diff --git a/beisi_rag/backend/__pycache__/backend.cpython-311.pyc b/beisi_rag/backend/__pycache__/backend.cpython-311.pyc
new file mode 100644
index 00000000..2340c624
Binary files /dev/null and b/beisi_rag/backend/__pycache__/backend.cpython-311.pyc differ
diff --git a/beisi_rag/backend/backend.py b/beisi_rag/backend/backend.py
new file mode 100644
index 00000000..d9a1c7b4
--- /dev/null
+++ b/beisi_rag/backend/backend.py
@@ -0,0 +1,170 @@
+from __future__ import annotations
+import os
+from pathlib import Path
+from typing import List, AsyncGenerator
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel, Field
+from dotenv import load_dotenv
+import dashscope
+from dashscope.aigc.generation import AioGeneration
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import DashScopeEmbeddings
+from langchain_core.documents import Document
+
+# 加载环境变量
+load_dotenv(dotenv_path=Path(__file__).resolve().parents[1] / "config" / ".env")
+
+# 初始化DashScope API
+DASHSCOPE_API_KEY = os.getenv("DASHSCOPE_API_KEY")
+BASE_URL = os.getenv("OPENAI_BASE_URL")
+MODEL_NAME = os.getenv("MODEL_NAME", "qwen-plus")
+TOP_K_DEFAULT = int(os.getenv("TOP_K", "4"))
+
+# 初始化 DashScope 生成客户端
+client = AioGeneration()
+
+# 加载 FAISS 向量数据库
+INDEX_DIR = Path(__file__).parent.parent / "vectordb"
+emb = DashScopeEmbeddings(dashscope_api_key=DASHSCOPE_API_KEY, model="text-embedding-v3")
+vectordb = FAISS.load_local(str(INDEX_DIR), emb, allow_dangerous_deserialization=True)
+
+# 格式化上下文
+def formatContext(docs: List[Document]) -> str:
+    parts = []
+    for i, d in enumerate(docs, 1):
+        src = (d.metadata or {}).get("source", "unknown")
+        txt = (d.page_content or "").replace("\n", " ")
+        if len(txt) > 500:
+            txt = txt[:500] + "…"
+        parts.append(f"[{i}] ({src}) {txt}")
+    return "\n".join(parts)
+
+# 同步检索
+def retrieve(question: str, k: int) -> List[Document]:
+    retriever = vectordb.as_retriever(search_type="similarity", search_kwargs={"k": k})
+    return retriever.invoke(question)
+
+SYSTEM_PROMPT = (
+    "你是严谨的中文检索增强助手。严格依据给定上下文回答；"
+    "若上下文没有答案，请明确说“不确定”，并给出你能确认的线索。"
+)
+
+# 构建用户提示
+def buildUserPrompt(question: str, context: str) -> str:
+    return (
+        "结合<已检索上下文>作答：\n"
+        f"<已检索上下文>\n{context}\n</已检索上下文>\n\n"
+        f"用户问题：{question}\n"
+        "要求：若答案不在上下文里，明确说明不确定；用中文、分点作答，必要时给出引用的原句摘要。"
+    )
+
+# FastAPI应用设置
+app = FastAPI(title="Qwen RAG QA Backend", version="1.0.0")
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# 请求模型
+class ChatReq(BaseModel):
+    question: str
+    top_k: int = Field(default=TOP_K_DEFAULT, ge=1, le=12)
+    model: str = Field(default=MODEL_NAME)
+    temperature: float = Field(default=0.3, ge=0, le=1)
+
+class ChatResp(BaseModel):
+    answer: str
+
+# 流式接口
+async def _sse_event(data: str) -> bytes:
+    print(f"发送流数据: {data}")  # 增加调试信息
+    return f"data: {data}\n\n".encode("utf-8")
+
+@app.post("/chat/stream")
+async def chat_stream(req: ChatReq):
+    try:
+        docs = retrieve(req.question, req.top_k)
+        context = formatContext(docs)
+        prompt = buildUserPrompt(req.question, context)
+
+        async def event_gen() -> AsyncGenerator[bytes, None]:
+            yield await _sse_event("{\"type\":\"meta\",\"message\":\"stream-start\"}")
+            
+            # 调试: 输出流的开始
+            print("流式处理开始...")
