koishi-plugin-chatluna-google-gemini-adapter 1.2.18 → 1.2.20

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", platform: "适配器的平台名。（不懂请不要修改）", 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。", thinkingBudget: "思考预算，范围：(-1~24576)，设置的数值越大，思考时花费的 Token 越多,-1 为动态思考。目前仅支持 gemini 2.5 系列模型。", groundingContentDisplay: "是否显示谷歌搜索结果。", imageGeneration: "为模型启用图像生成。目前仅支持 `gemini-2.0-flash-exp` 模型。", searchThreshold: "搜索的[置信度阈值](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval)，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。（仅支持 `gemini-1.5` 系列模型。gemini 2.0 模型起使用动态的工具调用）", includeThoughts: "是否获取模型的思考内容。", codeExecution: "为模型启用代码执行工具。", urlContext: "为模型启用 URL 内容获取工具。" }] };
+    module.exports = { $inner: [{}, { $desc: "请求选项", platform: "适配器的平台名。（不懂请不要修改）", 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。", thinkingBudget: "思考预算，范围：(-1~24576)，设置的数值越大，思考时花费的 Token 越多,-1 为动态思考。目前仅支持 gemini 2.5 系列模型。", groundingContentDisplay: "是否显示谷歌搜索结果。", imageGeneration: "为模型启用图像生成。目前仅支持 `gemini-2.0-flash-exp` 和 `gemini-2.5-flash-image-preview` 模型。", searchThreshold: "搜索的[置信度阈值](https://ai.google.dev/gemini-api/docs/grounding?lang=rest#dynamic-retrieval)，范围：0~1，设置的数值越低，则越倾向于使用谷歌搜索。（仅支持 `gemini-1.5` 系列模型。gemini 2.0 模型起使用动态的工具调用）", includeThoughts: "是否获取模型的思考内容。", codeExecution: "为模型启用代码执行工具。", urlContext: "为模型启用 URL 内容获取工具。", nonStreaming: "强制不启用流式返回。开启后，将总是以非流式发起请求，即便配置了 stream 参数。" }] };
   }
 });
 
 // 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", platform: "Adapter platform name. (Do not modify if you do not understand)", 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", thinkingBudget: "Thinking budget (-1-24576). (0: dynamic thinking) Higher: more tokens spent on thinking. Currently only supports `gemini-2.5` series models.", groundingContentDisplay: "Enable display of search results", imageGeneration: "Enable image generation (only for `gemini-2.0-flash-exp` model)", 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", includeThoughts: "Enable retrieval of model thoughts", codeExecution: "Enable code execution tool", urlContext: "Enable URL context retrieval tool" }] };
+    module.exports = { $inner: [{}, { $desc: "API Configuration", platform: "Adapter platform name. (Do not modify if you do not understand)", 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", thinkingBudget: "Thinking budget (-1-24576). (0: dynamic thinking) Higher: more tokens spent on thinking. Currently only supports `gemini-2.5` series models.", groundingContentDisplay: "Enable display of search results", imageGeneration: "Enable image generation (only for `gemini-2.0-flash-exp` and `gemini-2.5-flash-image-preview` model)", 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", includeThoughts: "Enable retrieval of model thoughts", codeExecution: "Enable code execution tool", urlContext: "Enable URL context retrieval tool", nonStreaming: "Force disable streaming response. When enabled, requests will always be made in non-streaming mode, even if the stream parameter is configured." }] };
   }
 });
 
@@ -38,7 +38,9 @@ import {
 } from "koishi-plugin-chatluna/utils/error";
 
 // src/requester.ts
-import { AIMessageChunk as AIMessageChunk2 } from "@langchain/core/messages";
+import {
+  AIMessageChunk as AIMessageChunk2
+} from "@langchain/core/messages";
 import { ChatGenerationChunk } from "@langchain/core/outputs";
 import {
   ModelRequester
@@ -309,8 +311,96 @@ function messageTypeToGeminiRole(type) {
   }
 }
 __name(messageTypeToGeminiRole, "messageTypeToGeminiRole");
+function prepareModelConfig(params, pluginConfig) {
+  let model = params.model;
+  let enabledThinking = null;
+  if (model.includes("-thinking") && model.includes("gemini-2.5")) {
+    enabledThinking = !model.includes("-non-thinking");
+    model = model.replace("-nom-thinking", "").replace("-thinking", "");
+  }
+  let thinkingBudget = pluginConfig.thinkingBudget ?? -1;
+  if (!enabledThinking && !model.includes("2.5-pro")) {
+    thinkingBudget = 0;
+  } else if (thinkingBudget >= 0 && thinkingBudget < 128) {
+    thinkingBudget = 128;
+  }
