koishi-plugin-chatluna-google-gemini-adapter 1.3.0-alpha.8 → 1.3.0

package/README.md

@@ -1,7 +1,7 @@
-## chatluna-google-gemini-adapter
-
-## [![npm](https://img.shields.io/npm/v/koishi-plugin-chatluna-google-gemini-adapter)](https://www.npmjs.com/package/koishi-plugin-chatluna-google-gemini) [![npm](https://img.shields.io/npm/dm/koishi-plugin-chatluna-google-gemini-adapter)](https://www.npmjs.com/package//koishi-plugin-chatluna-google-gemini-adapter)
-
-> 为 ChatLuna 提供 Google Gemini 支持的适配器
-
-[Gemini 适配器文档](https://chatluna.chat/guide/configure-model-platform/google-gemini.html)
+## chatluna-google-gemini-adapter
+
+## [![npm](https://img.shields.io/npm/v/koishi-plugin-chatluna-google-gemini-adapter)](https://www.npmjs.com/package/koishi-plugin-chatluna-google-gemini) [![npm](https://img.shields.io/npm/dm/koishi-plugin-chatluna-google-gemini-adapter)](https://www.npmjs.com/package//koishi-plugin-chatluna-google-gemini-adapter)
+
+> 为 ChatLuna 提供 Google Gemini 支持的适配器
+
+[Gemini 适配器文档](https://chatluna.chat/guide/configure-model-platform/google-gemini.html)

package/lib/client.d.ts

@@ -5,6 +5,7 @@ import { ChatLunaBaseEmbeddings, ChatLunaChatModel } from 'koishi-plugin-chatlun
 import { ModelInfo } from 'koishi-plugin-chatluna/llm-core/platform/types';
 import { Config } from '.';
 import { ChatLunaPlugin } from 'koishi-plugin-chatluna/services/chat';
+import { RunnableConfig } from '@langchain/core/runnables';
 export declare class GeminiClient extends PlatformModelAndEmbeddingsClient<ClientConfig> {
     private _config;
     plugin: ChatLunaPlugin;
@@ -12,6 +13,6 @@ export declare class GeminiClient extends PlatformModelAndEmbeddingsClient<Clien
     private _requester;
     get logger(): import("reggol");
     constructor(ctx: Context, _config: Config, plugin: ChatLunaPlugin);
-    refreshModels(): Promise<ModelInfo[]>;
+    refreshModels(config?: RunnableConfig): Promise<ModelInfo[]>;
     protected _createModel(model: string): ChatLunaChatModel | ChatLunaBaseEmbeddings;
 }

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", 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。" }] };
+    module2.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 API Key", "Gemini API 请求地址", "是否启用此配置"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxContextRatio: "最大上下文使用比例（0~1），控制可用的模型上下文窗口大小的最大百分比。例如 0.35 表示最多使用模型上下文的 35%。", 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"(exports2, module2) {
-    module2.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." }] };
+    module2.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)", "Enabled"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxContextRatio: "Maximum context usage ratio (0-1). Controls the maximum percentage of model context window available for use. For example, 0.35 means at most 35% of the model context can be used.", 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." }] };
   }
 });
 
@@ -51,7 +51,7 @@ var import_koishi = require("koishi");
 // src/client.ts
 var import_client = require("koishi-plugin-chatluna/llm-core/platform/client");
 var import_model = require("koishi-plugin-chatluna/llm-core/platform/model");
-var import_types = require("koishi-plugin-chatluna/llm-core/platform/types");
+var import_types2 = require("koishi-plugin-chatluna/llm-core/platform/types");
 var import_error2 = require("koishi-plugin-chatluna/utils/error");
 
 // src/requester.ts
@@ -63,10 +63,10 @@ var import_sse = require("koishi-plugin-chatluna/utils/sse");
 var import_stream = require("koishi-plugin-chatluna/utils/stream");
 