+
+            try:
+                # 使用 DashScope 生成流式回答
+                stream = dashscope.Generation.call(
+                    api_key=DASHSCOPE_API_KEY,
+                    model=req.model,
+                    messages=[{"role": "system", "content": SYSTEM_PROMPT},
+                              {"role": "user", "content": prompt}],
+                    result_format="message",
+                    enable_thinking=True,
+                    stream=True,
+                    incremental_output=True,
+                )
+
+                reasoning_content = ""
+                answer_content = ""
+                is_answering = False
+
+                # 逐步处理流数据
+                for chunk in stream:
+                    print(f"处理流数据: {chunk.output}")  # 增加调试信息
+
+                    if chunk.output.choices[0].message.content == "" and chunk.output.choices[0].message.reasoning_content == "":
+                        continue
+                    elif chunk.output.choices[0].message.reasoning_content != "" and chunk.output.choices[0].message.content == "":
+                        reasoning_content += chunk.output.choices[0].message.reasoning_content
+                        yield await _sse_event(chunk.output.choices[0].message.reasoning_content)
+                    elif chunk.output.choices[0].message.content != "":
+                        if not is_answering:
+                            yield await _sse_event("\n" + "=" * 20 + "完整回复" + "=" * 20)
+                            is_answering = True
+                        answer_content += chunk.output.choices[0].message.content
+                        yield await _sse_event(chunk.output.choices[0].message.content)
+
+                yield await _sse_event("[DONE]")
+            except Exception as ie:
+                # 错误事件
+                err = str(ie).replace("\n", " ")
+                yield await _sse_event(f"{{\"type\":\"error\",\"message\":\"{err}\"}}")
+
+        return StreamingResponse(event_gen(), media_type="text/event-stream")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+# 非流式接口
+@app.post("/chat", response_model=ChatResp)
+async def chat(req: ChatReq):
+    try:
+        docs = retrieve(req.question, req.top_k)
+        context = formatContext(docs)
+        prompt = buildUserPrompt(req.question, context)
+
+        response = await AioGeneration.call(
+            api_key=DASHSCOPE_API_KEY,
+            model=req.model,
+            messages=[{"role": "system", "content": SYSTEM_PROMPT},
+                      {"role": "user", "content": prompt}],
+            result_format="message",
+        )
+        answer = response.output.choices[0].message.content.strip()
+        return ChatResp(answer=answer)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+if __name__ == "__main__":
+    import uvicorn
+    HOST = os.getenv("HOST", "0.0.0.0")
+    PORT = int(os.getenv("PORT", "8000"))
+    uvicorn.run(app, host=HOST, port=PORT, reload=False)
diff --git a/beisi_rag/config/.env b/beisi_rag/config/.env
new file mode 100644
index 00000000..1e44a6dc
--- /dev/null
+++ b/beisi_rag/config/.env
@@ -0,0 +1,7 @@
+export DASHSCOPE_API_KEY="***************************************************"
+export OPENAI_API_KEY="$DASHSCOPE_API_KEY"
+export OPENAI_BASE_URL="https://dashscope-intl.aliyuncs.com/compatible-mode/v1"
+export MODEL_NAME="qwen-plus"
+export TOP_K="4"
+
+