+  let imageGeneration = pluginConfig.imageGeneration ?? false;
+  if (imageGeneration) {
+    imageGeneration = params.model.includes("gemini-2.0-flash-exp") || params.model.includes("gemini-2.5-flash-image");
+  }
+  return { model, enabledThinking, thinkingBudget, imageGeneration };
+}
+__name(prepareModelConfig, "prepareModelConfig");
+function createSafetySettings(model) {
+  const isGemini2 = model.includes("gemini-2");
+  return [
+    {
+      category: "HARM_CATEGORY_HARASSMENT",
+      threshold: isGemini2 ? "OFF" : "BLOCK_NONE"
+    },
+    {
+      category: "HARM_CATEGORY_HATE_SPEECH",
+      threshold: isGemini2 ? "OFF" : "BLOCK_NONE"
+    },
+    {
+      category: "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+      threshold: isGemini2 ? "OFF" : "BLOCK_NONE"
+    },
+    {
+      category: "HARM_CATEGORY_DANGEROUS_CONTENT",
+      threshold: isGemini2 ? "OFF" : "BLOCK_NONE"
+    },
+    {
+      category: "HARM_CATEGORY_CIVIC_INTEGRITY",
+      threshold: isGemini2 ? "OFF" : "BLOCK_NONE"
+    }
+  ];
+}
+__name(createSafetySettings, "createSafetySettings");
+function createGenerationConfig(params, modelConfig, pluginConfig) {
+  return {
+    stopSequences: params.stop,
+    temperature: params.temperature,
+    maxOutputTokens: params.model.includes("vision") ? void 0 : params.maxTokens,
+    topP: params.topP,
+    responseModalities: modelConfig.imageGeneration ? ["TEXT", "IMAGE"] : void 0,
+    thinkingConfig: modelConfig.enabledThinking != null || pluginConfig.includeThoughts ? {
+      thinkingBudget: modelConfig.thinkingBudget,
+      includeThoughts: pluginConfig.includeThoughts
+    } : void 0
+  };
+}
+__name(createGenerationConfig, "createGenerationConfig");
+async function createChatGenerationParams(params, modelConfig, pluginConfig) {
+  const geminiMessages = await langchainMessageToGeminiMessage(
+    params.input,
+    modelConfig.model
+  );
+  const [systemInstruction, modelMessages] = extractSystemMessages(geminiMessages);
+  return {
+    contents: modelMessages,
+    safetySettings: createSafetySettings(params.model),
+    generationConfig: createGenerationConfig(
+      params,
+      modelConfig,
+      pluginConfig
+    ),
+    system_instruction: systemInstruction != null ? systemInstruction : void 0,
+    tools: params.tools != null || pluginConfig.googleSearch || pluginConfig.codeExecution || pluginConfig.urlContext ? formatToolsToGeminiAITools(
+      params.tools ?? [],
+      pluginConfig,
+      params.model
+    ) : void 0
+  };
+}
+__name(createChatGenerationParams, "createChatGenerationParams");
+function isChatResponse(response) {
+  return "candidates" in response;
+}
+__name(isChatResponse, "isChatResponse");
 
 // src/requester.ts
+import { getMessageContent } from "koishi-plugin-chatluna/utils/string";
 var GeminiRequester = class extends ModelRequester {
   constructor(ctx, _configPool, _pluginConfig, _plugin) {
     super(ctx, _configPool, _pluginConfig, _plugin);
@@ -319,211 +409,38 @@ var GeminiRequester = class extends ModelRequester {
   static {
     __name(this, "GeminiRequester");
   }
+  async completion(params) {
+    if (!this._pluginConfig.nonStreaming) {
+      return super.completion(params);
+    }
+    return await this.completionInternal(params);
+  }
   async *completionStreamInternal(params) {
+    if (this._pluginConfig.nonStreaming) {
+      const generation = await this.completion(params);
+      yield new ChatGenerationChunk({
+        generationInfo: generation.generationInfo,
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        message: generation.message,
+        text: generation.text
+      });
+      return;
+    }
+    const modelConfig = prepareModelConfig(params, this._pluginConfig);
     try {
-      let model = params.model;
-      let enabledThinking = null;
-      if (model.includes("-thinking") && model.includes("gemini-2.5")) {
-        enabledThinking = !model.includes("-non-thinking");
-        model = model.replace("-nom-thinking", "").replace("-thinking", "");
-      }
-      const geminiMessages = await langchainMessageToGeminiMessage(