 // src/utils.ts
-var import_zod_to_json_schema = require("zod-to-json-schema");
 var import_v1_shared_adapter = require("@chatluna/v1-shared-adapter");
 var import_string = require("koishi-plugin-chatluna/utils/string");
-var import_zod = require("zod");
+var import_types = require("@langchain/core/utils/types");
+var import_zod_openapi = require("@anatine/zod-openapi");
 async function langchainMessageToGeminiMessage(messages, plugin, model) {
   return Promise.all(
     messages.map(async (message) => {
@@ -287,9 +287,7 @@ function formatToolsToGeminiAITools(tools, config, model) {
 __name(formatToolsToGeminiAITools, "formatToolsToGeminiAITools");
 function formatToolToGeminiAITool(tool) {
   const parameters = (0, import_v1_shared_adapter.removeAdditionalProperties)(
-    tool.schema instanceof import_zod.ZodSchema ? (0, import_zod_to_json_schema.zodToJsonSchema)(tool.schema, {
-      allowedAdditionalProperties: void 0
-    }) : tool.schema
+    (0, import_types.isZodSchemaV3)(tool.schema) ? (0, import_zod_openapi.generateSchema)(tool.schema, true, "3.0") : tool.schema
   );
   return {
     name: tool.name,
@@ -405,6 +403,7 @@ __name(isChatResponse, "isChatResponse");
 
