Merge branch 'ChatGPTNextWeb:main' into cloudflare-dev

app/client/api.ts

@@ -1,22 +1,33 @@
 import { getClientConfig } from "../config/client";
 import {
   ACCESS_CODE_PREFIX,
-  Azure,
   ModelProvider,
   ServiceProvider,
 } from "../constant";
-import { ChatMessage, ModelType, useAccessStore, useChatStore } from "../store";
-import { ChatGPTApi } from "./platforms/openai";
+import {
+  ChatMessageTool,
+  ChatMessage,
+  ModelType,
+  useAccessStore,
+  useChatStore,
+} from "../store";
+import { ChatGPTApi, DalleRequestPayload } from "./platforms/openai";
 import { GeminiProApi } from "./platforms/google";
 import { ClaudeApi } from "./platforms/anthropic";
 import { ErnieApi } from "./platforms/baidu";
 import { DoubaoApi } from "./platforms/bytedance";
 import { QwenApi } from "./platforms/alibaba";
+import { HunyuanApi } from "./platforms/tencent";
+import { MoonshotApi } from "./platforms/moonshot";
+import { SparkApi } from "./platforms/iflytek";
+import { XAIApi } from "./platforms/xai";
+import { ChatGLMApi } from "./platforms/glm";
 
 export const ROLES = ["system", "user", "assistant"] as const;
 export type MessageRole = (typeof ROLES)[number];
 
 export const Models = ["gpt-3.5-turbo", "gpt-4"] as const;
+export const TTSModels = ["tts-1", "tts-1-hd"] as const;
 export type ChatModel = ModelType;
 
 export interface MultimodalContent {
@@ -40,6 +51,18 @@ export interface LLMConfig {
   stream?: boolean;
   presence_penalty?: number;
   frequency_penalty?: number;
+  size?: DalleRequestPayload["size"];
+  quality?: DalleRequestPayload["quality"];
+  style?: DalleRequestPayload["style"];
+}
+
+export interface SpeechOptions {
+  model: string;
+  input: string;
+  voice: string;
+  response_format?: string;
+  speed?: number;
+  onController?: (controller: AbortController) => void;
 }
 
 export interface ChatOptions {
@@ -47,9 +70,11 @@ export interface ChatOptions {
   config: LLMConfig;
 
   onUpdate?: (message: string, chunk: string) => void;
-  onFinish: (message: string) => void;
+  onFinish: (message: string, responseRes: Response) => void;
   onError?: (err: Error) => void;
   onController?: (controller: AbortController) => void;
+  onBeforeTool?: (tool: ChatMessageTool) => void;
+  onAfterTool?: (tool: ChatMessageTool) => void;
 }
 
 export interface LLMUsage {
@@ -62,16 +87,19 @@ export interface LLMModel {
   displayName?: string;
   available: boolean;
   provider: LLMModelProvider;
+  sorted: number;
 }
 
 export interface LLMModelProvider {
   id: string;
   providerName: string;
   providerType: string;
+  sorted: number;
 }
 
 export abstract class LLMApi {
   abstract chat(options: ChatOptions): Promise<void>;
+  abstract speech(options: SpeechOptions): Promise<ArrayBuffer>;
   abstract usage(): Promise<LLMUsage>;
   abstract models(): Promise<LLMModel[]>;
 }
@@ -117,6 +145,21 @@ export class ClientApi {
       case ModelProvider.Qwen:
         this.llm = new QwenApi();
         break;
+      case ModelProvider.Hunyuan:
+        this.llm = new HunyuanApi();
+        break;
+      case ModelProvider.Moonshot:
+        this.llm = new MoonshotApi();
+        break;
+      case ModelProvider.Iflytek:
+        this.llm = new SparkApi();
+        break;
+      case ModelProvider.XAI:
+        this.llm = new XAIApi();
+        break;
+      case ModelProvider.ChatGLM:
+        this.llm = new ChatGLMApi();
+        break;
       default:
         this.llm = new ChatGPTApi();
     }
@@ -168,24 +211,44 @@ export class ClientApi {
   }
 }
 
-export function getHeaders() {
+export function getBearerToken(
+  apiKey: string,
+  noBearer: boolean = false,
+): string {
+  return validString(apiKey)
+    ? `${noBearer ? "" : "Bearer "}${apiKey.trim()}`
+    : "";
+}
+
+export function validString(x: string): boolean {
+  return x?.length > 0;
+}
+
+export function getHeaders(ignoreHeaders: boolean = false) {
   const accessStore = useAccessStore.getState();
   const chatStore = useChatStore.getState();
-  const headers: Record<string, string> = {
-    "Content-Type": "application/json",
-    Accept: "application/json",
-  };
+  let headers: Record<string, string> = {};
+  if (!ignoreHeaders) {
+    headers = {
+      "Content-Type": "application/json",
+      Accept: "application/json",
+    };
+  }
 
   const clientConfig = getClientConfig();
 
   function getConfig() {
     const modelConfig = chatStore.currentSession().mask.modelConfig;
-    const isGoogle = modelConfig.providerName == ServiceProvider.Google;
+    const isGoogle = modelConfig.providerName === ServiceProvider.Google;
     const isAzure = modelConfig.providerName === ServiceProvider.Azure;
     const isAnthropic = modelConfig.providerName === ServiceProvider.Anthropic;
     const isBaidu = modelConfig.providerName == ServiceProvider.Baidu;
     const isByteDance = modelConfig.providerName === ServiceProvider.ByteDance;
     const isAlibaba = modelConfig.providerName === ServiceProvider.Alibaba;
+    const isMoonshot = modelConfig.providerName === ServiceProvider.Moonshot;
+    const isIflytek = modelConfig.providerName === ServiceProvider.Iflytek;
+    const isXAI = modelConfig.providerName === ServiceProvider.XAI;
+    const isChatGLM = modelConfig.providerName === ServiceProvider.ChatGLM;
     const isEnabledAccessControl = accessStore.enabledAccessControl();
     const apiKey = isGoogle
       ? accessStore.googleApiKey
@@ -197,6 +260,16 @@ export function getHeaders() {
       ? accessStore.bytedanceApiKey
       : isAlibaba
       ? accessStore.alibabaApiKey
+      : isMoonshot
+      ? accessStore.moonshotApiKey
+      : isXAI
+      ? accessStore.xaiApiKey
+      : isChatGLM
+      ? accessStore.chatglmApiKey
+      : isIflytek
+      ? accessStore.iflytekApiKey && accessStore.iflytekApiSecret
+        ? accessStore.iflytekApiKey + ":" + accessStore.iflytekApiSecret
+        : ""
       : accessStore.openaiApiKey;
     return {
       isGoogle,
@@ -205,24 +278,25 @@ export function getHeaders() {
       isBaidu,
       isByteDance,
       isAlibaba,
+      isMoonshot,
+      isIflytek,
+      isXAI,
+      isChatGLM,
       apiKey,
       isEnabledAccessControl,
     };
   }
 
   function getAuthHeader(): string {
-    return isAzure ? "api-key" : isAnthropic ? "x-api-key" : "Authorization";
+    return isAzure
+      ? "api-key"
+      : isAnthropic
+      ? "x-api-key"
+      : isGoogle
+      ? "x-goog-api-key"
+      : "Authorization";
   }
 
-  function getBearerToken(apiKey: string, noBearer: boolean = false): string {
-    return validString(apiKey)
-      ? `${noBearer ? "" : "Bearer "}${apiKey.trim()}`
-      : "";
-  }
-
-  function validString(x: string): boolean {
-    return x?.length > 0;
-  }
   const {
     isGoogle,
     isAzure,
@@ -231,14 +305,15 @@ export function getHeaders() {
     apiKey,
     isEnabledAccessControl,
   } = getConfig();
-  // when using google api in app, not set auth header
-  if (isGoogle && clientConfig?.isApp) return headers;
   // when using baidu api in app, not set auth header
   if (isBaidu && clientConfig?.isApp) return headers;
 
   const authHeader = getAuthHeader();
 
-  const bearerToken = getBearerToken(apiKey, isAzure || isAnthropic);
+  const bearerToken = getBearerToken(
+    apiKey,
+    isAzure || isAnthropic || isGoogle,
+  );
 
   if (bearerToken) {
     headers[authHeader] = bearerToken;
@@ -262,6 +337,16 @@ export function getClientApi(provider: ServiceProvider): ClientApi {
       return new ClientApi(ModelProvider.Doubao);
     case ServiceProvider.Alibaba:
       return new ClientApi(ModelProvider.Qwen);
+    case ServiceProvider.Tencent:
+      return new ClientApi(ModelProvider.Hunyuan);
+    case ServiceProvider.Moonshot:
+      return new ClientApi(ModelProvider.Moonshot);
+    case ServiceProvider.Iflytek:
+      return new ClientApi(ModelProvider.Iflytek);
+    case ServiceProvider.XAI:
+      return new ClientApi(ModelProvider.XAI);
+    case ServiceProvider.ChatGLM:
+      return new ClientApi(ModelProvider.ChatGLM);
     default:
       return new ClientApi(ModelProvider.GPT);
   }

app/client/platforms/alibaba.ts

@@ -12,6 +12,7 @@ import {
   getHeaders,
   LLMApi,
   LLMModel,
+  SpeechOptions,
   MultimodalContent,
 } from "../api";
 import Locale from "../../locales";
@@ -21,7 +22,8 @@ import {
 } from "@fortaine/fetch-event-source";
 import { prettyObject } from "@/app/utils/format";
 import { getClientConfig } from "@/app/config/client";
-import { getMessageTextContent, isVisionModel } from "@/app/utils";
+import { getMessageTextContent } from "@/app/utils";
+import { fetch } from "@/app/utils/stream";
 
 export interface OpenAIListModelResponse {
   object: string;
@@ -83,6 +85,10 @@ export class QwenApi implements LLMApi {
     return res?.output?.choices?.at(0)?.message?.content ?? "";
   }
 