-        params.input,
-        model
-      );
-      const [systemInstruction, modelMessages] = extractSystemMessages(geminiMessages);
-      let thinkingBudget = this._pluginConfig.thinkingBudget ?? -1;
-      if (!enabledThinking && !model.includes("2.5-pro")) {
-        thinkingBudget = 0;
-      } else if (thinkingBudget >= 0 && thinkingBudget < 128) {
-        thinkingBudget = 128;
-      }
-      let imageGeneration = this._pluginConfig.imageGeneration ?? false;
-      if (imageGeneration) {
-        imageGeneration = params.model.includes("gemini-2.0-flash-exp") || params.model.includes("gemini-2.5-flash-image");
-      }
       const response = await this._post(
-        `models/${model}:streamGenerateContent?alt=sse`,
-        {
-          contents: modelMessages,
-          safetySettings: [
-            {
-              category: "HARM_CATEGORY_HARASSMENT",
-              threshold: params.model.includes("gemini-2") ? "OFF" : "BLOCK_NONE"
-            },
-            {
-              category: "HARM_CATEGORY_HATE_SPEECH",
-              threshold: params.model.includes("gemini-2") ? "OFF" : "BLOCK_NONE"
-            },
-            {
-              category: "HARM_CATEGORY_SEXUALLY_EXPLICIT",
-              threshold: params.model.includes("gemini-2") ? "OFF" : "BLOCK_NONE"
-            },
-            {
-              category: "HARM_CATEGORY_DANGEROUS_CONTENT",
-              threshold: params.model.includes("gemini-2") ? "OFF" : "BLOCK_NONE"
-            },
-            {
-              category: "HARM_CATEGORY_CIVIC_INTEGRITY",
-              threshold: params.model.includes("gemini-2.0") ? "OFF" : "BLOCK_NONE"
-            }
-          ],
-          generationConfig: {
-            stopSequences: params.stop,
-            temperature: params.temperature,
-            maxOutputTokens: params.model.includes("vision") ? void 0 : params.maxTokens,
-            topP: params.topP,
-            responseModalities: imageGeneration ? ["TEXT", "IMAGE"] : void 0,
-            thinkingConfig: enabledThinking != null || this._pluginConfig.includeThoughts ? {
-              thinkingBudget,
-              includeThoughts: this._pluginConfig.includeThoughts
-            } : void 0
-          },
-          system_instruction: systemInstruction != null ? systemInstruction : void 0,
-          tools: params.tools != null || this._pluginConfig.googleSearch || this._pluginConfig.codeExecution || this._pluginConfig.urlContext ? formatToolsToGeminiAITools(
-            params.tools ?? [],
-            this._pluginConfig,
-            params.model
-          ) : void 0
-        },
+        `models/${modelConfig.model}:streamGenerateContent?alt=sse`,
+        await createChatGenerationParams(
+          params,
+          modelConfig,
+          this._pluginConfig
+        ),
         {
           signal: params.signal
         }
       );
-      let errorCount = 0;
-      let groundingContent = "";
-      let currentGroundingIndex = 0;
       await checkResponse(response);
-      const readableStream = new ReadableStream({
-        async start(controller) {
-          for await (const chunk of sseIterable(response)) {
-            controller.enqueue(chunk.data);
-          }
-          controller.close();
-        }
-      });
-      const transformToChatPartStream = new TransformStream({
-        async transform(chunk, controller) {
-          if (chunk === "undefined") {
-            return;
-          }
-          const parsedValue = JSON.parse(chunk);
-          const transformValue = parsedValue;
-          if (!transformValue.candidates) {
-            return;
-          }
-          for (const candidate of transformValue.candidates) {
-            const parts = candidate.content?.parts;
-            if ((parts == null || parts.length < 1) && candidate.finishReason !== "STOP") {
-              throw new Error(chunk);
-            } else if (candidate.finishReason === "STOP" && parts == null) {
-              continue;
-            }
-            for (const part of parts) {
-              controller.enqueue(part);