 // src/requester.ts
 var import_string2 = require("koishi-plugin-chatluna/utils/string");
+var import_logger = require("koishi-plugin-chatluna/utils/logger");
 var GeminiRequester = class extends import_api.ModelRequester {
   constructor(ctx, _configPool, _pluginConfig, _plugin) {
     super(ctx, _configPool, _pluginConfig, _plugin);
@@ -448,6 +447,13 @@ var GeminiRequester = class extends import_api.ModelRequester {
       await (0, import_sse.checkResponse)(response);
       yield* this._processResponseStream(response);
     } catch (e) {
+      if (this.ctx.chatluna.config.isLog) {
+        await (0, import_logger.trackLogToLocal)(
+          "Request",
+          JSON.stringify(chatGenerationParams),
+          logger
+        );
+      }
       if (e instanceof import_error.ChatLunaError) {
         throw e;
       } else {
@@ -457,15 +463,16 @@ var GeminiRequester = class extends import_api.ModelRequester {
   }
   async completionInternal(params) {
     const modelConfig = prepareModelConfig(params, this._pluginConfig);
+    const chatGenerationParams = await createChatGenerationParams(
+      params,
+      this._plugin,
+      modelConfig,
+      this._pluginConfig
+    );
     try {
       const response = await this._post(
         `models/${modelConfig.model}:generateContent`,
-        await createChatGenerationParams(
-          params,
-          this._plugin,
-          modelConfig,
-          this._pluginConfig
-        ),
+        chatGenerationParams,
         {
           signal: params.signal
         }
@@ -473,6 +480,13 @@ var GeminiRequester = class extends import_api.ModelRequester {
       await (0, import_sse.checkResponse)(response);
       return await this._processResponse(response);
     } catch (e) {
+      if (this.ctx.chatluna.config.isLog) {
+        await (0, import_logger.trackLogToLocal)(
+          "Request",
+          JSON.stringify(chatGenerationParams),
+          logger
+        );
+      }
       if (e instanceof import_error.ChatLunaError) {
         throw e;
       } else {
@@ -522,12 +536,17 @@ var GeminiRequester = class extends import_api.ModelRequester {
       "error when calling gemini embeddings, Result: " + JSON.stringify(data)
     );
   }
-  async getModels() {
+  async getModels(config) {
     try {
-      const response = await this._get("models");
+      const response = await this._get("models", {
+        signal: config?.signal
+      });
       const data = await this._parseModelsResponse(response);
       return this._filterAndTransformModels(data.models);
     } catch (e) {
+      if (e instanceof import_error.ChatLunaError) {
+        throw e;
+      }
       const error = new Error(
         "error when listing gemini models, Error: " + e.message
       );
@@ -664,6 +683,18 @@ var GeminiRequester = class extends import_api.ModelRequester {
           return;
         }
         const transformValue = typeof chunk === "string" ? JSON.parse(chunk) : chunk;
+        if (transformValue.usageMetadata) {
+          const promptTokens = transformValue.usageMetadata.promptTokenCount;
+          const totalTokens = transformValue.usageMetadata.totalTokenCount;
+          const completionTokens = transformValue.usageMetadata.candidatesTokenCount ?? totalTokens - promptTokens;
+          controller.enqueue({
+            usage: {
+              promptTokens,
+              completionTokens,
+              totalTokens
+            }
+          });
+        }
         if (!transformValue?.candidates) {
           return;
         }
@@ -702,6 +733,22 @@ var GeminiRequester = class extends import_api.ModelRequester {
     let errorCount = 0;
     let functionIndex = 0;
     for await (const chunk of iterable) {
+      let parsedChunk;
+      if (parsedChunk = partAsTypeCheck(
+        chunk,
+        (chunk2) => chunk2["usage"] != null
+      )) {
+        const generationChunk = new import_outputs.ChatGenerationChunk({
+          message: new import_messages.AIMessageChunk({
+            content: "",
+            response_metadata: {
+              tokenUsage: parsedChunk.usage
+            }
+          }),
+          text: ""
+        });
+        yield { type: "generation", generation: generationChunk };
+      }
       try {
         const { updatedContent, updatedReasoning, updatedToolCalling } = await this._processChunk(
           chunk,
@@ -849,11 +896,12 @@ ${groundingContent}`
       ...params
     });
   }
-  _get(url) {
+  _get(url, params = {}) {
     const requestUrl = this._concatUrl(url);
     return this._plugin.fetch(requestUrl, {
       method: "GET",
-      headers: this._buildHeaders()
+      headers: this._buildHeaders(),
+      ...params
     });
   }
   _concatUrl(url) {
@@ -901,9 +949,9 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
   get logger() {
     return logger;
   }
-  async refreshModels() {
+  async refreshModels(config) {
     try {
-      const rawModels = await this._requester.getModels();
+      const rawModels = await this._requester.getModels(config);
       if (!rawModels.length) {
         throw new import_error2.ChatLunaError(
           import_error2.ChatLunaErrorCode.MODEL_INIT_ERROR,
@@ -915,10 +963,10 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
         const info = {
           name: model.name,
           maxTokens: model.inputTokenLimit,
-          type: model.name.includes("embedding") ? import_types.ModelType.embeddings : import_types.ModelType.llm,
+          type: model.name.includes("embedding") ? import_types2.ModelType.embeddings : import_types2.ModelType.llm,
           capabilities: [
-            import_types.ModelCapabilities.ImageInput,
-            import_types.ModelCapabilities.ToolCall
+            import_types2.ModelCapabilities.ImageInput,
+            import_types2.ModelCapabilities.ToolCall
           ]
         };
         if (model.name.includes("gemini-2.5") && !model.name.includes("pro") && !model.name.includes("image")) {
@@ -937,6 +985,9 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
       }
       return models;
     } catch (e) {
+      if (e instanceof import_error2.ChatLunaError) {
+        throw e;
+      }
       throw new import_error2.ChatLunaError(import_error2.ChatLunaErrorCode.MODEL_INIT_ERROR, e);
     }
   }
@@ -945,13 +996,15 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
     if (info == null) {
       throw new import_error2.ChatLunaError(import_error2.ChatLunaErrorCode.MODEL_NOT_FOUND);
     }
-    if (info.type === import_types.ModelType.llm) {
+    if (info.type === import_types2.ModelType.llm) {
       return new import_model.ChatLunaChatModel({
         modelInfo: info,
         requester: this._requester,
         model,
         modelMaxContextSize: info.maxTokens,
-        maxTokenLimit: this._config.maxTokens,
+        maxTokenLimit: Math.floor(
+          (info.maxTokens || 1e5) * this._config.maxContextRatio
+        ),
         timeout: this._config.timeout,
         temperature: this._config.temperature,
         maxRetries: this._config.maxRetries,
@@ -967,16 +1020,17 @@ var GeminiClient = class extends import_client.PlatformModelAndEmbeddingsClient
 };
 