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
   async chat(options: ChatOptions) {
     const messages = options.messages.map((v) => ({
       role: v.role,
@@ -137,6 +143,7 @@ export class QwenApi implements LLMApi {
         let responseText = "";
         let remainText = "";
         let finished = false;
+        let responseRes: Response;
 
         // animate response to make it looks smooth
         function animateResponseText() {
@@ -166,13 +173,14 @@ export class QwenApi implements LLMApi {
         const finish = () => {
           if (!finished) {
             finished = true;
-            options.onFinish(responseText + remainText);
+            options.onFinish(responseText + remainText, responseRes);
           }
         };
 
         controller.signal.onabort = finish;
 
         fetchEventSource(chatPath, {
+          fetch: fetch as any,
           ...chatPayload,
           async onopen(res) {
             clearTimeout(requestTimeoutId);
@@ -181,6 +189,7 @@ export class QwenApi implements LLMApi {
               "[Alibaba] request response content type: ",
               contentType,
             );
+            responseRes = res;
 
             if (contentType?.startsWith("text/plain")) {
               responseText = await res.clone().text();
@@ -247,7 +256,7 @@ export class QwenApi implements LLMApi {
 
         const resJson = await res.json();
         const message = this.extractMessage(resJson);
-        options.onFinish(message);
+        options.onFinish(message, res);
       }
     } catch (e) {
       console.log("[Request] failed to make a chat request", e);

app/client/platforms/anthropic.ts

@@ -1,18 +1,19 @@
-import { ACCESS_CODE_PREFIX, Anthropic, ApiPath } from "@/app/constant";
-import { ChatOptions, getHeaders, LLMApi, MultimodalContent } from "../api";
-import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
-import { getClientConfig } from "@/app/config/client";
-import { DEFAULT_API_HOST } from "@/app/constant";
-import { RequestMessage } from "@/app/typing";
+import { Anthropic, ApiPath } from "@/app/constant";
+import { ChatOptions, getHeaders, LLMApi, SpeechOptions } from "../api";
 import {
-  EventStreamContentType,
-  fetchEventSource,
-} from "@fortaine/fetch-event-source";
-
-import Locale from "../../locales";
-import { prettyObject } from "@/app/utils/format";
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  usePluginStore,
+  ChatMessageTool,
+} from "@/app/store";
+import { getClientConfig } from "@/app/config/client";
+import { ANTHROPIC_BASE_URL } from "@/app/constant";
 import { getMessageTextContent, isVisionModel } from "@/app/utils";
+import { preProcessImageContent, stream } from "@/app/utils/chat";
 import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
+import { RequestPayload } from "./openai";
+import { fetch } from "@/app/utils/stream";
 
 export type MultiBlockContent = {
   type: "image" | "text";
@@ -73,6 +74,10 @@ const ClaudeMapper = {
 const keys = ["claude-2, claude-instant-1"];
 
 export class ClaudeApi implements LLMApi {
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
   extractMessage(res: any) {
     console.log("[Response] claude response: ", res);
 
@@ -93,7 +98,12 @@ export class ClaudeApi implements LLMApi {
       },
     };
 
-    const messages = [...options.messages];
+    // try get base64image from local cache image_url
+    const messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = await preProcessImageContent(v.content);
+      messages.push({ role: v.role, content });
+    }
 
     const keys = ["system", "user"];
 
@@ -186,120 +196,135 @@ export class ClaudeApi implements LLMApi {
     const controller = new AbortController();
     options.onController?.(controller);
 
-    const payload = {
-      method: "POST",
-      body: JSON.stringify(requestBody),
-      signal: controller.signal,
-      headers: {
-        ...getHeaders(), // get common headers
-        "anthropic-version": accessStore.anthropicApiVersion,
-        // do not send `anthropicApiKey` in browser!!!
-        // Authorization: getAuthKey(accessStore.anthropicApiKey),
-      },
-    };
-
     if (shouldStream) {
-      try {
-        const context = {
-          text: "",
-          finished: false,
-        };
-
-        const finish = () => {
-          if (!context.finished) {
-            options.onFinish(context.text);
-            context.finished = true;
-          }
-        };
-
-        controller.signal.onabort = finish;
-        fetchEventSource(path, {
-          ...payload,
-          async onopen(res) {
-            const contentType = res.headers.get("content-type");
-            console.log("response content type: ", contentType);
-
-            if (contentType?.startsWith("text/plain")) {
-              context.text = await res.clone().text();
-              return finish();
-            }
-
-            if (
-              !res.ok ||
-              !res.headers
-                .get("content-type")
-                ?.startsWith(EventStreamContentType) ||
-              res.status !== 200
-            ) {
-              const responseTexts = [context.text];
-              let extraInfo = await res.clone().text();
-              try {
-                const resJson = await res.clone().json();
-                extraInfo = prettyObject(resJson);
-              } catch {}
-
-              if (res.status === 401) {
-                responseTexts.push(Locale.Error.Unauthorized);
-              }
-
-              if (extraInfo) {
-                responseTexts.push(extraInfo);
-              }
-
-              context.text = responseTexts.join("\n\n");
-
-              return finish();
-            }
-          },
-          onmessage(msg) {
-            let chunkJson:
-              | undefined
-              | {
-                  type: "content_block_delta" | "content_block_stop";
-                  delta?: {
-                    type: "text_delta";
-                    text: string;
-                  };
-                  index: number;
+      let index = -1;
+      const [tools, funcs] = usePluginStore
+        .getState()
+        .getAsTools(
+          useChatStore.getState().currentSession().mask?.plugin || [],
+        );
+      return stream(
+        path,
+        requestBody,
+        {
+          ...getHeaders(),
+          "anthropic-version": accessStore.anthropicApiVersion,
+        },
+        // @ts-ignore
+        tools.map((tool) => ({
+          name: tool?.function?.name,
+          description: tool?.function?.description,
+          input_schema: tool?.function?.parameters,
+        })),
+        funcs,
+        controller,
+        // parseSSE
+        (text: string, runTools: ChatMessageTool[]) => {
+          // console.log("parseSSE", text, runTools);
+          let chunkJson:
+            | undefined
+            | {
+                type: "content_block_delta" | "content_block_stop";
+                content_block?: {
+                  type: "tool_use";
+                  id: string;
+                  name: string;
                 };
-            try {
-              chunkJson = JSON.parse(msg.data);
-            } catch (e) {
-              console.error("[Response] parse error", msg.data);
-            }
+                delta?: {
+                  type: "text_delta" | "input_json_delta";
+                  text?: string;
+                  partial_json?: string;
+                };
+                index: number;
+              };
+          chunkJson = JSON.parse(text);
 
-            if (!chunkJson || chunkJson.type === "content_block_stop") {
-              return finish();
-            }
-
-            const { delta } = chunkJson;
-            if (delta?.text) {
-              context.text += delta.text;
-              options.onUpdate?.(context.text, delta.text);
-            }
-          },
-          onclose() {
-            finish();
-          },
-          onerror(e) {
-            options.onError?.(e);
-            throw e;
-          },
-          openWhenHidden: true,
-        });
-      } catch (e) {
-        console.error("failed to chat", e);
-        options.onError?.(e as Error);
-      }
+          if (chunkJson?.content_block?.type == "tool_use") {
+            index += 1;
+            const id = chunkJson?.content_block.id;
+            const name = chunkJson?.content_block.name;
+            runTools.push({
+              id,
+              type: "function",
+              function: {
+                name,
+                arguments: "",
+              },
+            });
+          }
+          if (
+            chunkJson?.delta?.type == "input_json_delta" &&
+            chunkJson?.delta?.partial_json
+          ) {
+            // @ts-ignore
+            runTools[index]["function"]["arguments"] +=
+              chunkJson?.delta?.partial_json;
+          }
+          return chunkJson?.delta?.text;
+        },
+        // processToolMessage, include tool_calls message and tool call results
+        (
+          requestPayload: RequestPayload,
+          toolCallMessage: any,
+          toolCallResult: any[],
+        ) => {
+          // reset index value
+          index = -1;
+          // @ts-ignore
+          requestPayload?.messages?.splice(
+            // @ts-ignore
+            requestPayload?.messages?.length,
+            0,
+            {
+              role: "assistant",
+              content: toolCallMessage.tool_calls.map(
+                (tool: ChatMessageTool) => ({
+                  type: "tool_use",
+                  id: tool.id,
+                  name: tool?.function?.name,
+                  input: tool?.function?.arguments
+                    ? JSON.parse(tool?.function?.arguments)
+                    : {},
+                }),
+              ),
+            },
+            // @ts-ignore
+            ...toolCallResult.map((result) => ({
+              role: "user",
+              content: [
+                {
+                  type: "tool_result",
+                  tool_use_id: result.tool_call_id,
+                  content: result.content,
+                },
+              ],
+            })),
+          );
+        },
+        options,
+      );
     } else {
+      const payload = {
+        method: "POST",
+        body: JSON.stringify(requestBody),
+        signal: controller.signal,
+        headers: {
+          ...getHeaders(), // get common headers
+          "anthropic-version": accessStore.anthropicApiVersion,
+          // do not send `anthropicApiKey` in browser!!!
+          // Authorization: getAuthKey(accessStore.anthropicApiKey),
+        },
+      };
+
       try {
-        controller.signal.onabort = () => options.onFinish("");
+        controller.signal.onabort = () =>
+          options.onFinish("", new Response(null, { status: 400 }));
 