-            }
-            for (const source of candidate.groundingMetadata?.groundingChunks ?? []) {
-              groundingContent += `[^${currentGroundingIndex++}]: [${source.web.title}](${source.web.uri})
-`;
-            }
-          }
-        }
-      });
-      const iterable = readableStreamToAsyncIterable(
-        readableStream.pipeThrough(transformToChatPartStream)
-      );
-      let reasoningContent = "";
-      let content = "";
-      const functionCall = {
-        name: "",
-        args: "",
-        arguments: ""
-      };
-      for await (const chunk of iterable) {
-        const messagePart = partAsType(chunk);
-        const chatFunctionCallingPart = partAsType(chunk);
-        const imagePart = partAsTypeCheck(
-          chunk,
-          (part) => part["inlineData"] != null
-        );
-        if (messagePart.text) {
-          if (messagePart.thought) {
-            reasoningContent += messagePart.text;
-            continue;
-          }
-          content = messagePart.text;
-        } else if (imagePart) {
-          messagePart.text = `![image](data:${imagePart.inlineData.mimeType ?? "image/png"};base64,${imagePart.inlineData.data})`;
-          content = messagePart.text;
-        }
-        const deltaFunctionCall = chatFunctionCallingPart?.functionCall;
-        if (deltaFunctionCall) {
-          let args = deltaFunctionCall.args;
-          try {
-            let parsedArgs = JSON.parse(args);
-            if (typeof parsedArgs !== "string") {
-              args = parsedArgs;
-            }
-            parsedArgs = JSON.parse(args);
-            if (typeof parsedArgs !== "string") {
-              args = parsedArgs;
-            }
-          } catch (e) {
-          }
-          functionCall.args = JSON.stringify(args);
-          functionCall.name = deltaFunctionCall.name;
-          functionCall.arguments = deltaFunctionCall.args;
-        }
-        try {
-          const messageChunk = new AIMessageChunk2(content);
-          messageChunk.additional_kwargs = {
-            function_call: functionCall.name.length > 0 ? {
-              name: functionCall.name,
-              arguments: functionCall.args,
-              args: functionCall.arguments
-            } : void 0,
-            images: imagePart ? [
-              `data:${imagePart.inlineData.mimeType ?? "image/png"};base64,${imagePart.inlineData.data}`
-            ] : void 0
-            // eslint-disable-next-line @typescript-eslint/no-explicit-any
-          };
-          messageChunk.content = content;
-          const generationChunk = new ChatGenerationChunk({
-            message: messageChunk,
-            text: messageChunk.content
-          });
-          yield generationChunk;
-          content = messageChunk.content;
-        } catch (e) {
-          if (errorCount > 5) {
-            logger.error("error with chunk", chunk);
-            throw new ChatLunaError(
-              ChatLunaErrorCode.API_REQUEST_FAILED,
-              e
-            );
-          } else {
-            errorCount++;
-            continue;
-          }
-        }
-      }
-      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;
-        }
-      }
+      yield* this._processResponseStream(response);
     } catch (e) {
       if (e instanceof ChatLunaError) {
         throw e;
@@ -532,42 +449,41 @@ ${groundingContent}`
       }
     }
   }
-  async embeddings(params) {
-    let data;
-    if (typeof params.input === "string") {
-      params.input = [params.input];
-    }
+  async completionInternal(params) {
+    const modelConfig = prepareModelConfig(params, this._pluginConfig);
     try {
       const response = await this._post(
-        `models/${params.model}:batchEmbedContents`,
+        `models/${modelConfig.model}:generateContent?alt=sse`,
+        await createChatGenerationParams(
+          params,
+          modelConfig,
+          this._pluginConfig
+        ),
         {
-          requests: params.input.map((input) => {
-            return {
-              model: `models/${params.model}`,
-              content: {