 // src/index.ts
-var import_logger = require("koishi-plugin-chatluna/utils/logger");
+var import_logger2 = require("koishi-plugin-chatluna/utils/logger");
 var logger;
 var reusable = true;
 function apply(ctx, config) {
-  const plugin = new import_chat.ChatLunaPlugin(ctx, config, config.platform);
-  logger = (0, import_logger.createLogger)(ctx, "chatluna-gemini-adapter");
+  logger = (0, import_logger2.createLogger)(ctx, "chatluna-gemini-adapter");
   ctx.on("ready", async () => {
-    plugin.registerToService();
-    await plugin.parseConfig((config2) => {
-      return config2.apiKeys.map(([apiKey, apiEndpoint]) => {
+    const plugin = new import_chat.ChatLunaPlugin(ctx, config, config.platform);
+    plugin.parseConfig((config2) => {
+      return config2.apiKeys.filter(([apiKey, _, enabled]) => {
+        return apiKey.length > 0 && enabled;
+      }).map(([apiKey, apiEndpoint]) => {
         return {
           apiKey,
           apiEndpoint,
@@ -988,8 +1042,8 @@ function apply(ctx, config) {
         };
       });
     });
-    plugin.registerClient((ctx2) => new GeminiClient(ctx2, config, plugin));
-    await plugin.initClients();
+    plugin.registerClient(() => new GeminiClient(ctx, config, plugin));
+    await plugin.initClient();
   });
 }
 __name(apply, "apply");
@@ -999,15 +1053,16 @@ var Config4 = import_koishi.Schema.intersect([
     platform: import_koishi.Schema.string().default("gemini"),
     apiKeys: import_koishi.Schema.array(
       import_koishi.Schema.tuple([
-        import_koishi.Schema.string().role("secret"),
+        import_koishi.Schema.string().role("secret").default(""),
         import_koishi.Schema.string().default(
           "https://generativelanguage.googleapis.com/v1beta"
-        )
+        ),
+        import_koishi.Schema.boolean().default(true)
       ])
-    ).default([["", "https://generativelanguage.googleapis.com/v1beta"]])
+    ).default([[]]).role("table")
   }),
   import_koishi.Schema.object({
-    maxTokens: import_koishi.Schema.number().min(16).max(2097e3).step(16).default(8064),
+    maxContextRatio: import_koishi.Schema.number().min(0).max(1).step(1e-4).role("slider").default(0.35),
     temperature: import_koishi.Schema.percent().min(0).max(2).step(0.1).default(1),
     googleSearch: import_koishi.Schema.boolean().default(false),
     codeExecution: import_koishi.Schema.boolean().default(false),

package/lib/index.d.ts

@@ -4,8 +4,8 @@ export declare let logger: Logger;
 export declare const reusable = true;
 export declare function apply(ctx: Context, config: Config): void;
 export interface Config extends ChatLunaPlugin.Config {
-    apiKeys: [string, string][];
-    maxTokens: number;
+    apiKeys: [string, string, boolean][];
+    maxContextRatio: number;
     platform: string;
     temperature: number;
     googleSearch: boolean;

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` 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。" }] };
+    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 API Key", "Gemini API 请求地址", "是否启用此配置"], $desc: "Gemini 的 API Key 和请求地址列表。" } }, { $desc: "模型配置", maxContextRatio: "最大上下文使用比例（0~1），控制可用的模型上下文窗口大小的最大百分比。例如 0.35 表示最多使用模型上下文的 35%。", 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` 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." }] };
+    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)", "Enabled"], $desc: "Gemini API access credentials" } }, { $desc: "Model Parameters", maxContextRatio: "Maximum context usage ratio (0-1). Controls the maximum percentage of model context window available for use. For example, 0.35 means at most 35% of the model context can be used.", 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." }] };
   }
 });
 
@@ -54,7 +54,6 @@ import { checkResponse, sseIterable } from "koishi-plugin-chatluna/utils/sse";
 import { readableStreamToAsyncIterable } from "koishi-plugin-chatluna/utils/stream";
 
 // src/utils.ts
-import { zodToJsonSchema } from "zod-to-json-schema";
 import {
   fetchImageUrl,
   removeAdditionalProperties
@@ -63,7 +62,8 @@ import {
   isMessageContentImageUrl,
   isMessageContentText
 } from "koishi-plugin-chatluna/utils/string";
-import { ZodSchema } from "zod";
+import { isZodSchemaV3 } from "@langchain/core/utils/types";
+import { generateSchema } from "@anatine/zod-openapi";
 async function langchainMessageToGeminiMessage(messages, plugin, model) {
   return Promise.all(
     messages.map(async (message) => {
@@ -284,9 +284,7 @@ function formatToolsToGeminiAITools(tools, config, model) {
 __name(formatToolsToGeminiAITools, "formatToolsToGeminiAITools");
 function formatToolToGeminiAITool(tool) {
   const parameters = removeAdditionalProperties(
-    tool.schema instanceof ZodSchema ? zodToJsonSchema(tool.schema, {
-      allowedAdditionalProperties: void 0
-    }) : tool.schema
+    isZodSchemaV3(tool.schema) ? generateSchema(tool.schema, true, "3.0") : tool.schema
   );
   return {
     name: tool.name,
@@ -402,6 +400,7 @@ __name(isChatResponse, "isChatResponse");
 