         const res = await fetch(path, payload);
         const resJson = await res.json();
 
         const message = this.extractMessage(resJson);
-        options.onFinish(message);
+        options.onFinish(message, res);
       } catch (e) {
         console.error("failed to chat", e);
         options.onError?.(e as Error);
@@ -365,9 +390,7 @@ export class ClaudeApi implements LLMApi {
     if (baseUrl.trim().length === 0) {
       const isApp = !!getClientConfig()?.isApp;
 
-      baseUrl = isApp
-        ? DEFAULT_API_HOST + "/api/proxy/anthropic"
-        : ApiPath.Anthropic;
+      baseUrl = isApp ? ANTHROPIC_BASE_URL : ApiPath.Anthropic;
     }
 
     if (!baseUrl.startsWith("http") && !baseUrl.startsWith("/api")) {

app/client/platforms/baidu.ts

@@ -14,6 +14,7 @@ import {
   LLMApi,
   LLMModel,
   MultimodalContent,
+  SpeechOptions,
 } from "../api";
 import Locale from "../../locales";
 import {
@@ -23,6 +24,7 @@ import {
 import { prettyObject } from "@/app/utils/format";
 import { getClientConfig } from "@/app/config/client";
 import { getMessageTextContent } from "@/app/utils";
+import { fetch } from "@/app/utils/stream";
 
 export interface OpenAIListModelResponse {
   object: string;
@@ -75,18 +77,30 @@ export class ErnieApi implements LLMApi {
     return [baseUrl, path].join("/");
   }
 
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
   async chat(options: ChatOptions) {
     const messages = options.messages.map((v) => ({
-      role: v.role,
+      // "error_code": 336006, "error_msg": "the role of message with even index in the messages must be user or function",
+      role: v.role === "system" ? "user" : v.role,
       content: getMessageTextContent(v),
     }));
 
     // "error_code": 336006, "error_msg": "the length of messages must be an odd number",
     if (messages.length % 2 === 0) {
-      messages.unshift({
-        role: "user",
-        content: " ",
-      });
+      if (messages.at(0)?.role === "user") {
+        messages.splice(1, 0, {
+          role: "assistant",
+          content: " ",
+        });
+      } else {
+        messages.unshift({
+          role: "user",
+          content: " ",
+        });
+      }
     }
 
     const modelConfig = {
@@ -148,6 +162,7 @@ export class ErnieApi implements LLMApi {
         let responseText = "";
         let remainText = "";
         let finished = false;
+        let responseRes: Response;
 
         // animate response to make it looks smooth
         function animateResponseText() {
@@ -177,19 +192,20 @@ export class ErnieApi implements LLMApi {
         const finish = () => {
           if (!finished) {
             finished = true;
-            options.onFinish(responseText + remainText);
+            options.onFinish(responseText + remainText, responseRes);
           }
         };
 
         controller.signal.onabort = finish;
 
         fetchEventSource(chatPath, {
+          fetch: fetch as any,
           ...chatPayload,
           async onopen(res) {
             clearTimeout(requestTimeoutId);
             const contentType = res.headers.get("content-type");
             console.log("[Baidu] request response content type: ", contentType);
-
+            responseRes = res;
             if (contentType?.startsWith("text/plain")) {
               responseText = await res.clone().text();
               return finish();
@@ -252,7 +268,7 @@ export class ErnieApi implements LLMApi {
 
         const resJson = await res.json();
         const message = resJson?.result;
-        options.onFinish(message);
+        options.onFinish(message, res);
       }
     } catch (e) {
       console.log("[Request] failed to make a chat request", e);

app/client/platforms/bytedance.ts

@@ -13,6 +13,7 @@ import {
   LLMApi,
   LLMModel,
   MultimodalContent,
+  SpeechOptions,
 } from "../api";
 import Locale from "../../locales";
 import {
@@ -22,6 +23,7 @@ import {
 import { prettyObject } from "@/app/utils/format";
 import { getClientConfig } from "@/app/config/client";
 import { getMessageTextContent } from "@/app/utils";
+import { fetch } from "@/app/utils/stream";
 
 export interface OpenAIListModelResponse {
   object: string;
@@ -77,6 +79,10 @@ export class DoubaoApi implements LLMApi {
     return res.choices?.at(0)?.message?.content ?? "";
   }
 
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
   async chat(options: ChatOptions) {
     const messages = options.messages.map((v) => ({
       role: v.role,
@@ -124,6 +130,7 @@ export class DoubaoApi implements LLMApi {
         let responseText = "";
         let remainText = "";
         let finished = false;
+        let responseRes: Response;
 
         // animate response to make it looks smooth
         function animateResponseText() {
@@ -153,13 +160,14 @@ export class DoubaoApi implements LLMApi {
         const finish = () => {
           if (!finished) {
             finished = true;
-            options.onFinish(responseText + remainText);
+            options.onFinish(responseText + remainText, responseRes);
           }
         };
 
         controller.signal.onabort = finish;
 
         fetchEventSource(chatPath, {
+          fetch: fetch as any,
           ...chatPayload,
           async onopen(res) {
             clearTimeout(requestTimeoutId);
@@ -168,7 +176,7 @@ export class DoubaoApi implements LLMApi {
               "[ByteDance] request response content type: ",
               contentType,
             );
-
+            responseRes = res;
             if (contentType?.startsWith("text/plain")) {
               responseText = await res.clone().text();
               return finish();
@@ -234,7 +242,7 @@ export class DoubaoApi implements LLMApi {
 
         const resJson = await res.json();
         const message = this.extractMessage(resJson);
-        options.onFinish(message);
+        options.onFinish(message, res);
       }
     } catch (e) {
       console.log("[Request] failed to make a chat request", e);

app/client/platforms/glm.ts

@@ -0,0 +1,197 @@
+"use client";
+import {
+  ApiPath,
+  CHATGLM_BASE_URL,
+  ChatGLM,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import {
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  ChatMessageTool,
+  usePluginStore,
+} from "@/app/store";
+import { stream } from "@/app/utils/chat";
+import {
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  SpeechOptions,
+} from "../api";
+import { getClientConfig } from "@/app/config/client";
+import { getMessageTextContent } from "@/app/utils";
+import { RequestPayload } from "./openai";
+import { fetch } from "@/app/utils/stream";
+
+export class ChatGLMApi implements LLMApi {
+  private disableListModels = true;
+
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.chatglmUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      const apiPath = ApiPath.ChatGLM;
+      baseUrl = isApp ? CHATGLM_BASE_URL : apiPath;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.ChatGLM)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res.choices?.at(0)?.message?.content ?? "";
+  }
+
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async chat(options: ChatOptions) {
+    const messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = getMessageTextContent(v);
+      messages.push({ role: v.role, content });
+    }
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+        providerName: options.config.providerName,
+      },
+    };
+
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: options.config.stream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+    };
+
+    console.log("[Request] glm payload: ", requestPayload);
+
+    const shouldStream = !!options.config.stream;
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(ChatGLM.ChatPath);
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      if (shouldStream) {
+        const [tools, funcs] = usePluginStore
+          .getState()
+          .getAsTools(
+            useChatStore.getState().currentSession().mask?.plugin || [],
+          );
+        return stream(
+          chatPath,
+          requestPayload,
+          getHeaders(),
+          tools as any,
+          funcs,
+          controller,
+          // parseSSE
+          (text: string, runTools: ChatMessageTool[]) => {
+            // console.log("parseSSE", text, runTools);
+            const json = JSON.parse(text);
+            const choices = json.choices as Array<{
+              delta: {
+                content: string;
+                tool_calls: ChatMessageTool[];
+              };
+            }>;
+            const tool_calls = choices[0]?.delta?.tool_calls;
+            if (tool_calls?.length > 0) {
+              const index = tool_calls[0]?.index;
+              const id = tool_calls[0]?.id;
+              const args = tool_calls[0]?.function?.arguments;
+              if (id) {
+                runTools.push({
+                  id,
+                  type: tool_calls[0]?.type,
+                  function: {
+                    name: tool_calls[0]?.function?.name as string,
+                    arguments: args,
+                  },
+                });
+              } else {
+                // @ts-ignore
+                runTools[index]["function"]["arguments"] += args;
+              }
+            }
+            return choices[0]?.delta?.content;
+          },
+          // processToolMessage, include tool_calls message and tool call results
+          (
+            requestPayload: RequestPayload,
+            toolCallMessage: any,
+            toolCallResult: any[],
+          ) => {
+            // @ts-ignore
+            requestPayload?.messages?.splice(
+              // @ts-ignore
+              requestPayload?.messages?.length,
+              0,
+              toolCallMessage,
+              ...toolCallResult,
+            );
+          },
+          options,
+        );
+      } else {
+        const res = await fetch(chatPath, chatPayload);
+        clearTimeout(requestTimeoutId);
+
+        const resJson = await res.json();
+        const message = this.extractMessage(resJson);
+        options.onFinish(message, res);
+      }
+    } catch (e) {
+      console.log("[Request] failed to make a chat request", e);
+      options.onError?.(e as Error);
+    }
+  }
+  async usage() {
+    return {
+      used: 0,
+      total: 0,
+    };
+  }
+
+  async models(): Promise<LLMModel[]> {
+    return [];
+  }
+}