-                parts: [
-                  {
-                    text: input
-                  }
-                ]
-              }
-            };
-          })
+          signal: params.signal
         }
       );
-      data = await response.text();
-      data = JSON.parse(data);
-      if (data.embeddings && data.embeddings.length > 0) {
-        return data.embeddings.map((embedding) => {
-          return embedding.values;
-        });
+      await checkResponse(response);
+      return await this._processResponse(response);
+    } catch (e) {
+      if (e instanceof ChatLunaError) {
+        throw e;
+      } else {
+        throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, e);
       }
-      throw new Error(
-        "error when calling gemini embeddings, Result: " + JSON.stringify(data)
+    }
+  }
+  async embeddings(params) {
+    const input = this._prepareEmbeddingsInput(params.input);
+    try {
+      const response = await this._post(
+        `models/${params.model}:batchEmbedContents`,
+        this._createEmbeddingsRequest(params.model, input)
       );
+      return await this._processEmbeddingsResponse(response);
     } catch (e) {
       const error = new Error(
-        "error when calling gemini embeddings, Result: " + JSON.stringify(data)
+        "error when calling gemini embeddings, Error: " + e.message
       );
       error.stack = e.stack;
       error.cause = e.cause;
@@ -575,34 +491,333 @@ ${groundingContent}`
       throw new ChatLunaError(ChatLunaErrorCode.API_REQUEST_FAILED, error);
     }
   }
+  _prepareEmbeddingsInput(input) {
+    return typeof input === "string" ? [input] : input;
+  }
+  _createEmbeddingsRequest(model, input) {
+    return {
+      requests: input.map((text) => ({
+        model: `models/${model}`,
+        content: {
+          parts: [{ text }]
+        }
+      }))
+    };
+  }
+  async _processEmbeddingsResponse(response) {
+    const data = JSON.parse(
+      await response.text()
+    );
+    if (data.embeddings?.length > 0) {
+      return data.embeddings.map((embedding) => embedding.values);
+    }
+    throw new Error(
+      "error when calling gemini embeddings, Result: " + JSON.stringify(data)
+    );
+  }
   async getModels() {
-    let data;
     try {
       const response = await this._get("models");
-      data = await response.text();
-      data = JSON.parse(data);
-      if (!data.models || !data.models.length) {
-        throw new Error(
-          "error when listing gemini models, Result:" + JSON.stringify(data)
-        );
-      }
-      return data.models.filter(
-        (model) => model.name.includes("gemini") || model.name.includes("gemma") || model.name.includes("embedding")
-      ).map((model) => {
-        return {
-          ...model,
-          name: model.name.replace("models/", "")
-        };
-      });
+      const data = await this._parseModelsResponse(response);
+      return this._filterAndTransformModels(data.models);
     } catch (e) {
       const error = new Error(
-        "error when listing gemini models, Result: " + JSON.stringify(data)
+        "error when listing gemini models, Error: " + e.message
       );
       error.stack = e.stack;
       error.cause = e.cause;
       throw error;
     }
   }
+  async _parseModelsResponse(response) {
+    const text = await response.text();
+    const data = JSON.parse(text);
+    if (!data.models?.length) {
+      throw new Error(
+        "error when listing gemini models, Result:" + JSON.stringify(data)
+      );
+    }
+    return data;
+  }
+  _filterAndTransformModels(models) {
+    return models.filter(
+      (model) => ["gemini", "gemma", "embedding"].some(
+        (keyword) => model.name.includes(keyword)
+      )
+    ).map((model) => ({
+      ...model,
+      name: model.name.replace("models/", "")
+    }));
+  }
+  async _processResponse(response) {
+    const { groundingContent, currentGroundingIndex } = this._createStreamContext();
+    const responseText = await response.text();
+    let parsedResponse;
+    try {
+      parsedResponse = JSON.parse(responseText);