 // src/requester.ts
 import { getMessageContent } from "koishi-plugin-chatluna/utils/string";
+import { trackLogToLocal } from "koishi-plugin-chatluna/utils/logger";
 var GeminiRequester = class extends ModelRequester {
   constructor(ctx, _configPool, _pluginConfig, _plugin) {
     super(ctx, _configPool, _pluginConfig, _plugin);
@@ -445,6 +444,13 @@ var GeminiRequester = class extends ModelRequester {
       await checkResponse(response);
       yield* this._processResponseStream(response);
     } catch (e) {
+      if (this.ctx.chatluna.config.isLog) {
+        await trackLogToLocal(
+          "Request",
+          JSON.stringify(chatGenerationParams),
+          logger
+        );
+      }
       if (e instanceof ChatLunaError) {
         throw e;
       } else {
@@ -454,15 +460,16 @@ var GeminiRequester = class extends ModelRequester {
   }
   async completionInternal(params) {
     const modelConfig = prepareModelConfig(params, this._pluginConfig);
+    const chatGenerationParams = await createChatGenerationParams(
+      params,
+      this._plugin,
+      modelConfig,
+      this._pluginConfig
+    );
     try {
       const response = await this._post(
         `models/${modelConfig.model}:generateContent`,
-        await createChatGenerationParams(
-          params,
-          this._plugin,
-          modelConfig,
-          this._pluginConfig
-        ),
+        chatGenerationParams,
         {
           signal: params.signal
         }
@@ -470,6 +477,13 @@ var GeminiRequester = class extends ModelRequester {
       await checkResponse(response);
       return await this._processResponse(response);
     } catch (e) {
+      if (this.ctx.chatluna.config.isLog) {
+        await trackLogToLocal(
+          "Request",
+          JSON.stringify(chatGenerationParams),
+          logger
+        );
+      }
       if (e instanceof ChatLunaError) {
         throw e;
       } else {
@@ -519,12 +533,17 @@ var GeminiRequester = class extends ModelRequester {
       "error when calling gemini embeddings, Result: " + JSON.stringify(data)
     );
   }
-  async getModels() {
+  async getModels(config) {
     try {
-      const response = await this._get("models");
+      const response = await this._get("models", {
+        signal: config?.signal
+      });
       const data = await this._parseModelsResponse(response);
       return this._filterAndTransformModels(data.models);
     } catch (e) {
+      if (e instanceof ChatLunaError) {
+        throw e;
+      }
       const error = new Error(
         "error when listing gemini models, Error: " + e.message
       );
@@ -661,6 +680,18 @@ var GeminiRequester = class extends ModelRequester {
           return;
         }
         const transformValue = typeof chunk === "string" ? JSON.parse(chunk) : chunk;
+        if (transformValue.usageMetadata) {
+          const promptTokens = transformValue.usageMetadata.promptTokenCount;
+          const totalTokens = transformValue.usageMetadata.totalTokenCount;
+          const completionTokens = transformValue.usageMetadata.candidatesTokenCount ?? totalTokens - promptTokens;
+          controller.enqueue({
+            usage: {
+              promptTokens,
+              completionTokens,
+              totalTokens
+            }
+          });
+        }
         if (!transformValue?.candidates) {
           return;
         }
@@ -699,6 +730,22 @@ var GeminiRequester = class extends ModelRequester {
     let errorCount = 0;
     let functionIndex = 0;
     for await (const chunk of iterable) {
+      let parsedChunk;
+      if (parsedChunk = partAsTypeCheck(
+        chunk,
+        (chunk2) => chunk2["usage"] != null
+      )) {
+        const generationChunk = new ChatGenerationChunk({
+          message: new AIMessageChunk({
+            content: "",
+            response_metadata: {
+              tokenUsage: parsedChunk.usage