app/client/platforms/google.ts

@@ -1,34 +1,87 @@
-import { Google, REQUEST_TIMEOUT_MS } from "@/app/constant";
-import { ChatOptions, getHeaders, LLMApi, LLMModel, LLMUsage } from "../api";
-import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
-import { getClientConfig } from "@/app/config/client";
-import { DEFAULT_API_HOST } from "@/app/constant";
-import Locale from "../../locales";
+import { ApiPath, Google, REQUEST_TIMEOUT_MS } from "@/app/constant";
 import {
-  EventStreamContentType,
-  fetchEventSource,
-} from "@fortaine/fetch-event-source";
-import { prettyObject } from "@/app/utils/format";
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  LLMUsage,
+  SpeechOptions,
+} from "../api";
+import {
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  usePluginStore,
+  ChatMessageTool,
+} from "@/app/store";
+import { stream } from "@/app/utils/chat";
+import { getClientConfig } from "@/app/config/client";
+import { GEMINI_BASE_URL } from "@/app/constant";
+
 import {
   getMessageTextContent,
   getMessageImages,
   isVisionModel,
 } from "@/app/utils";
+import { preProcessImageContent } from "@/app/utils/chat";
+import { nanoid } from "nanoid";
+import { RequestPayload } from "./openai";
+import { fetch } from "@/app/utils/stream";
 
 export class GeminiProApi implements LLMApi {
+  path(path: string, shouldStream = false): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.googleUrl;
+    }
+
+    const isApp = !!getClientConfig()?.isApp;
+    if (baseUrl.length === 0) {
+      baseUrl = isApp ? GEMINI_BASE_URL : ApiPath.Google;
+    }
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Google)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    let chatPath = [baseUrl, path].join("/");
+    if (shouldStream) {
+      chatPath += chatPath.includes("?") ? "&alt=sse" : "?alt=sse";
+    }
+
+    return chatPath;
+  }
   extractMessage(res: any) {
     console.log("[Response] gemini-pro response: ", res);
 
     return (
       res?.candidates?.at(0)?.content?.parts.at(0)?.text ||
+      res?.at(0)?.candidates?.at(0)?.content?.parts.at(0)?.text ||
       res?.error?.message ||
       ""
     );
   }
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
   async chat(options: ChatOptions): Promise<void> {
     const apiClient = this;
     let multimodal = false;
-    const messages = options.messages.map((v) => {
+
+    // try get base64image from local cache image_url
+    const _messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = await preProcessImageContent(v.content);
+      _messages.push({ role: v.role, content });
+    }
+    const messages = _messages.map((v) => {
       let parts: any[] = [{ text: getMessageTextContent(v) }];
       if (isVisionModel(options.config.model)) {
         const images = getMessageImages(v);
@@ -70,6 +123,9 @@ export class GeminiProApi implements LLMApi {
     // if (visionModel && messages.length > 1) {
     //   options.onError?.(new Error("Multiturn chat is not enabled for models/gemini-pro-vision"));
     // }
+
+    const accessStore = useAccessStore.getState();
+
     const modelConfig = {
       ...useAppConfig.getState().modelConfig,
       ...useChatStore.getState().currentSession().mask.modelConfig,
@@ -91,47 +147,33 @@ export class GeminiProApi implements LLMApi {
       safetySettings: [
         {
           category: "HARM_CATEGORY_HARASSMENT",
-          threshold: "BLOCK_ONLY_HIGH",
+          threshold: accessStore.googleSafetySettings,
         },
         {
           category: "HARM_CATEGORY_HATE_SPEECH",
-          threshold: "BLOCK_ONLY_HIGH",
+          threshold: accessStore.googleSafetySettings,
         },
         {
           category: "HARM_CATEGORY_SEXUALLY_EXPLICIT",
-          threshold: "BLOCK_ONLY_HIGH",
+          threshold: accessStore.googleSafetySettings,
         },
         {
           category: "HARM_CATEGORY_DANGEROUS_CONTENT",
-          threshold: "BLOCK_ONLY_HIGH",
+          threshold: accessStore.googleSafetySettings,
         },
       ],
     };
 
-    const accessStore = useAccessStore.getState();
-
-    let baseUrl = "";
-
-    if (accessStore.useCustomConfig) {
-      baseUrl = accessStore.googleUrl;
-    }
-
-    const isApp = !!getClientConfig()?.isApp;
-
     let shouldStream = !!options.config.stream;
     const controller = new AbortController();
     options.onController?.(controller);
     try {
-      if (!baseUrl && isApp) {
-        baseUrl = DEFAULT_API_HOST + "/api/proxy/google/";
-      }
-      baseUrl = `${baseUrl}/${Google.ChatPath(modelConfig.model)}`.replaceAll(
-        "//",
-        "/",
+      // https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/Streaming_REST.ipynb
+      const chatPath = this.path(
+        Google.ChatPath(modelConfig.model),
+        shouldStream,
       );
-      if (isApp) {
-        baseUrl += `?key=${accessStore.googleApiKey}`;
-      }
+
       const chatPayload = {
         method: "POST",
         body: JSON.stringify(requestPayload),
@@ -146,120 +188,86 @@ export class GeminiProApi implements LLMApi {
       );
 
       if (shouldStream) {
-        let responseText = "";
-        let remainText = "";
-        let finished = false;
+        const [tools, funcs] = usePluginStore
+          .getState()
+          .getAsTools(
+            useChatStore.getState().currentSession().mask?.plugin || [],
+          );
+        return stream(
+          chatPath,
+          requestPayload,
+          getHeaders(),
+          // @ts-ignore
+          tools.length > 0
+            ? // @ts-ignore
+              [{ functionDeclarations: tools.map((tool) => tool.function) }]
+            : [],
+          funcs,
+          controller,
+          // parseSSE
+          (text: string, runTools: ChatMessageTool[]) => {
+            // console.log("parseSSE", text, runTools);
+            const chunkJson = JSON.parse(text);
 