+      if (!parsedResponse.candidates) {
+        throw new ChatLunaError(
+          ChatLunaErrorCode.API_REQUEST_FAILED,
+          new Error(
+            "error when calling gemini, Result: " + responseText
+          )
+        );
+      }
+    } catch (e) {
+      if (e instanceof ChatLunaError) {
+        throw e;
+      } else {
+        throw new ChatLunaError(
+          ChatLunaErrorCode.API_REQUEST_FAILED,
+          new Error(
+            "error when calling gemini, Result: " + responseText
+          )
+        );
+      }
+    }
+    const iterable = this._setupStreamTransform(
+      parsedResponse,
+      groundingContent,
+      currentGroundingIndex
+    );
+    let result;
+    let reasoningContent = "";
+    for await (const chunk of this._processChunks(iterable)) {
+      if (chunk.type === "reasoning") {
+        reasoningContent = chunk.content;
+      } else {
+        result = result != null ? result.concat(chunk.generation) : chunk.generation;
+      }
+    }
+    const finalChunk = this._handleFinalContent(
+      reasoningContent,
+      groundingContent.value
+    );
+    if (finalChunk != null) {
+      result = result.concat(finalChunk);
+    }
+    return result;
+  }
+  async *_processResponseStream(response) {
+    const { groundingContent, currentGroundingIndex } = this._createStreamContext();
+    const iterable = this._setupStreamTransform(
+      response,
+      groundingContent,
+      currentGroundingIndex
+    );
+    let reasoningContent = "";
+    for await (const chunk of this._processChunks(iterable)) {
+      if (chunk.type === "reasoning") {
+        reasoningContent = chunk.content;
+      } else {
+        yield chunk.generation;
+      }
+    }
+    const finalContent = this._handleFinalContent(reasoningContent, groundingContent.value);
+    if (finalContent != null) {
+      yield finalContent;
+    }
+  }
+  _createStreamContext() {
+    return {
+      groundingContent: { value: "" },
+      currentGroundingIndex: { value: 0 }
+    };
+  }
+  _setupStreamTransform(response, groundingContent, currentGroundingIndex) {
+    const transformToChatPartStream = this._createTransformStream(
+      groundingContent,
+      currentGroundingIndex
+    );
+    const readableStream = new ReadableStream({
+      async start(controller) {
+        if (isChatResponse(response)) {
+          controller.enqueue(response);
+          controller.close();
+          return;
+        }
+        for await (const chunk of sseIterable(response)) {
+          controller.enqueue(chunk.data);
+        }
+        controller.close();
+      }
+    });
+    return readableStreamToAsyncIterable(
+      readableStream.pipeThrough(transformToChatPartStream)
+    );
+  }
+  _createTransformStream(groundingContent, currentGroundingIndex) {
+    const that = this;
+    return new TransformStream({
+      async transform(chunk, controller) {
+        if (chunk === "undefined") {
+          return;
+        }
+        const transformValue = typeof chunk === "string" ? JSON.parse(chunk) : chunk;
+        if (!transformValue?.candidates) {
+          return;
+        }
+        for (const candidate of transformValue.candidates) {
+          that._processCandidateChunk(
+            candidate,
+            controller,
+            JSON.stringify(transformValue),
+            groundingContent,
+            currentGroundingIndex
+          );
+        }
+      }
+    });
+  }
+  _processCandidateChunk(candidate, controller, chunk, groundingContent, currentGroundingIndex) {
+    const parts = candidate.content?.parts;
+    if ((parts == null || parts.length < 1) && candidate.finishReason !== "STOP" && candidate.content === null) {
+      throw new Error(chunk);
+    } else if (candidate.finishReason === "STOP" && parts == null) {
+      return;
+    }
+    if (parts == null) {
+      return;
+    }
+    for (const part of parts) {
+      controller.enqueue(part);
+    }
+    for (const source of candidate.groundingMetadata?.groundingChunks ?? []) {
