Merge remote-tracking branch 'upstream/main'

app/client/platforms/alibaba.ts

@@ -0,0 +1,284 @@
+"use client";
+import {
+  ApiPath,
+  Alibaba,
+  ALIBABA_BASE_URL,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+
+import {
+  AgentChatOptions,
+  ChatOptions,
+  CreateRAGStoreOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  MultimodalContent,
+  SpeechOptions,
+  TranscriptionOptions,
+} 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";
+
+export interface OpenAIListModelResponse {
+  object: string;
+  data: Array<{
+    id: string;
+    object: string;
+    root: string;
+  }>;
+}
+
+interface RequestInput {
+  messages: {
+    role: "system" | "user" | "assistant";
+    content: string | MultimodalContent[];
+  }[];
+}
+interface RequestParam {
+  result_format: string;
+  incremental_output?: boolean;
+  temperature: number;
+  repetition_penalty?: number;
+  top_p: number;
+  max_tokens?: number;
+}
+interface RequestPayload {
+  model: string;
+  input: RequestInput;
+  parameters: RequestParam;
+}
+
+export class QwenApi implements LLMApi {
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+  transcription(options: TranscriptionOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  toolAgentChat(options: AgentChatOptions): Promise<void> {
+    throw new Error("Method not implemented.");
+  }
+  createRAGStore(options: CreateRAGStoreOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.alibabaUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      baseUrl = isApp ? ALIBABA_BASE_URL : ApiPath.Alibaba;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Alibaba)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res?.output?.choices?.at(0)?.message?.content ?? "";
+  }
+
+  async chat(options: ChatOptions) {
+    const messages = options.messages.map((v) => ({
+      role: v.role,
+      content: getMessageTextContent(v),
+    }));
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+      },
+    };
+
+    const shouldStream = !!options.config.stream;
+    const requestPayload: RequestPayload = {
+      model: modelConfig.model,
+      input: {
+        messages,
+      },
+      parameters: {
+        result_format: "message",
+        incremental_output: shouldStream,
+        temperature: modelConfig.temperature,
+        // max_tokens: modelConfig.max_tokens,
+        top_p: modelConfig.top_p === 1 ? 0.99 : modelConfig.top_p, // qwen top_p is should be < 1
+      },
+    };
+
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(Alibaba.ChatPath);
+      const chatPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: {
+          ...getHeaders(),
+          "X-DashScope-SSE": shouldStream ? "enable" : "disable",
+        },
+      };
+
+      // 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"));
+            }
+            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);
+          }
+        };
+
+        controller.signal.onabort = finish;
+
+        fetchEventSource(chatPath, {
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log(
+              "[Alibaba] 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.output.choices as Array<{
+                message: { content: string };
+              }>;
+              const delta = choices[0]?.message?.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);
+      }
+    } 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 [];
+  }
+}
+export { Alibaba };

app/client/platforms/anthropic.ts

@@ -3,6 +3,7 @@ import {
   AgentChatOptions,
   ChatOptions,
   CreateRAGStoreOptions,
+  getHeaders,
   LLMApi,
   MultimodalContent,
   SpeechOptions,
@@ -11,7 +12,6 @@ import {
 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 {
   EventStreamContentType,
   fetchEventSource,
@@ -20,6 +20,8 @@ import {
 import Locale from "../../locales";
 import { prettyObject } from "@/app/utils/format";
 import { getMessageTextContent, isVisionModel } from "@/app/utils";
+import { preProcessImageContent } from "@/app/utils/chat";
+import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
 
 export type MultiBlockContent = {
   type: "image" | "text";
@@ -112,7 +114,12 @@ 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"];
 
@@ -210,11 +217,10 @@ export class ClaudeApi implements LLMApi {
       body: JSON.stringify(requestBody),
       signal: controller.signal,
       headers: {
-        "Content-Type": "application/json",
-        Accept: "application/json",
-        "x-api-key": accessStore.anthropicApiKey,
+        ...getHeaders(), // get common headers
         "anthropic-version": accessStore.anthropicApiVersion,
-        Authorization: getAuthKey(accessStore.anthropicApiKey),
+        // do not send `anthropicApiKey` in browser!!!
+        // Authorization: getAuthKey(accessStore.anthropicApiKey),
       },
     };
 
@@ -396,7 +402,8 @@ export class ClaudeApi implements LLMApi {
 
     baseUrl = trimEnd(baseUrl, "/");
 
