Release 0.0.33

更新了英文的README_EN.md

.github/workflows/sync.yml

@@ -0,0 +1,48 @@
+name: Upstream Sync
+
+permissions:
+  contents: write
+  issues: write
+  actions: write
+
+on:
+  schedule:
+    - cron: '0 * * * *' # every hour
+  workflow_dispatch:
+
+jobs:
+  sync_latest_from_upstream:
+    name: Sync latest commits from upstream repo
+    runs-on: ubuntu-latest
+    if: ${{ github.event.repository.fork }}
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Clean issue notice
+        uses: actions-cool/issues-helper@v3
+        with:
+          actions: 'close-issues'
+          labels: '🚨 Sync Fail'
+
+      - name: Sync upstream changes
+        id: sync
+        uses: aormsby/Fork-Sync-With-Upstream-action@v3.4
+        with:
+          upstream_sync_repo: LLM-Red-Team/kimi-free-api
+          upstream_sync_branch: master
+          target_sync_branch: master
+          target_repo_token: ${{ secrets.GITHUB_TOKEN }} # automatically generated, no need to set
+          test_mode: false
+
+      - name: Sync check
+        if: failure()
+        uses: actions-cool/issues-helper@v3
+        with:
+          actions: 'create-issue'
+          title: '🚨 同步失败 | Sync Fail'
+          labels: '🚨 Sync Fail'
+          body: |
+            Due to a change in the workflow file of the LLM-Red-Team/kimi-free-api upstream repository, GitHub has automatically suspended the scheduled automatic update. You need to manually sync your fork. Please refer to the detailed [Tutorial][tutorial-en-US] for instructions.
+
+            由于 LLM-Red-Team/kimi-free-api 上游仓库的 workflow 文件变更，导致 GitHub 自动暂停了本次自动更新，你需要手动 Sync Fork 一次，

README.md

@@ -11,7 +11,7 @@
 ![](https://img.shields.io/github/forks/llm-red-team/kimi-free-api.svg)
 ![](https://img.shields.io/docker/pulls/vinlic/kimi-free-api.svg)
 
-支持高速流式输出、支持多轮对话、支持联网搜索、支持长文档解读、支持图像解析，零配置部署，多路token支持，自动清理会话痕迹。
+支持高速流式输出、支持多轮对话、支持联网搜索、支持智能体对话、支持长文档解读、支持图像解析，零配置部署，多路token支持，自动清理会话痕迹。
 
 与ChatGPT接口完全兼容。
 
@@ -42,6 +42,7 @@
 * [Vercel部署](#Vercel部署)
 * [Zeabur部署](#Zeabur部署)
 * [原生部署](#原生部署)
+* [推荐使用客户端](#推荐使用客户端)
 * [接口列表](#接口列表)
   * [对话补全](#对话补全)
   * [文档解读](#文档解读)
@@ -84,6 +85,12 @@ https://udify.app/chat/Po0F6BMJ15q5vu2P
 
 ![联网搜索](./doc/example-2.png)
 
+### 智能体对话Demo
+
+此处使用 [翻译通](https://kimi.moonshot.cn/chat/coo6l3pkqq4ri39f36bg) 智能体。
+
+![智能体对话](./doc/example-7.png)
+
 ### 长文档解读Demo
 