+      groundingContent.value += `[^${currentGroundingIndex.value++}]: [${source.web.title}](${source.web.uri})
+`;
+    }
+  }
+  async *_processChunks(iterable) {
+    let reasoningContent = "";
+    let errorCount = 0;
+    const functionCall = {
+      name: "",
+      args: "",
+      arguments: ""
+    };
+    for await (const chunk of iterable) {
+      try {
+        const { updatedContent, updatedReasoning } = this._processChunk(
+          chunk,
+          reasoningContent,
+          functionCall
+        );
+        if (updatedReasoning !== reasoningContent) {
+          reasoningContent = updatedReasoning;
+          yield { type: "reasoning", content: reasoningContent };
+          continue;
+        }
+        if (updatedContent || functionCall.name) {
+          const messageChunk = this._createMessageChunk(
+            updatedContent,
+            functionCall,
+            partAsTypeCheck(
+              chunk,
+              (part) => part["inlineData"] != null
+            )
+          );
+          const generationChunk = new ChatGenerationChunk({
+            message: messageChunk,
+            text: getMessageContent(messageChunk.content) ?? ""
+          });
+          yield { type: "generation", generation: generationChunk };
+        }
+      } catch (e) {
+        if (errorCount > 5) {
+          logger.error("error with chunk", chunk);
+          throw new ChatLunaError(
+            ChatLunaErrorCode.API_REQUEST_FAILED,
+            e
+          );
+        } else {
+          errorCount++;
+          continue;
+        }
+      }
+    }
+  }
+  _processChunk(chunk, reasoningContent, functionCall) {
+    const messagePart = partAsType(chunk);
+    const chatFunctionCallingPart = partAsType(chunk);
+    const imagePart = partAsTypeCheck(
+      chunk,
+      (part) => part["inlineData"] != null
+    );
+    let messageContent;
+    if (messagePart.text) {
+      if (messagePart.thought) {
+        return {
+          updatedContent: messageContent,
+          updatedReasoning: reasoningContent + messagePart.text
+        };
+      }
+      messageContent = messagePart.text;
+    } else if (imagePart) {
+      messagePart.text = `![image](data:${imagePart.inlineData.mimeType ?? "image/png"};base64,${imagePart.inlineData.data})`;
+      messageContent = messagePart.text;
+    }
+    const deltaFunctionCall = chatFunctionCallingPart?.functionCall;
+    if (deltaFunctionCall) {
+      this._updateFunctionCall(functionCall, deltaFunctionCall);
+    }
+    return {
+      updatedContent: messageContent,
+      updatedReasoning: reasoningContent
+    };
+  }
+  _updateFunctionCall(functionCall, deltaFunctionCall) {
+    let args = 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;
+  }
+  _handleFinalContent(reasoningContent, groundingContent) {
+    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
+        });
+        return generationChunk;
+      }
+    }
+  }
+  _createMessageChunk(content, functionCall, imagePart) {
+    const messageChunk = new AIMessageChunk2({
+      content
+    });
+    messageChunk.additional_kwargs = {
+      function_call: functionCall.name.length > 0 ? {
+        name: functionCall.name,
+        arguments: functionCall.args,
+        args: functionCall.arguments
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      } : void 0,
+      images: imagePart ? [
+        `data:${imagePart.inlineData.mimeType ?? "image/png"};base64,${imagePart.inlineData.data}`
+      ] : void 0
+    };
+    return messageChunk;
+  }
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
   _post(url, data, params = {}) {
     const requestUrl = this._concatUrl(url);
@@ -782,6 +997,7 @@ var Config4 = Schema.intersect([
     urlContext: Schema.boolean().default(false),
     thinkingBudget: Schema.number().min(-1).max(24576).default(-1),
     includeThoughts: Schema.boolean().default(false),
+    nonStreaming: Schema.boolean().default(false),
     imageGeneration: Schema.boolean().default(false),
     groundingContentDisplay: Schema.boolean().default(false),
     searchThreshold: Schema.number().min(0).max(1).step(0.1).default(0.5)