-        const finish = () => {
-          if (!finished) {
-            finished = true;
-            options.onFinish(responseText + remainText);
-          }
-        };
-
-        // animate response to make it looks smooth
-        function animateResponseText() {
-          if (finished || controller.signal.aborted) {
-            responseText += remainText;
-            finish();
-            return;
-          }
-
-          if (remainText.length > 0) {
-            const fetchCount = Math.max(1, Math.round(remainText.length / 60));
-            const fetchText = remainText.slice(0, fetchCount);
-            responseText += fetchText;
-            remainText = remainText.slice(fetchCount);
-            options.onUpdate?.(responseText, fetchText);
-          }
-
-          requestAnimationFrame(animateResponseText);
-        }
-
-        // start animaion
-        animateResponseText();
-
-        controller.signal.onabort = finish;
-
-        // https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/Streaming_REST.ipynb
-        const chatPath =
-          baseUrl.replace("generateContent", "streamGenerateContent") +
-          (baseUrl.indexOf("?") > -1 ? "&alt=sse" : "?alt=sse");
-        fetchEventSource(chatPath, {
-          ...chatPayload,
-          async onopen(res) {
-            clearTimeout(requestTimeoutId);
-            const contentType = res.headers.get("content-type");
-            console.log(
-              "[Gemini] request response content type: ",
-              contentType,
+            const functionCall = chunkJson?.candidates
+              ?.at(0)
+              ?.content.parts.at(0)?.functionCall;
+            if (functionCall) {
+              const { name, args } = functionCall;
+              runTools.push({
+                id: nanoid(),
+                type: "function",
+                function: {
+                  name,
+                  arguments: JSON.stringify(args), // utils.chat call function, using JSON.parse
+                },
+              });
+            }
+            return chunkJson?.candidates?.at(0)?.content.parts.at(0)?.text;
+          },
+          // processToolMessage, include tool_calls message and tool call results
+          (
+            requestPayload: RequestPayload,
+            toolCallMessage: any,
+            toolCallResult: any[],
+          ) => {
+            // @ts-ignore
+            requestPayload?.contents?.splice(
+              // @ts-ignore
+              requestPayload?.contents?.length,
+              0,
+              {
+                role: "model",
+                parts: toolCallMessage.tool_calls.map(
+                  (tool: ChatMessageTool) => ({
+                    functionCall: {
+                      name: tool?.function?.name,
+                      args: JSON.parse(tool?.function?.arguments as string),
+                    },
+                  }),
+                ),
+              },
+              // @ts-ignore
+              ...toolCallResult.map((result) => ({
+                role: "function",
+                parts: [
+                  {
+                    functionResponse: {
+                      name: result.name,
+                      response: {
+                        name: result.name,
+                        content: result.content, // TODO just text content...
+                      },
+                    },
+                  },
+                ],
+              })),
             );
-
-            if (contentType?.startsWith("text/plain")) {
-              responseText = await res.clone().text();
-              return finish();
-            }
-
-            if (
-              !res.ok ||
-              !res.headers
-                .get("content-type")
-                ?.startsWith(EventStreamContentType) ||
-              res.status !== 200
-            ) {
-              const responseTexts = [responseText];
-              let extraInfo = await res.clone().text();
-              try {
-                const resJson = await res.clone().json();
-                extraInfo = prettyObject(resJson);
-              } catch {}
-
-              if (res.status === 401) {
-                responseTexts.push(Locale.Error.Unauthorized);
-              }
-
-              if (extraInfo) {
-                responseTexts.push(extraInfo);
-              }
-
-              responseText = responseTexts.join("\n\n");
-
-              return finish();
-            }
           },
-          onmessage(msg) {
-            if (msg.data === "[DONE]" || finished) {
-              return finish();
-            }
-            const text = msg.data;
-            try {
-              const json = JSON.parse(text);
-              const delta = apiClient.extractMessage(json);
-
-              if (delta) {
-                remainText += delta;
-              }
-
-              const blockReason = json?.promptFeedback?.blockReason;
-              if (blockReason) {
-                // being blocked
-                console.log(`[Google] [Safety Ratings] result:`, blockReason);
-              }
-            } catch (e) {
-              console.error("[Request] parse error", text, msg);
-            }
-          },
-          onclose() {
-            finish();
-          },
-          onerror(e) {
-            options.onError?.(e);
-            throw e;
-          },
-          openWhenHidden: true,
-        });
+          options,
+        );
       } else {
-        const res = await fetch(baseUrl, chatPayload);
+        const res = await fetch(chatPath, chatPayload);
         clearTimeout(requestTimeoutId);
         const resJson = await res.json();
         if (resJson?.promptFeedback?.blockReason) {
@@ -272,7 +280,7 @@ export class GeminiProApi implements LLMApi {
           );
         }
         const message = apiClient.extractMessage(resJson);
-        options.onFinish(message);
+        options.onFinish(message, res);
       }
     } catch (e) {
       console.log("[Request] failed to make a chat request", e);
@@ -285,14 +293,4 @@ export class GeminiProApi implements LLMApi {
   async models(): Promise<LLMModel[]> {
     return [];
   }
-  path(path: string): string {
-    return "/api/google/" + path;
-  }
-}
-
-function ensureProperEnding(str: string) {
-  if (str.startsWith("[") && !str.endsWith("]")) {
-    return str + "]";
-  }
-  return str;
 }

app/client/platforms/iflytek.ts

@@ -0,0 +1,253 @@
+"use client";
+import {
+  ApiPath,
+  IFLYTEK_BASE_URL,
+  Iflytek,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+
+import {
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  SpeechOptions,
+} from "../api";
+import Locale from "../../locales";
+import {
+  EventStreamContentType,
+  fetchEventSource,
+} from "@fortaine/fetch-event-source";
+import { prettyObject } from "@/app/utils/format";
+import { getClientConfig } from "@/app/config/client";
+import { getMessageTextContent } from "@/app/utils";
+import { fetch } from "@/app/utils/stream";
+
+import { RequestPayload } from "./openai";
+
+export class SparkApi implements LLMApi {
+  private disableListModels = true;
+
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.iflytekUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      const apiPath = ApiPath.Iflytek;
+      baseUrl = isApp ? IFLYTEK_BASE_URL : apiPath;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Iflytek)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res.choices?.at(0)?.message?.content ?? "";
+  }
+
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async chat(options: ChatOptions) {
+    const messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = getMessageTextContent(v);
+      messages.push({ role: v.role, content });
+    }
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+        providerName: options.config.providerName,
+      },
+    };
+
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: options.config.stream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+      // max_tokens: Math.max(modelConfig.max_tokens, 1024),
+      // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
+    };
+
+    console.log("[Request] Spark payload: ", requestPayload);
+
+    const shouldStream = !!options.config.stream;
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(Iflytek.ChatPath);
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // Make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      if (shouldStream) {
+        let responseText = "";
+        let remainText = "";
+        let finished = false;
+        let responseRes: Response;
+
+        // Animate response text to make it look smooth
+        function animateResponseText() {
+          if (finished || controller.signal.aborted) {
+            responseText += remainText;
+            console.log("[Response Animation] finished");
+            return;
+          }
+
+          if (remainText.length > 0) {
+            const fetchCount = Math.max(1, Math.round(remainText.length / 60));
+            const fetchText = remainText.slice(0, fetchCount);
+            responseText += fetchText;
+            remainText = remainText.slice(fetchCount);
+            options.onUpdate?.(responseText, fetchText);
+          }
+
+          requestAnimationFrame(animateResponseText);
+        }
+
+        // Start animation
+        animateResponseText();
+
+        const finish = () => {
+          if (!finished) {
+            finished = true;
+            options.onFinish(responseText + remainText, responseRes);
+          }
+        };
+
+        controller.signal.onabort = finish;
+
+        fetchEventSource(chatPath, {
+          fetch: fetch as any,
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log("[Spark] request response content type: ", contentType);
+            responseRes = res;
+            if (contentType?.startsWith("text/plain")) {
+              responseText = await res.clone().text();
+              return finish();
+            }
+
+            // Handle different error scenarios
+            if (
+              !res.ok ||
+              !res.headers
+                .get("content-type")
+                ?.startsWith(EventStreamContentType) ||
+              res.status !== 200
+            ) {
+              let extraInfo = await res.clone().text();
+              try {
+                const resJson = await res.clone().json();
+                extraInfo = prettyObject(resJson);
+              } catch {}
+
+              if (res.status === 401) {
+                extraInfo = Locale.Error.Unauthorized;
+              }
+
+              options.onError?.(
+                new Error(
+                  `Request failed with status ${res.status}: ${extraInfo}`,
+                ),
+              );
+              return finish();
+            }
+          },
+          onmessage(msg) {
+            if (msg.data === "[DONE]" || finished) {
+              return finish();
+            }
+            const text = msg.data;
+            try {
+              const json = JSON.parse(text);
+              const choices = json.choices as Array<{
+                delta: { content: string };
+              }>;
+              const delta = choices[0]?.delta?.content;
+
+              if (delta) {
+                remainText += delta;
+              }
+            } catch (e) {
+              console.error("[Request] parse error", text);
+              options.onError?.(new Error(`Failed to parse response: ${text}`));
+            }
+          },
+          onclose() {
+            finish();
+          },
+          onerror(e) {
+            options.onError?.(e);
+            throw e;
+          },
+          openWhenHidden: true,
+        });
+      } else {
+        const res = await fetch(chatPath, chatPayload);
+        clearTimeout(requestTimeoutId);
+
+        if (!res.ok) {
+          const errorText = await res.text();
+          options.onError?.(
+            new Error(`Request failed with status ${res.status}: ${errorText}`),
+          );
+          return;
+        }
+
+        const resJson = await res.json();
+        const message = this.extractMessage(resJson);
+        options.onFinish(message, res);
+      }
+    } catch (e) {
+      console.log("[Request] failed to make a chat request", e);
+      options.onError?.(e as Error);
+    }
+  }
+
+  async usage() {
+    return {
+      used: 0,
+      total: 0,
+    };
+  }
+
+  async models(): Promise<LLMModel[]> {
+    return [];
+  }
+}