-    return `${baseUrl}/${path}`;
+    // try rebuild url, when using cloudflare ai gateway in client
+    return cloudflareAIGatewayUrl(`${baseUrl}/${path}`);
   }
 }
 
@@ -409,27 +416,3 @@ function trimEnd(s: string, end = " ") {
 
   return s;
 }
-
-function bearer(value: string) {
-  return `Bearer ${value.trim()}`;
-}
-
-function getAuthKey(apiKey = "") {
-  const accessStore = useAccessStore.getState();
-  const isApp = !!getClientConfig()?.isApp;
-  let authKey = "";
-
-  if (apiKey) {
-    // use user's api key first
-    authKey = bearer(apiKey);
-  } else if (
-    accessStore.enabledAccessControl() &&
-    !isApp &&
-    !!accessStore.accessCode
-  ) {
-    // or use access code
-    authKey = bearer(ACCESS_CODE_PREFIX + accessStore.accessCode);
-  }
-
-  return authKey;
-}

app/client/platforms/baidu.ts

@@ -0,0 +1,289 @@
+"use client";
+import {
+  ApiPath,
+  Baidu,
+  BAIDU_BASE_URL,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+import { getAccessToken } from "@/app/utils/baidu";
+
+import {
+  AgentChatOptions,
+  ChatOptions,
+  CreateRAGStoreOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  MultimodalContent,
+  SpeechOptions,
+  TranscriptionOptions,
+} 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";
+
+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;
+  presence_penalty: number;
+  frequency_penalty: number;
+  top_p: number;
+  max_tokens?: number;
+}
+
+export class ErnieApi implements LLMApi {
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+  transcription(options: TranscriptionOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  toolAgentChat(options: AgentChatOptions): Promise<void> {
+    throw new Error("Method not implemented.");
+  }
+  createRAGStore(options: CreateRAGStoreOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.baiduUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      // do not use proxy for baidubce api
+      baseUrl = isApp ? BAIDU_BASE_URL : ApiPath.Baidu;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.Baidu)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  async chat(options: ChatOptions) {
+    const messages = options.messages.map((v) => ({
+      role: 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: " ",
+      });
+    }
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+      },
+    };
+
+    const shouldStream = !!options.config.stream;
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: shouldStream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+    };
+
+    console.log("[Request] Baidu payload: ", requestPayload);
+
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      let chatPath = this.path(Baidu.ChatPath(modelConfig.model));
+
+      // getAccessToken can not run in browser, because cors error
+      if (!!getClientConfig()?.isApp) {
+        const accessStore = useAccessStore.getState();
+        if (accessStore.useCustomConfig) {
+          if (accessStore.isValidBaidu()) {
+            const { access_token } = await getAccessToken(
+              accessStore.baiduApiKey,
+              accessStore.baiduSecretKey,
+            );
+            chatPath = `${chatPath}${
+              chatPath.includes("?") ? "&" : "?"
+            }access_token=${access_token}`;
+          }
+        }
+      }
+      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"));
+            }
+            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);
+          }
+        };
+
+        controller.signal.onabort = finish;
+
+        fetchEventSource(chatPath, {
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log("[Baidu] 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 delta = json?.result;
+              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 = resJson?.result;
+        options.onFinish(message);
+      }
+    } 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 [];
+  }
+}
+export { Baidu };