 ![长文档解读](./doc/example-5.png)
@@ -241,6 +248,14 @@ pm2 reload kimi-free-api
 pm2 stop kimi-free-api
 ```
 
+## 推荐使用客户端
+
+使用以下二次开发客户端接入free-api系列项目更快更简单，支持文档/图像上传！
+
+由 [Clivia](https://github.com/Yanyutin753/lobe-chat) 二次开发的LobeChat [https://github.com/Yanyutin753/lobe-chat](https://github.com/Yanyutin753/lobe-chat)
+
+由 [时光@](https://github.com/SuYxh) 二次开发的ChatGPT Web [https://github.com/SuYxh/chatgpt-web-sea](https://github.com/SuYxh/chatgpt-web-sea)
+
 ## 接口列表
 
 目前支持与openai兼容的 `/v1/chat/completions` 接口，可自行使用与openai或其他兼容的客户端接入接口，或者使用 [dify](https://dify.ai/) 等线上服务接入使用。
@@ -260,8 +275,12 @@ Authorization: Bearer [refresh_token]
 请求数据：
 ```json
 {
-    // 模型名称随意填写，如果不希望输出检索过程模型名称请包含silent_search
+    // model随意填写，如果不希望输出检索过程模型名称请包含silent_search
+    // 如果使用kimi+智能体，model请填写智能体ID，就是浏览器地址栏上尾部的一串英文+数字20个字符的ID
     "model": "kimi",
+    // 目前多轮对话基于消息合并实现，某些场景可能导致能力下降且受单轮最大Token数限制
+    // 如果您想获得原生的多轮对话体验，可以传入首轮消息获得的id，来接续上下文，注意如果使用这个，首轮必须传none，否则第二轮会空响应！
+    // "conversation_id": "cnndivilnl96vah411dg",
     "messages": [
         {
             "role": "user",
@@ -278,6 +297,7 @@ Authorization: Bearer [refresh_token]
 响应数据：
 ```json
 {
+    // 如果想获得原生多轮对话体验，此id，你可以传入到下一轮对话的conversation_id来接续上下文
     "id": "cnndivilnl96vah411dg",
     "model": "kimi",
     "object": "chat.completion",

README_EN.md

@@ -20,7 +20,7 @@ Ali Tongyi (Qwen) API to API [qwen-free-api](https://github.com/LLM-Red-Team/qwe
 
 ZhipuAI (ChatGLM) API to API [glm-free-api](https://github.com/LLM-Red-Team/glm-free-api)
 
-MetaAI (metaso) API to API [metaso-free-api](https://github.com/LLM-Red-Team/metaso-free-api)
+Meta Sota (metaso) API to API [metaso-free-api](https://github.com/LLM-Red-Team/metaso-free-api)
 
 Iflytek Spark (Spark) API to API [spark-free-api](https://github.com/LLM-Red-Team/spark-free-api)
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "kimi-free-api",
-  "version": "0.0.30",
+  "version": "0.0.33",
   "description": "Kimi Free API Server",
   "type": "module",
   "main": "dist/index.js",

src/api/controllers/chat.ts

@@ -237,9 +237,10 @@ async function promptSnippetSubmit(query: string, refreshToken: string) {
  * @param messages 参考gpt系列消息格式，多轮对话请完整提供上下文
  * @param refreshToken 用于刷新access_token的refresh_token
  * @param useSearch 是否开启联网搜索
+ * @param refConvId 引用会话ID
  * @param retryCount 重试次数
  */
-async function createCompletion(model = MODEL_NAME, messages: any[], refreshToken: string, useSearch = true, retryCount = 0) {
+async function createCompletion(model = MODEL_NAME, messages: any[], refreshToken: string, useSearch = true, refConvId?: string, retryCount = 0) {
   return (async () => {
     logger.info(messages);
 
@@ -252,15 +253,16 @@ async function createCompletion(model = MODEL_NAME, messages: any[], refreshToke
       .catch(err => logger.error(err));
 
     // 创建会话
-    const convId = await createConversation("未命名会话", refreshToken);
+    const convId = /[0-9a-zA-Z]{20}/.test(refConvId) ? refConvId : await createConversation("未命名会话", refreshToken);
 
     // 请求流
     const {
       accessToken,
       userId
     } = await acquireToken(refreshToken);
-    const sendMessages = messagesPrepare(messages);
+    const sendMessages = messagesPrepare(messages, !!refConvId);
     const result = await axios.post(`https://kimi.moonshot.cn/api/chat/${convId}/completion/stream`, {
+      kimiplus_id: /^[0-9a-z]{20}$/.test(model) ? model : undefined,
       messages: sendMessages,
       refs,
       use_search: useSearch
@@ -268,6 +270,7 @@ async function createCompletion(model = MODEL_NAME, messages: any[], refreshToke
       headers: {
         Authorization: `Bearer ${accessToken}`,
         Referer: `https://kimi.moonshot.cn/chat/${convId}`,
+        'Priority': 'u=1, i',
         'X-Traffic-Id': userId,
         ...FAKE_HEADERS
       },
@@ -283,7 +286,8 @@ async function createCompletion(model = MODEL_NAME, messages: any[], refreshToke
     logger.success(`Stream has completed transfer ${util.timestamp() - streamStartTime}ms`);
 
     // 异步移除会话，如果消息不合规，此操作可能会抛出数据库错误异常，请忽略
-    removeConversation(convId, refreshToken)