app/client/platforms/moonshot.ts

@@ -0,0 +1,200 @@
+"use client";
+// azure and openai, using same models. so using same LLMApi.
+import {
+  ApiPath,
+  MOONSHOT_BASE_URL,
+  Moonshot,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import {
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  ChatMessageTool,
+  usePluginStore,
+} from "@/app/store";
+import { stream } from "@/app/utils/chat";
+import {
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  SpeechOptions,
+} from "../api";
+import { getClientConfig } from "@/app/config/client";
+import { getMessageTextContent } from "@/app/utils";
+import { RequestPayload } from "./openai";
+import { fetch } from "@/app/utils/stream";
+
+export class MoonshotApi implements LLMApi {
+  private disableListModels = true;
+
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.moonshotUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      const apiPath = ApiPath.Moonshot;
+      baseUrl = isApp ? MOONSHOT_BASE_URL : apiPath;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Moonshot)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res.choices?.at(0)?.message?.content ?? "";
+  }
+
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async chat(options: ChatOptions) {
+    const messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = getMessageTextContent(v);
+      messages.push({ role: v.role, content });
+    }
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+        providerName: options.config.providerName,
+      },
+    };
+
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: options.config.stream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+      // max_tokens: Math.max(modelConfig.max_tokens, 1024),
+      // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
+    };
+
+    console.log("[Request] openai payload: ", requestPayload);
+
+    const shouldStream = !!options.config.stream;
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(Moonshot.ChatPath);
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      if (shouldStream) {
+        const [tools, funcs] = usePluginStore
+          .getState()
+          .getAsTools(
+            useChatStore.getState().currentSession().mask?.plugin || [],
+          );
+        return stream(
+          chatPath,
+          requestPayload,
+          getHeaders(),
+          tools as any,
+          funcs,
+          controller,
+          // parseSSE
+          (text: string, runTools: ChatMessageTool[]) => {
+            // console.log("parseSSE", text, runTools);
+            const json = JSON.parse(text);
+            const choices = json.choices as Array<{
+              delta: {
+                content: string;
+                tool_calls: ChatMessageTool[];
+              };
+            }>;
+            const tool_calls = choices[0]?.delta?.tool_calls;
+            if (tool_calls?.length > 0) {
+              const index = tool_calls[0]?.index;
+              const id = tool_calls[0]?.id;
+              const args = tool_calls[0]?.function?.arguments;
+              if (id) {
+                runTools.push({
+                  id,
+                  type: tool_calls[0]?.type,
+                  function: {
+                    name: tool_calls[0]?.function?.name as string,
+                    arguments: args,
+                  },
+                });
+              } else {
+                // @ts-ignore
+                runTools[index]["function"]["arguments"] += args;
+              }
+            }
+            return choices[0]?.delta?.content;
+          },
+          // processToolMessage, include tool_calls message and tool call results
+          (
+            requestPayload: RequestPayload,
+            toolCallMessage: any,
+            toolCallResult: any[],
+          ) => {
+            // @ts-ignore
+            requestPayload?.messages?.splice(
+              // @ts-ignore
+              requestPayload?.messages?.length,
+              0,
+              toolCallMessage,
+              ...toolCallResult,
+            );
+          },
+          options,
+        );
+      } else {
+        const res = await fetch(chatPath, chatPayload);
+        clearTimeout(requestTimeoutId);
+
+        const resJson = await res.json();
+        const message = this.extractMessage(resJson);
+        options.onFinish(message, res);
+      }
+    } catch (e) {
+      console.log("[Request] failed to make a chat request", e);
+      options.onError?.(e as Error);
+    }
+  }
+  async usage() {
+    return {
+      used: 0,
+      total: 0,
+    };
+  }
+
+  async models(): Promise<LLMModel[]> {
+    return [];
+  }
+}

app/client/platforms/openai.ts

@@ -2,16 +2,29 @@
 // azure and openai, using same models. so using same LLMApi.
 import {
   ApiPath,
-  DEFAULT_API_HOST,
+  OPENAI_BASE_URL,
   DEFAULT_MODELS,
   OpenaiPath,
   Azure,
   REQUEST_TIMEOUT_MS,
   ServiceProvider,
 } from "@/app/constant";
-import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+import {
+  ChatMessageTool,
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  usePluginStore,
+} from "@/app/store";
 import { collectModelsWithDefaultModel } from "@/app/utils/model";
+import {
+  preProcessImageContent,
+  uploadImage,
+  base64Image2Blob,
+  stream,
+} from "@/app/utils/chat";
 import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
+import { DalleSize, DalleQuality, DalleStyle } from "@/app/typing";
 
 import {
   ChatOptions,
@@ -20,19 +33,16 @@ import {
   LLMModel,
   LLMUsage,
   MultimodalContent,
+  SpeechOptions,
 } from "../api";
 import Locale from "../../locales";
-import {
-  EventStreamContentType,
-  fetchEventSource,
-} from "@fortaine/fetch-event-source";
-import { prettyObject } from "@/app/utils/format";
 import { getClientConfig } from "@/app/config/client";
 import {
   getMessageTextContent,
-  getMessageImages,
   isVisionModel,
+  isDalle3 as _isDalle3,
 } from "@/app/utils";
+import { fetch } from "@/app/utils/stream";
 
 export interface OpenAIListModelResponse {
   object: string;
@@ -55,6 +65,17 @@ export interface RequestPayload {
   frequency_penalty: number;
   top_p: number;
   max_tokens?: number;
+  max_completion_tokens?: number;
+}
+
+export interface DalleRequestPayload {
+  model: string;
+  prompt: string;
+  response_format: "url" | "b64_json";
+  n: number;
+  size: DalleSize;
+  quality: DalleQuality;
+  style: DalleStyle;
 }
 
 export class ChatGPTApi implements LLMApi {
@@ -79,7 +100,7 @@ export class ChatGPTApi implements LLMApi {
     if (baseUrl.length === 0) {
       const isApp = !!getClientConfig()?.isApp;
       const apiPath = isAzure ? ApiPath.Azure : ApiPath.OpenAI;
-      baseUrl = isApp ? DEFAULT_API_HOST + "/proxy" + apiPath : apiPath;
+      baseUrl = isApp ? OPENAI_BASE_URL : apiPath;
     }
 
     if (baseUrl.endsWith("/")) {
@@ -99,17 +120,69 @@ export class ChatGPTApi implements LLMApi {
     return cloudflareAIGatewayUrl([baseUrl, path].join("/"));
   }
 
-  extractMessage(res: any) {
-    return res.choices?.at(0)?.message?.content ?? "";
+  async extractMessage(res: any) {
+    if (res.error) {
+      return "```\n" + JSON.stringify(res, null, 4) + "\n```";
+    }
+    // dalle3 model return url, using url create image message
+    if (res.data) {
+      let url = res.data?.at(0)?.url ?? "";
+      const b64_json = res.data?.at(0)?.b64_json ?? "";
+      if (!url && b64_json) {
+        // uploadImage
+        url = await uploadImage(base64Image2Blob(b64_json, "image/png"));
+      }
+      return [
+        {
+          type: "image_url",
+          image_url: {
+            url,
+          },
+        },
+      ];
+    }
+    return res.choices?.at(0)?.message?.content ?? res;
+  }
+
+  async speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    const requestPayload = {
+      model: options.model,
+      input: options.input,
+      voice: options.voice,
+      response_format: options.response_format,
+      speed: options.speed,
+    };
+
+    console.log("[Request] openai speech payload: ", requestPayload);
+
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const speechPath = this.path(OpenaiPath.SpeechPath);
+      const speechPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      const res = await fetch(speechPath, speechPayload);
+      clearTimeout(requestTimeoutId);
+      return await res.arrayBuffer();
+    } catch (e) {
+      console.log("[Request] failed to make a speech request", e);
+      throw e;
+    }
   }
 
   async chat(options: ChatOptions) {
-    const visionModel = isVisionModel(options.config.model);
-    const messages = options.messages.map((v) => ({
-      role: v.role,
-      content: visionModel ? v.content : getMessageTextContent(v),
-    }));
-
     const modelConfig = {
       ...useAppConfig.getState().modelConfig,
       ...useChatStore.getState().currentSession().mask.modelConfig,
@@ -119,26 +192,62 @@ export class ChatGPTApi implements LLMApi {
       },
     };
 
-    const requestPayload: RequestPayload = {
-      messages,
-      stream: options.config.stream,
-      model: modelConfig.model,
-      temperature: modelConfig.temperature,
-      presence_penalty: modelConfig.presence_penalty,
-      frequency_penalty: modelConfig.frequency_penalty,
-      top_p: modelConfig.top_p,
-      // max_tokens: Math.max(modelConfig.max_tokens, 1024),
-      // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
-    };
+    let requestPayload: RequestPayload | DalleRequestPayload;
 
-    // add max_tokens to vision model
-    if (visionModel && modelConfig.model.includes("preview")) {
-      requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
+    const isDalle3 = _isDalle3(options.config.model);
+    const isO1 = options.config.model.startsWith("o1");
+    if (isDalle3) {
+      const prompt = getMessageTextContent(
+        options.messages.slice(-1)?.pop() as any,
+      );
+      requestPayload = {
+        model: options.config.model,
+        prompt,
+        // URLs are only valid for 60 minutes after the image has been generated.
+        response_format: "b64_json", // using b64_json, and save image in CacheStorage
+        n: 1,
+        size: options.config?.size ?? "1024x1024",
+        quality: options.config?.quality ?? "standard",
+        style: options.config?.style ?? "vivid",
+      };
+    } else {
+      const visionModel = isVisionModel(options.config.model);
+      const messages: ChatOptions["messages"] = [];
+      for (const v of options.messages) {
+        const content = visionModel
+          ? await preProcessImageContent(v.content)
+          : getMessageTextContent(v);
+        if (!(isO1 && v.role === "system"))
+          messages.push({ role: v.role, content });
+      }
+
+      // O1 not support image, tools (plugin in ChatGPTNextWeb) and system, stream, logprobs, temperature, top_p, n, presence_penalty, frequency_penalty yet.
+      requestPayload = {
+        messages,
+        stream: options.config.stream,
+        model: modelConfig.model,
+        temperature: !isO1 ? modelConfig.temperature : 1,
+        presence_penalty: !isO1 ? modelConfig.presence_penalty : 0,
+        frequency_penalty: !isO1 ? modelConfig.frequency_penalty : 0,
+        top_p: !isO1 ? modelConfig.top_p : 1,
+        // max_tokens: Math.max(modelConfig.max_tokens, 1024),
+        // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
+      };
+
+      // O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
+      if (isO1) {
+        requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
+      }
+
+      // add max_tokens to vision model
+      if (visionModel) {
+        requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
+      }
     }
 
     console.log("[Request] openai payload: ", requestPayload);
 
