningContent.length > 0) {
         logger.debug(`reasoning content: ${reasoningContent}`);
       }
+      if (groundingContent.length > 0) {
+        logger.debug(`grounding content: ${groundingContent}`);
+        if (this._pluginConfig.groundingContentDisplay) {
+          const groundingMessage = new AIMessageChunk2(
+            `
+${groundingContent}`
+          );
+          const generationChunk = new ChatGenerationChunk({
+            message: groundingMessage,
+            text: "\n" + groundingContent
+          });
+          yield generationChunk;
+        }
+      }
     } catch (e) {
       if (e instanceof ChatLunaError) {
         throw e;
@@ -612,13 +778,13 @@ var GeminiClient = class extends PlatformModelAndEmbeddingsClient {
             if (model2.includes("gemini-1.5-pro")) {
               return 1048576;
             }
-            if (model2.includes("gemini-1.5-flash")) {
+            if (model2.includes("gemini-1.5-flash") || model2.includes("gemini-2.0-pro")) {
               return 2097152;
             }
             if (model2.includes("gemini-1.0-pro")) {
               return 30720;
             }
-            return 30720;
+            return 1048576;
           })(model),
           type: model.includes("embedding") ? ModelType.embeddings : ModelType.llm,
           functionCall: !model.includes("vision"),
@@ -709,6 +875,7 @@ var Config3 = Schema.intersect([
     maxTokens: Schema.number().min(16).max(2097e3).step(16).default(8064),
     temperature: Schema.percent().min(0).max(2).step(0.1).default(0.8),
     googleSearch: Schema.boolean().default(false),
+    groundingContentDisplay: Schema.boolean().default(false),
     searchThreshold: Schema.number().min(0).max(1).step(0.1).default(0.5)
   })
 ]).i18n({

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "koishi-plugin-chatluna-google-gemini-adapter",
   "description": "google-gemini adapter for chatluna",
-  "version": "1.1.0-alpha.0",
+  "version": "1.1.0-alpha.2",
   "main": "lib/index.cjs",
   "module": "lib/index.mjs",
   "typings": "lib/index.d.ts",
@@ -73,7 +73,7 @@
   },
   "peerDependencies": {
     "koishi": "^4.18.4",
-    "koishi-plugin-chatluna": "^1.1.0-alpha.0"
+    "koishi-plugin-chatluna": "^1.1.0-alpha.2"
   },
   "koishi": {
     "description": {