+    // 如果引用会话将不会清除，因为我们不知道什么时候你会结束会话
+    !refConvId && removeConversation(convId, refreshToken)
       .catch(err => console.error(err));
     promptSnippetSubmit(sendMessages[0].content, refreshToken)
       .catch(err => console.error(err));
@@ -296,7 +300,7 @@ async function createCompletion(model = MODEL_NAME, messages: any[], refreshToke
         logger.warn(`Try again after ${RETRY_DELAY / 1000}s...`);
         return (async () => {
           await new Promise(resolve => setTimeout(resolve, RETRY_DELAY));
-          return createCompletion(model, messages, refreshToken, useSearch, retryCount + 1);
+          return createCompletion(model, messages, refreshToken, useSearch, refConvId, retryCount + 1);
         })();
       }
       throw err;
@@ -310,9 +314,10 @@ async function createCompletion(model = MODEL_NAME, messages: any[], refreshToke
  * @param messages 参考gpt系列消息格式，多轮对话请完整提供上下文
  * @param refreshToken 用于刷新access_token的refresh_token
  * @param useSearch 是否开启联网搜索
+ * @param refConvId 引用会话ID
  * @param retryCount 重试次数
  */
-async function createCompletionStream(model = MODEL_NAME, messages: any[], refreshToken: string, useSearch = true, retryCount = 0) {
+async function createCompletionStream(model = MODEL_NAME, messages: any[], refreshToken: string, useSearch = true, refConvId?: string, retryCount = 0) {
   return (async () => {
     logger.info(messages);
 
@@ -325,15 +330,16 @@ async function createCompletionStream(model = MODEL_NAME, messages: any[], refre
       .catch(err => logger.error(err));
 
     // 创建会话
-    const convId = await createConversation("未命名会话", refreshToken);
+    const convId = /[0-9a-zA-Z]{20}/.test(refConvId) ? refConvId : await createConversation("未命名会话", refreshToken);
 
     // 请求流
     const {
       accessToken,
       userId
     } = await acquireToken(refreshToken);
-    const sendMessages = messagesPrepare(messages);
+    const sendMessages = messagesPrepare(messages, !!refConvId);
     const result = await axios.post(`https://kimi.moonshot.cn/api/chat/${convId}/completion/stream`, {
+      kimiplus_id: /^[0-9a-z]{20}$/.test(model) ? model : undefined,
       messages: sendMessages,
       refs,
       use_search: useSearch
@@ -343,6 +349,7 @@ async function createCompletionStream(model = MODEL_NAME, messages: any[], refre
       headers: {
         Authorization: `Bearer ${accessToken}`,
         Referer: `https://kimi.moonshot.cn/chat/${convId}`,
+        'Priority': 'u=1, i',
         'X-Traffic-Id': userId,
         ...FAKE_HEADERS
       },
@@ -354,7 +361,8 @@ async function createCompletionStream(model = MODEL_NAME, messages: any[], refre
     return createTransStream(model, convId, result.data, () => {
       logger.success(`Stream has completed transfer ${util.timestamp() - streamStartTime}ms`);
       // 流传输结束后异步移除会话，如果消息不合规，此操作可能会抛出数据库错误异常，请忽略
-      removeConversation(convId, refreshToken)
+      // 如果引用会话将不会清除，因为我们不知道什么时候你会结束会话
+      !refConvId && removeConversation(convId, refreshToken)
         .catch(err => console.error(err));
       promptSnippetSubmit(sendMessages[0].content, refreshToken)
         .catch(err => console.error(err));
@@ -366,7 +374,7 @@ async function createCompletionStream(model = MODEL_NAME, messages: any[], refre
         logger.warn(`Try again after ${RETRY_DELAY / 1000}s...`);
         return (async () => {
           await new Promise(resolve => setTimeout(resolve, RETRY_DELAY));
-          return createCompletionStream(model, messages, refreshToken, useSearch, retryCount + 1);
+          return createCompletionStream(model, messages, refreshToken, useSearch, refConvId, retryCount + 1);
         })();
       }
       throw err;
@@ -447,14 +455,28 @@ function extractRefFileUrls(messages: any[]) {
  * user:新消息
  * 
  * @param messages 参考gpt系列消息格式，多轮对话请完整提供上下文
+ * @param isRefConv 是否为引用会话
  */
-function messagesPrepare(messages: any[]) {
-  // 注入消息提升注意力
-  let latestMessage = messages[messages.length - 1];
-  let hasFileOrImage = Array.isArray(latestMessage.content)