-    const shouldStream = !!options.config.stream;
+    const shouldStream = !isDalle3 && !!options.config.stream;
     const controller = new AbortController();
     options.onController?.(controller);
 
@@ -164,157 +273,101 @@ export class ChatGPTApi implements LLMApi {
             model?.provider?.providerName === ServiceProvider.Azure,
         );
         chatPath = this.path(
-          Azure.ChatPath(
+          (isDalle3 ? Azure.ImagePath : Azure.ChatPath)(
             (model?.displayName ?? model?.name) as string,
             useCustomConfig ? useAccessStore.getState().azureApiVersion : "",
           ),
         );
       } else {
-        chatPath = this.path(OpenaiPath.ChatPath);
+        chatPath = this.path(
+          isDalle3 ? OpenaiPath.ImagePath : OpenaiPath.ChatPath,
+        );
       }
-      const chatPayload = {
-        method: "POST",
-        body: JSON.stringify(requestPayload),
-        signal: controller.signal,
-        headers: getHeaders(),
-      };
-
-      // make a fetch request
-      const requestTimeoutId = setTimeout(
-        () => controller.abort(),
-        REQUEST_TIMEOUT_MS,
-      );
-
       if (shouldStream) {
-        let responseText = "";
-        let remainText = "";
-        let finished = false;
-
-        // animate response to make it looks smooth
-        function animateResponseText() {
-          if (finished || controller.signal.aborted) {
-            responseText += remainText;
-            console.log("[Response Animation] finished");
-            if (responseText?.length === 0) {
-              options.onError?.(new Error("empty response from server"));
+        let index = -1;
+        const [tools, funcs] = usePluginStore
+          .getState()
+          .getAsTools(
+            useChatStore.getState().currentSession().mask?.plugin || [],
+          );
+        // console.log("getAsTools", tools, funcs);
+        stream(
+          chatPath,
+          requestPayload,
+          getHeaders(),
+          tools as any,
+          funcs,
+          controller,
+          // parseSSE
+          (text: string, runTools: ChatMessageTool[]) => {
+            // console.log("parseSSE", text, runTools);
+            const json = JSON.parse(text);
+            const choices = json.choices as Array<{
+              delta: {
+                content: string;
+                tool_calls: ChatMessageTool[];
+              };
+            }>;
+            const tool_calls = choices[0]?.delta?.tool_calls;
+            if (tool_calls?.length > 0) {
+              const id = tool_calls[0]?.id;
+              const args = tool_calls[0]?.function?.arguments;
+              if (id) {
+                index += 1;
+                runTools.push({
+                  id,
+                  type: tool_calls[0]?.type,
+                  function: {
+                    name: tool_calls[0]?.function?.name as string,
+                    arguments: args,
+                  },
+                });
+              } else {
+                // @ts-ignore
+                runTools[index]["function"]["arguments"] += args;
+              }
             }
-            return;
-          }
-
-          if (remainText.length > 0) {
-            const fetchCount = Math.max(1, Math.round(remainText.length / 60));
-            const fetchText = remainText.slice(0, fetchCount);
-            responseText += fetchText;
-            remainText = remainText.slice(fetchCount);
-            options.onUpdate?.(responseText, fetchText);
-          }
-
-          requestAnimationFrame(animateResponseText);
-        }
-
-        // start animaion
-        animateResponseText();
-
-        const finish = () => {
-          if (!finished) {
-            finished = true;
-            options.onFinish(responseText + remainText);
-          }
+            return choices[0]?.delta?.content;
+          },
+          // processToolMessage, include tool_calls message and tool call results
+          (
+            requestPayload: RequestPayload,
+            toolCallMessage: any,
+            toolCallResult: any[],
+          ) => {
+            // reset index value
+            index = -1;
+            // @ts-ignore
+            requestPayload?.messages?.splice(
+              // @ts-ignore
+              requestPayload?.messages?.length,
+              0,
+              toolCallMessage,
+              ...toolCallResult,
+            );
+          },
+          options,
+        );
+      } else {
+        const chatPayload = {
+          method: "POST",
+          body: JSON.stringify(requestPayload),
+          signal: controller.signal,
+          headers: getHeaders(),
         };
 
-        controller.signal.onabort = finish;
+        // make a fetch request
+        const requestTimeoutId = setTimeout(
+          () => controller.abort(),
+          isDalle3 || isO1 ? REQUEST_TIMEOUT_MS * 4 : REQUEST_TIMEOUT_MS, // dalle3 using b64_json is slow.
+        );
 
-        fetchEventSource(chatPath, {
-          ...chatPayload,
-          async onopen(res) {
-            clearTimeout(requestTimeoutId);
-            const contentType = res.headers.get("content-type");
-            console.log(
-              "[OpenAI] request response content type: ",
-              contentType,
-            );
-
-            if (contentType?.startsWith("text/plain")) {
-              responseText = await res.clone().text();
-              return finish();
-            }
-
-            if (
-              !res.ok ||
-              !res.headers
-                .get("content-type")
-                ?.startsWith(EventStreamContentType) ||
-              res.status !== 200
-            ) {
-              const responseTexts = [responseText];
-              let extraInfo = await res.clone().text();
-              try {
-                const resJson = await res.clone().json();
-                extraInfo = prettyObject(resJson);
-              } catch {}
-
-              if (res.status === 401) {
-                responseTexts.push(Locale.Error.Unauthorized);
-              }
-
-              if (extraInfo) {
-                responseTexts.push(extraInfo);
-              }
-
-              responseText = responseTexts.join("\n\n");
-
-              return finish();
-            }
-          },
-          onmessage(msg) {
-            if (msg.data === "[DONE]" || finished) {
-              return finish();
-            }
-            const text = msg.data;
-            try {
-              const json = JSON.parse(text);
-              const choices = json.choices as Array<{
-                delta: { content: string };
-              }>;
-              const delta = choices[0]?.delta?.content;
-              const textmoderation = json?.prompt_filter_results;
-
-              if (delta) {
-                remainText += delta;
-              }
-
-              if (
-                textmoderation &&
-                textmoderation.length > 0 &&
-                ServiceProvider.Azure
-              ) {
-                const contentFilterResults =
-                  textmoderation[0]?.content_filter_results;
-                console.log(
-                  `[${ServiceProvider.Azure}] [Text Moderation] flagged categories result:`,
-                  contentFilterResults,
-                );
-              }
-            } catch (e) {
-              console.error("[Request] parse error", text, msg);
-            }
-          },
-          onclose() {
-            finish();
-          },
-          onerror(e) {
-            options.onError?.(e);
-            throw e;
-          },
-          openWhenHidden: true,
-        });
-      } else {
         const res = await fetch(chatPath, chatPayload);
         clearTimeout(requestTimeoutId);
 
         const resJson = await res.json();
-        const message = this.extractMessage(resJson);
-        options.onFinish(message);
+        const message = await this.extractMessage(resJson);
+        options.onFinish(message, res);
       }
     } catch (e) {
       console.log("[Request] failed to make a chat request", e);
@@ -400,20 +453,26 @@ export class ChatGPTApi implements LLMApi {
     });
 
     const resJson = (await res.json()) as OpenAIListModelResponse;
-    const chatModels = resJson.data?.filter((m) => m.id.startsWith("gpt-"));
+    const chatModels = resJson.data?.filter(
+      (m) => m.id.startsWith("gpt-") || m.id.startsWith("chatgpt-"),
+    );
     console.log("[Models]", chatModels);
 
     if (!chatModels) {
       return [];
     }
 
+    //由于目前 OpenAI 的 disableListModels 默认为 true，所以当前实际不会运行到这场
+    let seq = 1000; //同 Constant.ts 中的排序保持一致
     return chatModels.map((m) => ({
       name: m.id,
       available: true,
+      sorted: seq++,
       provider: {
         id: "openai",
         providerName: "OpenAI",
         providerType: "openai",
+        sorted: 1,
       },
     }));
   }