app/client/platforms/bytedance.ts

@@ -0,0 +1,271 @@
+"use client";
+import {
+  ApiPath,
+  ByteDance,
+  BYTEDANCE_BASE_URL,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
+import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+
+import {
+  AgentChatOptions,
+  ChatOptions,
+  CreateRAGStoreOptions,
+  getHeaders,
+  LLMApi,
+  LLMModel,
+  MultimodalContent,
+  SpeechOptions,
+  TranscriptionOptions,
+} 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";
+
+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;
+  presence_penalty: number;
+  frequency_penalty: number;
+  top_p: number;
+  max_tokens?: number;
+}
+
+export class DoubaoApi implements LLMApi {
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+  transcription(options: TranscriptionOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  toolAgentChat(options: AgentChatOptions): Promise<void> {
+    throw new Error("Method not implemented.");
+  }
+  createRAGStore(options: CreateRAGStoreOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.bytedanceUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      baseUrl = isApp ? BYTEDANCE_BASE_URL : ApiPath.ByteDance;
+    }
+
+    if (baseUrl.endsWith("/")) {
+      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
+    }
+    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.ByteDance)) {
+      baseUrl = "https://" + baseUrl;
+    }
+
+    console.log("[Proxy Endpoint] ", baseUrl, path);
+
+    return [baseUrl, path].join("/");
+  }
+
+  extractMessage(res: any) {
+    return res.choices?.at(0)?.message?.content ?? "";
+  }
+
+  async chat(options: ChatOptions) {
+    const messages = options.messages.map((v) => ({
+      role: v.role,
+      content: getMessageTextContent(v),
+    }));
+
+    const modelConfig = {
+      ...useAppConfig.getState().modelConfig,
+      ...useChatStore.getState().currentSession().mask.modelConfig,
+      ...{
+        model: options.config.model,
+      },
+    };
+
+    const shouldStream = !!options.config.stream;
+    const requestPayload: RequestPayload = {
+      messages,
+      stream: shouldStream,
+      model: modelConfig.model,
+      temperature: modelConfig.temperature,
+      presence_penalty: modelConfig.presence_penalty,
+      frequency_penalty: modelConfig.frequency_penalty,
+      top_p: modelConfig.top_p,
+    };
+
+    const controller = new AbortController();
+    options.onController?.(controller);
+
+    try {
+      const chatPath = this.path(ByteDance.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"));
+            }
+            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);
+          }
+        };
+
+        controller.signal.onabort = finish;
+
+        fetchEventSource(chatPath, {
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log(
+              "[ByteDance] 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;
+              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);
+      }
+    } 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 [];
+  }
+}
+export { ByteDance };

app/client/platforms/google.ts

@@ -1,4 +1,4 @@
-import { Google, REQUEST_TIMEOUT_MS } from "@/app/constant";
+import { ApiPath, Google, REQUEST_TIMEOUT_MS } from "@/app/constant";
 import {
   AgentChatOptions,
   ChatOptions,
@@ -13,25 +13,61 @@ import {
 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 {
+  EventStreamContentType,
+  fetchEventSource,
+} from "@fortaine/fetch-event-source";
+import { prettyObject } from "@/app/utils/format";
 import {
   getMessageTextContent,
   getMessageImages,
   isVisionModel,
 } from "@/app/utils";
+import { preProcessImageContent } from "@/app/utils/chat";
+import options from "cheerio/lib/options";
 
 export class GeminiProApi implements LLMApi {
-  createRAGStore(options: CreateRAGStoreOptions): Promise<string> {
+  speech(options: SpeechOptions): Promise<ArrayBuffer> {
     throw new Error("Method not implemented.");
   }
   transcription(options: TranscriptionOptions): Promise<string> {
     throw new Error("Method not implemented.");
   }
-  speech(options: SpeechOptions): Promise<ArrayBuffer> {
-    throw new Error("Method not implemented.");
-  }
   toolAgentChat(options: AgentChatOptions): Promise<void> {
     throw new Error("Method not implemented.");
   }
+  createRAGStore(options: CreateRAGStoreOptions): Promise<string> {
+    throw new Error("Method not implemented.");
+  }
+  path(path: string): string {
+    const accessStore = useAccessStore.getState();
+
+    let baseUrl = "";
+    if (accessStore.useCustomConfig) {
+      baseUrl = accessStore.googleUrl;
+    }
+
+    if (baseUrl.length === 0) {
+      const isApp = !!getClientConfig()?.isApp;
+      baseUrl = isApp
+        ? DEFAULT_API_HOST + `/api/proxy/google?key=${accessStore.googleApiKey}`
+        : 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("/");
+
+    chatPath += chatPath.includes("?") ? "&alt=sse" : "?alt=sse";
+    return chatPath;
+  }
   extractMessage(res: any) {
     console.log("[Response] gemini-pro response: ", res);
 
@@ -42,9 +78,16 @@ export class GeminiProApi implements LLMApi {
     );
   }
   async chat(options: ChatOptions): Promise<void> {
-    // const apiClient = this;
+    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);
@@ -86,6 +129,9 @@ 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,
@@ -107,50 +153,30 @@ 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 {
-      // let baseUrl = accessStore.googleUrl;
+      // https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/Streaming_REST.ipynb
+      const chatPath = this.path(Google.ChatPath(modelConfig.model));
 
-      if (!baseUrl) {
-        baseUrl = isApp
-          ? DEFAULT_API_HOST +
-            "/api/proxy/google/" +
-            Google.ChatPath(modelConfig.model)
-          : this.path(Google.ChatPath(modelConfig.model));
-      }
-
-      if (isApp) {
-        baseUrl += `?key=${accessStore.googleApiKey}`;
-      }
       const chatPayload = {
         method: "POST",
         body: JSON.stringify(requestPayload),
@@ -169,10 +195,11 @@ export class GeminiProApi implements LLMApi {
         let remainText = "";
         let finished = false;
 