-    && latestMessage.content.some(v => (typeof v === 'object' && ['file', 'image_url'].includes(v['type'])));
-  // 第二轮开始注入system prompt
-  if (messages.length > 2) {
+function messagesPrepare(messages: any[], isRefConv = false) {
+  let content;
+  if (isRefConv || messages.length < 2) {
+    content = messages.reduce((content, message) => {
+      if (_.isArray(message.content)) {
+        return message.content.reduce((_content, v) => {
+          if (!_.isObject(v) || v['type'] != 'text') return _content;
+          return _content + `${v["text"] || ""}\n`;
+        }, content);
+      }
+      return content += `${message.role == 'user' ? wrapUrlsToTags(message.content) : message.content}\n`;
+    }, '')
+    logger.info("\n透传内容：\n" + content);
+  }
+  else {
+    // 注入消息提升注意力
+    let latestMessage = messages[messages.length - 1];
+    let hasFileOrImage = Array.isArray(latestMessage.content)
+      && latestMessage.content.some(v => (typeof v === 'object' && ['file', 'image_url'].includes(v['type'])));
+    // 第二轮开始注入system prompt
     if (hasFileOrImage) {
       let newFileMessage = {
         "content": "关注用户最新发送文件和消息",
@@ -470,18 +492,18 @@ function messagesPrepare(messages: any[]) {
       messages.splice(messages.length - 1, 0, newTextMessage);
       logger.info("注入提升尾部消息注意力system prompt");
     }
+    content = messages.reduce((content, message) => {
+      if (_.isArray(message.content)) {
+        return message.content.reduce((_content, v) => {
+          if (!_.isObject(v) || v['type'] != 'text') return _content;
+          return _content + `${message.role || "user"}:${v["text"] || ""}\n`;
+        }, content);
+      }
+      return content += `${message.role || "user"}:${message.role == 'user' ? wrapUrlsToTags(message.content) : message.content}\n`;
+    }, '')
+    logger.info("\n对话合并：\n" + content);
   }
 
-  const content = messages.reduce((content, message) => {
-    if (Array.isArray(message.content)) {
-      return message.content.reduce((_content, v) => {
-        if (!_.isObject(v) || v['type'] != 'text') return _content;
-        return _content + `${message.role || "user"}:${v["text"] || ""}\n`;
-      }, content);
-    }
-    return content += `${message.role || "user"}:${message.role == 'user' ? wrapUrlsToTags(message.content) : message.content}\n`;
-  }, '');
-  logger.info("\n对话合并：\n" + content);
   return [
     { role: 'user', content }
   ]
@@ -648,8 +670,8 @@ async function uploadFile(fileUrl: string, refreshToken: string) {
           ...FAKE_HEADERS
         }
       })
-      .then(() => resolve(true))
-      .catch(() => resolve(false));
+        .then(() => resolve(true))
+        .catch(() => resolve(false));
     });
   }
 

src/api/routes/chat.ts

@@ -13,22 +13,22 @@ export default {
 
         '/completions': async (request: Request) => {
             request
+                .validate('body.conversation_id', v => _.isUndefined(v) || _.isString(v))
                 .validate('body.messages', _.isArray)
                 .validate('headers.authorization', _.isString)
             // refresh_token切分
             const tokens = chat.tokenSplit(request.headers.authorization);
             // 随机挑选一个refresh_token
             const token = _.sample(tokens);
-            const model = request.body.model;
-            const messages =  request.body.messages;
-            if (request.body.stream) {
-                const stream = await chat.createCompletionStream(model, messages, token, request.body.use_search);
+            const { model, conversation_id: convId, messages, stream, use_search } = request.body;
+            if (stream) {
+                const stream = await chat.createCompletionStream(model, messages, token, use_search, convId);
                 return new Response(stream, {
                     type: "text/event-stream"
                 });
             }
             else
-                return await chat.createCompletion(model, messages, token, request.body.use_search);
+                return await chat.createCompletion(model, messages, token, use_search, convId);
         }
 
     }