app/client/platforms/tencent.ts

@@ -0,0 +1,274 @@
+"use client";
+import { ApiPath, TENCENT_BASE_URL, REQUEST_TIMEOUT_MS } from "@/app/constant";
+import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+
+import {
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  MultimodalContent,
+  SpeechOptions,
+} from "../api";
+import Locale from "../../locales";
+import {
+  EventStreamContentType,
+  fetchEventSource,
+} from "@fortaine/fetch-event-source";
+import { prettyObject } from "@/app/utils/format";
+import { getClientConfig } from "@/app/config/client";
+import { getMessageTextContent, isVisionModel } from "@/app/utils";
+import mapKeys from "lodash-es/mapKeys";
+import mapValues from "lodash-es/mapValues";
+import isArray from "lodash-es/isArray";
+import isObject from "lodash-es/isObject";
+import { fetch } from "@/app/utils/stream";
+
+export interface OpenAIListModelResponse {
+  object: string;
+  data: Array<{
+    id: string;
+    object: string;
+    root: string;
+  }>;
+}
+
+interface RequestPayload {
+  Messages: {
+    Role: "system" | "user" | "assistant";
+    Content: string | MultimodalContent[];
+  }[];
+  Stream?: boolean;
+  Model: string;
+  Temperature: number;
+  TopP: number;
+}
+
+function capitalizeKeys(obj: any): any {
+  if (isArray(obj)) {
+    return obj.map(capitalizeKeys);
+  } else if (isObject(obj)) {
+    return mapValues(
+      mapKeys(obj, (value: any, key: string) =>
+        key.replace(/(^|_)(\w)/g, (m, $1, $2) => $2.toUpperCase()),
+      ),
+      capitalizeKeys,
+    );
+  } else {
+    return obj;
+  }
+}
+
+export class HunyuanApi implements LLMApi {
+  path(): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.tencentUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      baseUrl = isApp ? TENCENT_BASE_URL : ApiPath.Tencent;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Tencent)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl);
+    return baseUrl;
+  }
+
+  extractMessage(res: any) {
+    return res.Choices?.at(0)?.Message?.Content ?? "";
+  }
+
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async chat(options: ChatOptions) {
+    const visionModel = isVisionModel(options.config.model);
+    const messages = options.messages.map((v, index) => ({
+      // "Messages 中 system 角色必须位于列表的最开始"
+      role: index !== 0 && v.role === "system" ? "user" : v.role,
+      content: visionModel ? v.content : getMessageTextContent(v),
+    }));
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+      },
+    };
+
+    const requestPayload: RequestPayload = capitalizeKeys({
+      model: modelConfig.model,
+      messages,
+      temperature: modelConfig.temperature,
+      top_p: modelConfig.top_p,
+      stream: options.config.stream,
+    });
+
+    console.log("[Request] Tencent payload: ", requestPayload);
+
+    const shouldStream = !!options.config.stream;
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path();
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      if (shouldStream) {
+        let responseText = "";
+        let remainText = "";
+        let finished = false;
+        let responseRes: Response;
+
+        // animate response to make it looks smooth
+        function animateResponseText() {
+          if (finished || controller.signal.aborted) {
+            responseText += remainText;
+            console.log("[Response Animation] finished");
+            if (responseText?.length === 0) {
+              options.onError?.(new Error("empty response from server"));
+            }
+            return;
+          }
+
+          if (remainText.length > 0) {
+            const fetchCount = Math.max(1, Math.round(remainText.length / 60));
+            const fetchText = remainText.slice(0, fetchCount);
+            responseText += fetchText;
+            remainText = remainText.slice(fetchCount);
+            options.onUpdate?.(responseText, fetchText);
+          }
+
+          requestAnimationFrame(animateResponseText);
+        }
+
+        // start animaion
+        animateResponseText();
+
+        const finish = () => {
+          if (!finished) {
+            finished = true;
+            options.onFinish(responseText + remainText, responseRes);
+          }
+        };
+
+        controller.signal.onabort = finish;
+
+        fetchEventSource(chatPath, {
+          fetch: fetch as any,
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log(
+              "[Tencent] request response content type: ",
+              contentType,
+            );
+            responseRes = res;
+            if (contentType?.startsWith("text/plain")) {
+              responseText = await res.clone().text();
+              return finish();
+            }
+
+            if (
+              !res.ok ||
+              !res.headers
+                .get("content-type")
+                ?.startsWith(EventStreamContentType) ||
+              res.status !== 200
+            ) {
+              const responseTexts = [responseText];
+              let extraInfo = await res.clone().text();
+              try {
+                const resJson = await res.clone().json();
+                extraInfo = prettyObject(resJson);
+              } catch {}
+
+              if (res.status === 401) {
+                responseTexts.push(Locale.Error.Unauthorized);
+              }
+
+              if (extraInfo) {
+                responseTexts.push(extraInfo);
+              }
+
+              responseText = responseTexts.join("\n\n");
+
+              return finish();
+            }
+          },
+          onmessage(msg) {
+            if (msg.data === "[DONE]" || finished) {
+              return finish();
+            }
+            const text = msg.data;
+            try {
+              const json = JSON.parse(text);
+              const choices = json.Choices as Array<{
+                Delta: { Content: string };
+              }>;
+              const delta = choices[0]?.Delta?.Content;
+              if (delta) {
+                remainText += delta;
+              }
+            } catch (e) {
+              console.error("[Request] parse error", text, msg);
+            }
+          },
+          onclose() {
+            finish();
+          },
+          onerror(e) {
+            options.onError?.(e);
+            throw e;
+          },
+          openWhenHidden: true,
+        });
+      } else {
+        const res = await fetch(chatPath, chatPayload);
+        clearTimeout(requestTimeoutId);
+
+        const resJson = await res.json();
+        const message = this.extractMessage(resJson);
+        options.onFinish(message, res);
+      }
+    } catch (e) {
+      console.log("[Request] failed to make a chat request", e);
+      options.onError?.(e as Error);
+    }
+  }
+  async usage() {
+    return {
+      used: 0,
+      total: 0,
+    };
+  }
+
+  async models(): Promise<LLMModel[]> {
+    return [];
+  }
+}

app/client/platforms/xai.ts

@@ -0,0 +1,193 @@
+"use client";
+// azure and openai, using same models. so using same LLMApi.
+import { ApiPath, XAI_BASE_URL, XAI, REQUEST_TIMEOUT_MS } from "@/app/constant";
+import {
+  useAccessStore,
+  useAppConfig,
+  useChatStore,
+  ChatMessageTool,
+  usePluginStore,
+} from "@/app/store";
+import { stream } from "@/app/utils/chat";
+import {
+  ChatOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  SpeechOptions,
+} from "../api";
+import { getClientConfig } from "@/app/config/client";
+import { getMessageTextContent } from "@/app/utils";
+import { RequestPayload } from "./openai";
+import { fetch } from "@/app/utils/stream";
+
+export class XAIApi implements LLMApi {
+  private disableListModels = true;
+
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.xaiUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      const apiPath = ApiPath.XAI;
+      baseUrl = isApp ? XAI_BASE_URL : apiPath;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.XAI)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res.choices?.at(0)?.message?.content ?? "";
+  }
+
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async chat(options: ChatOptions) {
+    const messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = getMessageTextContent(v);
+      messages.push({ role: v.role, content });
+    }
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+        providerName: options.config.providerName,
+      },
+    };
+
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: options.config.stream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+    };
+
+    console.log("[Request] xai payload: ", requestPayload);
+
+    const shouldStream = !!options.config.stream;
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(XAI.ChatPath);
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: getHeaders(),
+      };
+
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
+
+      if (shouldStream) {
+        const [tools, funcs] = usePluginStore
+          .getState()
+          .getAsTools(
+            useChatStore.getState().currentSession().mask?.plugin || [],
+          );
+        return stream(
+          chatPath,
+          requestPayload,
+          getHeaders(),
+          tools as any,
+          funcs,
+          controller,
+          // parseSSE
+          (text: string, runTools: ChatMessageTool[]) => {
+            // console.log("parseSSE", text, runTools);
+            const json = JSON.parse(text);
+            const choices = json.choices as Array<{
+              delta: {
+                content: string;
+                tool_calls: ChatMessageTool[];
+              };
+            }>;
+            const tool_calls = choices[0]?.delta?.tool_calls;
+            if (tool_calls?.length > 0) {
+              const index = tool_calls[0]?.index;
+              const id = tool_calls[0]?.id;
+              const args = tool_calls[0]?.function?.arguments;
+              if (id) {
+                runTools.push({
+                  id,
+                  type: tool_calls[0]?.type,
+                  function: {
+                    name: tool_calls[0]?.function?.name as string,
+                    arguments: args,
+                  },
+                });
+              } else {
+                // @ts-ignore
+                runTools[index]["function"]["arguments"] += args;
+              }
+            }
+            return choices[0]?.delta?.content;
+          },
+          // processToolMessage, include tool_calls message and tool call results
+          (
+            requestPayload: RequestPayload,
+            toolCallMessage: any,
+            toolCallResult: any[],
+          ) => {
+            // @ts-ignore
+            requestPayload?.messages?.splice(
+              // @ts-ignore
+              requestPayload?.messages?.length,
+              0,
+              toolCallMessage,
+              ...toolCallResult,
+            );
+          },
+          options,
+        );
+      } else {
+        const res = await fetch(chatPath, chatPayload);
+        clearTimeout(requestTimeoutId);
+
+        const resJson = await res.json();
+        const message = this.extractMessage(resJson);
+        options.onFinish(message, res);
+      }
+    } catch (e) {
+      console.log("[Request] failed to make a chat request", e);
+      options.onError?.(e as Error);
+    }
+  }
+  async usage() {
+    return {
+      used: 0,
+      total: 0,
+    };
+  }
+
+  async models(): Promise<LLMModel[]> {
+    return [];
+  }
+}