-        let existingTexts: string[] = [];
         const finish = () => {
-          finished = true;
-          options.onFinish(existingTexts.join(""));
+          if (!finished) {
+            finished = true;
+            options.onFinish(responseText + remainText);
+          }
         };
 
         // animate response to make it looks smooth
@@ -197,75 +224,83 @@ export class GeminiProApi implements LLMApi {
         // start animaion
         animateResponseText();
 
-        fetch(
-          baseUrl.replace("generateContent", "streamGenerateContent"),
-          chatPayload,
-        )
-          .then((response) => {
-            const reader = response?.body?.getReader();
-            const decoder = new TextDecoder();
-            let partialData = "";
+        controller.signal.onabort = finish;
 
-            return reader?.read().then(function processText({
-              done,
-              value,
-            }): Promise<any> {
-              if (done) {
-                if (response.status !== 200) {
-                  try {
-                    let data = JSON.parse(ensureProperEnding(partialData));
-                    if (data && data[0].error) {
-                      options.onError?.(new Error(data[0].error.message));
-                    } else {
-                      options.onError?.(new Error("Request failed"));
-                    }
-                  } catch (_) {
-                    options.onError?.(new Error("Request failed"));
-                  }
-                }
+        fetchEventSource(chatPath, {
+          ...chatPayload,
+          async onopen(res) {
+            clearTimeout(requestTimeoutId);
+            const contentType = res.headers.get("content-type");
+            console.log(
+              "[Gemini] request response content type: ",
+              contentType,
+            );
 
-                console.log("Stream complete");
-                // options.onFinish(responseText + remainText);
-                finished = true;
-                return Promise.resolve();
-              }
-
-              partialData += decoder.decode(value, { stream: true });
+            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 {
-                let data = JSON.parse(ensureProperEnding(partialData));
+                const resJson = await res.clone().json();
+                extraInfo = prettyObject(resJson);
+              } catch {}
 
-                const textArray = data.reduce(
-                  (acc: string[], item: { candidates: any[] }) => {
-                    const texts = item.candidates.map((candidate) =>
-                      candidate.content.parts
-                        .map((part: { text: any }) => part.text)
-                        .join(""),
-                    );
-                    return acc.concat(texts);
-                  },
-                  [],
-                );
-
-                if (textArray.length > existingTexts.length) {
-                  const deltaArray = textArray.slice(existingTexts.length);
-                  existingTexts = textArray;
-                  remainText += deltaArray.join("");
-                }
-              } catch (error) {
-                // console.log("[Response Animation] error: ", error,partialData);
-                // skip error message when parsing json
+              if (res.status === 401) {
+                responseTexts.push(Locale.Error.Unauthorized);
               }
 
-              return reader.read().then(processText);
-            });
-          })
-          .catch((error) => {
-            console.error("Error:", error);
-            options.onError?.(error as Error);
-          });
+              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,
+        });
       } else {
-        const res = await fetch(baseUrl, chatPayload);
+        const res = await fetch(chatPath, chatPayload);
         clearTimeout(requestTimeoutId);
 
         const resJson = await res.json();
@@ -279,7 +314,7 @@ export class GeminiProApi implements LLMApi {
             ),
           );
         }
-        const message = this.extractMessage(resJson);
+        const message = apiClient.extractMessage(resJson);
         options.onFinish(message);
       }
     } catch (e) {
@@ -293,14 +328,4 @@ 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/openai.ts

@@ -1,13 +1,18 @@
 "use client";
+// azure and openai, using same models. so using same LLMApi.
 import {
   ApiPath,
   DEFAULT_API_HOST,
   DEFAULT_MODELS,
   OpenaiPath,
+  Azure,
   REQUEST_TIMEOUT_MS,
   ServiceProvider,
 } from "@/app/constant";
 import { useAccessStore, useAppConfig, useChatStore } from "@/app/store";
+import { collectModelsWithDefaultModel } from "@/app/utils/model";
+import { preProcessImageContent } from "@/app/utils/chat";
+import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
 
 import {
   AgentChatOptions,
@@ -28,7 +33,6 @@ import {
 } from "@fortaine/fetch-event-source";
 import { prettyObject } from "@/app/utils/format";
 import { getClientConfig } from "@/app/config/client";
-import { makeAzurePath } from "@/app/azure";
 import {
   getMessageTextContent,
   getMessageImages,
@@ -44,7 +48,7 @@ export interface OpenAIListModelResponse {
   }>;
 }
 