+            }
+          }),
+          text: ""
+        });
+        yield { type: "generation", generation: generationChunk };
+      }
       try {
         const { updatedContent, updatedReasoning, updatedToolCalling } = await this._processChunk(
           chunk,
@@ -846,11 +893,12 @@ ${groundingContent}`
       ...params
     });
   }
-  _get(url) {
+  _get(url, params = {}) {
     const requestUrl = this._concatUrl(url);
     return this._plugin.fetch(requestUrl, {
       method: "GET",
-      headers: this._buildHeaders()
+      headers: this._buildHeaders(),
+      ...params
     });
   }
   _concatUrl(url) {
@@ -898,9 +946,9 @@ var GeminiClient = class extends PlatformModelAndEmbeddingsClient {
   get logger() {
     return logger;
   }
-  async refreshModels() {
+  async refreshModels(config) {
     try {
-      const rawModels = await this._requester.getModels();
+      const rawModels = await this._requester.getModels(config);
       if (!rawModels.length) {
         throw new ChatLunaError2(
           ChatLunaErrorCode2.MODEL_INIT_ERROR,
@@ -934,6 +982,9 @@ var GeminiClient = class extends PlatformModelAndEmbeddingsClient {
       }
       return models;
     } catch (e) {
+      if (e instanceof ChatLunaError2) {
+        throw e;
+      }
       throw new ChatLunaError2(ChatLunaErrorCode2.MODEL_INIT_ERROR, e);
     }
   }
@@ -948,7 +999,9 @@ var GeminiClient = class extends PlatformModelAndEmbeddingsClient {
         requester: this._requester,
         model,
         modelMaxContextSize: info.maxTokens,
-        maxTokenLimit: this._config.maxTokens,
+        maxTokenLimit: Math.floor(
+          (info.maxTokens || 1e5) * this._config.maxContextRatio
+        ),
         timeout: this._config.timeout,
         temperature: this._config.temperature,
         maxRetries: this._config.maxRetries,
@@ -968,12 +1021,13 @@ import { createLogger } from "koishi-plugin-chatluna/utils/logger";
 var logger;
 var reusable = true;
 function apply(ctx, config) {
-  const plugin = new ChatLunaPlugin(ctx, config, config.platform);
   logger = createLogger(ctx, "chatluna-gemini-adapter");
   ctx.on("ready", async () => {
-    plugin.registerToService();
-    await plugin.parseConfig((config2) => {
-      return config2.apiKeys.map(([apiKey, apiEndpoint]) => {
+    const plugin = new ChatLunaPlugin(ctx, config, config.platform);
+    plugin.parseConfig((config2) => {
+      return config2.apiKeys.filter(([apiKey, _, enabled]) => {
+        return apiKey.length > 0 && enabled;
+      }).map(([apiKey, apiEndpoint]) => {
         return {
           apiKey,
           apiEndpoint,
@@ -985,8 +1039,8 @@ function apply(ctx, config) {
         };
       });
     });
-    plugin.registerClient((ctx2) => new GeminiClient(ctx2, config, plugin));
-    await plugin.initClients();
+    plugin.registerClient(() => new GeminiClient(ctx, config, plugin));
+    await plugin.initClient();
   });
 }
 __name(apply, "apply");
@@ -996,15 +1050,16 @@ var Config4 = Schema.intersect([
     platform: Schema.string().default("gemini"),
     apiKeys: Schema.array(
       Schema.tuple([
-        Schema.string().role("secret"),
+        Schema.string().role("secret").default(""),
         Schema.string().default(
           "https://generativelanguage.googleapis.com/v1beta"
-        )
+        ),
+        Schema.boolean().default(true)
       ])
-    ).default([["", "https://generativelanguage.googleapis.com/v1beta"]])
+    ).default([[]]).role("table")
   }),
   Schema.object({
-    maxTokens: Schema.number().min(16).max(2097e3).step(16).default(8064),
+    maxContextRatio: Schema.number().min(0).max(1).step(1e-4).role("slider").default(0.35),
     temperature: Schema.percent().min(0).max(2).step(0.1).default(1),
     googleSearch: Schema.boolean().default(false),
     codeExecution: Schema.boolean().default(false),