-interface RequestPayload {
+export interface RequestPayload {
   messages: {
     role: "system" | "user" | "assistant";
     content: string | MultimodalContent[];
@@ -66,39 +70,38 @@ export class ChatGPTApi implements LLMApi {
 
     let baseUrl = "";
 
+    const isAzure = path.includes("deployments");
     if (accessStore.useCustomConfig) {
-      const isAzure = accessStore.provider === ServiceProvider.Azure;
-
       if (isAzure && !accessStore.isValidAzure()) {
         throw Error(
           "incomplete azure config, please check it in your settings page",
         );
       }
 
-      if (isAzure) {
-        path = makeAzurePath(path, accessStore.azureApiVersion);
-      }
-
       baseUrl = isAzure ? accessStore.azureUrl : accessStore.openaiUrl;
     }
 
     if (baseUrl.length === 0) {
       const isApp = !!getClientConfig()?.isApp;
-      baseUrl = isApp
-        ? DEFAULT_API_HOST + "/proxy" + ApiPath.OpenAI
-        : ApiPath.OpenAI;
+      const apiPath = isAzure ? ApiPath.Azure : ApiPath.OpenAI;
+      baseUrl = isApp ? DEFAULT_API_HOST + "/proxy" + apiPath : apiPath;
     }
 
     if (baseUrl.endsWith("/")) {
       baseUrl = baseUrl.slice(0, baseUrl.length - 1);
     }
-    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.OpenAI)) {
+    if (
+      !baseUrl.startsWith("http") &&
+      !isAzure &&
+      !baseUrl.startsWith(ApiPath.OpenAI)
+    ) {
       baseUrl = "https://" + baseUrl;
     }
 
     console.log("[Proxy Endpoint] ", baseUrl, path);
 
-    return [baseUrl, path].join("/");
+    // try rebuild url, when using cloudflare ai gateway in client
+    return cloudflareAIGatewayUrl([baseUrl, path].join("/"));
   }
 
   extractMessage(res: any) {
@@ -186,16 +189,20 @@ export class ChatGPTApi implements LLMApi {
 
   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 messages: ChatOptions["messages"] = [];
+    for (const v of options.messages) {
+      const content = visionModel
+        ? await preProcessImageContent(v.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,
       },
     };
 
@@ -223,7 +230,35 @@ export class ChatGPTApi implements LLMApi {
     options.onController?.(controller);
 
     try {
-      const chatPath = this.path(OpenaiPath.ChatPath, modelConfig.model);
+      let chatPath = "";
+      if (modelConfig.providerName === ServiceProvider.Azure) {
+        // find model, and get displayName as deployName
+        const { models: configModels, customModels: configCustomModels } =
+          useAppConfig.getState();
+        const {
+          defaultModel,
+          customModels: accessCustomModels,
+          useCustomConfig,
+        } = useAccessStore.getState();
+        const models = collectModelsWithDefaultModel(
+          configModels,
+          [configCustomModels, accessCustomModels].join(","),
+          defaultModel,
+        );
+        const model = models.find(
+          (model) =>
+            model.name === modelConfig.model &&
+            model?.provider?.providerName === ServiceProvider.Azure,
+        );
+        chatPath = this.path(
+          Azure.ChatPath(
+            (model?.displayName ?? model?.name) as string,
+            useCustomConfig ? useAccessStore.getState().azureApiVersion : "",
+          ),
+        );
+      } else {
+        chatPath = this.path(OpenaiPath.ChatPath);
+      }
       const chatPayload = {
         method: "POST",
         body: JSON.stringify(requestPayload),

app/client/platforms/utils.ts

@@ -1,5 +1,4 @@
-import { getClientApi } from "@/app/utils";
-import { ClientApi, getHeaders } from "../api";
+import { ClientApi, getClientApi, getHeaders } from "../api";
 import { ChatSession } from "@/app/store";
 
 export interface FileInfo {
@@ -37,7 +36,7 @@ export class FileApi {
 
   async uploadForRag(file: any, session: ChatSession): Promise<FileInfo> {
     var fileInfo = await this.upload(file);
-    var api: ClientApi = getClientApi(session.mask.modelConfig.model);
+    var api: ClientApi = getClientApi(session.mask.modelConfig.providerName);
     let partial = await api.llm.createRAGStore({
       chatSessionId: session.id,
       fileInfos: [fileInfo],