package/lib/requester.d.ts

@@ -5,6 +5,7 @@ import { Config } from '.';
 import { GeminiModelInfo } from './types';
 import { ChatLunaPlugin } from 'koishi-plugin-chatluna/services/chat';
 import { Context } from 'koishi';
+import { RunnableConfig } from '@langchain/core/runnables';
 export declare class GeminiRequester extends ModelRequester implements EmbeddingsRequester {
     _pluginConfig: Config;
     constructor(ctx: Context, _configPool: ClientConfigPool<ClientConfig>, _pluginConfig: Config, _plugin: ChatLunaPlugin);
@@ -15,7 +16,7 @@ export declare class GeminiRequester extends ModelRequester implements Embedding
     private _prepareEmbeddingsInput;
     private _createEmbeddingsRequest;
     private _processEmbeddingsResponse;
-    getModels(): Promise<GeminiModelInfo[]>;
+    getModels(config?: RunnableConfig): Promise<GeminiModelInfo[]>;
     private _parseModelsResponse;
     private _filterAndTransformModels;
     private _processResponse;

package/lib/types.d.ts

@@ -2,11 +2,18 @@ export interface ChatCompletionResponseMessage {
     role: string;
     parts?: ChatPart[];
 }
-export type ChatPart = ChatMessagePart | ChatInlineDataPart | ChatFunctionCallingPart | ChatFunctionResponsePart | ChatUploadDataPart;
+export type ChatPart = ChatMessagePart | ChatInlineDataPart | ChatFunctionCallingPart | ChatFunctionResponsePart | ChatUploadDataPart | ChatUsageMetadataPart;
 export type ChatMessagePart = {
     text: string;
     thought?: boolean;
 };
+export type ChatUsageMetadataPart = {
+    usage: {
+        promptTokens: number;
+        completionTokens: number;
+        totalTokens: number;
+    };
+};
 export type ChatInlineDataPart = {
     inlineData: {
         mimeType: string;

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "koishi-plugin-chatluna-google-gemini-adapter",
   "description": "google-gemini adapter for chatluna",
-  "version": "1.3.0-alpha.8",
+  "version": "1.3.0",
   "main": "lib/index.cjs",
   "module": "lib/index.mjs",
   "typings": "lib/index.d.ts",
@@ -22,13 +22,13 @@
   "repository": {
     "type": "git",
     "url": "https://github.com/ChatLunaLab/chatluna.git",
-    "directory": "packages/google-gemini-adapter"
+    "directory": "packages/adapter-gemini"
   },
   "license": "AGPL-3.0",
   "bugs": {
     "url": "https://github.com/ChatLunaLab/chatluna/issues"
   },
-  "homepage": "https://github.com/ChatLunaLab/chatluna/tree/v1-dev/packages/google-gemini-adapter#readme",
+  "homepage": "https://github.com/ChatLunaLab/chatluna/tree/v1-dev/packages/adapter-gemini#readme",
   "scripts": {
     "build": "atsc -b"
   },
@@ -62,19 +62,21 @@
     "adapter"
   ],
   "dependencies": {
-    "@chatluna/v1-shared-adapter": "^1.0.8",
+    "@anatine/zod-openapi": "^2.2.8",
+    "@chatluna/v1-shared-adapter": "^1.0.16",
     "@langchain/core": "0.3.62",
+    "openapi3-ts": "^4.5.0",
     "zod": "3.25.76",
-    "zod-to-json-schema": "^3.24.5"
+    "zod-to-json-schema": "^3.24.6"
   },
   "devDependencies": {
     "atsc": "^2.1.0",
-    "koishi": "^4.18.7"
+    "koishi": "^4.18.9"
   },
   "peerDependencies": {
-    "koishi": "^4.18.7",
-    "koishi-plugin-chatluna": "^1.3.0-alpha.40",
-    "koishi-plugin-chatluna-storage-service": "^0.0.9"
+    "koishi": "^4.18.9",
+    "koishi-plugin-chatluna": "^1.3.0",
+    "koishi-plugin-chatluna-storage-service": "^0.0.11"
   },
   "peerDependenciesMeta": {
     "koishi-plugin-chatluna-storage-service": {