diff --git a/beisi_rag/requirements.txt b/beisi_rag/requirements.txt
new file mode 100644
index 00000000..c6dfda1b
--- /dev/null
+++ b/beisi_rag/requirements.txt
@@ -0,0 +1,14 @@
+faiss-cpu==1.12.0
+importlib-metadata==8.0.0
+jaraco.collections==5.1.0
+langchain==1.0.3
+langchain-community==0.4.1
+langchain-openai==1.0.1
+pip-chill==1.0.3
+platformdirs==4.2.2
+tomli==2.0.1
+unicorn==2.1.4
+uvicorn==0.38.0
+fastapi
+dotenv
+dashscope
\ No newline at end of file
diff --git a/beisi_rag/src/chat_qwen_rag.py b/beisi_rag/src/chat_qwen_rag.py
new file mode 100644
index 00000000..d86c5a09
--- /dev/null
+++ b/beisi_rag/src/chat_qwen_rag.py
@@ -0,0 +1,161 @@
+# /Users/profighted/beisi-tech/docs/RAG-Anything/beisi_rag/chat_qwen_rag.py
+import os
+from pathlib import Path
+from dotenv import load_dotenv
+
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import DashScopeEmbeddings
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.runnables import RunnableParallel, RunnablePassthrough, RunnableLambda
+from langchain_core.messages import AIMessage, HumanMessage, SystemMessage, BaseMessage
+
+from openai import OpenAI
+
+load_dotenv(dotenv_path=Path(__file__).resolve().parents[1] / "config" / ".env")
+
+# 本地向量库目录（确保与 ingest 阶段一致）
+INDEX_DIR = Path(__file__).parent.parent / "vectordb"
+
+# ====== 配置 Qwen 兼容端点 ======
+DASHSCOPE_API_KEY = os.environ.get("DASHSCOPE_API_KEY")
+assert DASHSCOPE_API_KEY, "请先 export DASHSCOPE_API_KEY=你的通义 DashScope Key"
+
+# 如需新加坡地域，改成 https://dashscope-intl.aliyuncs.com/compatible-mode/v1
+DASHSCOPE_BASE_URL = os.environ.get(
+    "DASHSCOPE_BASE_URL",
+    "https://dashscope.aliyuncs.com/compatible-mode/v1",
+)
+
+# 对话模型 & 向量模型（按你账号可用情况调整）
+CHAT_MODEL = os.environ.get("QWEN_CHAT_MODEL", "qwen-plus")           # 也可用 qwen2.5-7b-instruct / qwen-turbo 等
+EMBEDDING_MODEL = os.environ.get("QWEN_EMBED_MODEL", "text-embedding-v3")
+
+
+def load_retriever():
+    """加载 FAISS 检索器（与 ingest 使用同一 Embedding 模型）"""
+    embeddings = DashScopeEmbeddings(
+        dashscope_api_key=DASHSCOPE_API_KEY,
+        model=EMBEDDING_MODEL,
+    )
+    vectordb = FAISS.load_local(str(INDEX_DIR), embeddings, allow_dangerous_deserialization=True)
+    return vectordb.as_retriever(search_type="similarity", search_kwargs={"k": 4})
+
+
+def build_llm_runnable():
+    """
+    用 OpenAI 兼容端点（Qwen）构建一个 LangChain Runnable。
+    避免使用 langchain_openai，降低版本依赖冲突风险。
+    """
+    client = OpenAI(api_key=DASHSCOPE_API_KEY, base_url=DASHSCOPE_BASE_URL)
+
+    def _lc_to_openai_messages(prompt_value) -> list[dict]:
+        """
+        将 LangChain 的 PromptValue / BaseMessage 列表转换为 OpenAI 兼容 messages。
+        """
+        if hasattr(prompt_value, "to_messages"):
+            msgs = prompt_value.to_messages()  # List[BaseMessage]
+        elif isinstance(prompt_value, list) and all(isinstance(m, BaseMessage) for m in prompt_value):
+            msgs = prompt_value
+        else:
+            # 兜底：当成用户单轮输入
+            msgs = [HumanMessage(content=str(prompt_value))]
+
+        out = []
+        for m in msgs:
+            if isinstance(m, SystemMessage):
+                role = "system"
+            elif isinstance(m, HumanMessage):
+                role = "user"
+            elif isinstance(m, AIMessage):
+                role = "assistant"
+            else:
+                role = "user"
+            out.append({"role": role, "content": m.content})
+        return out
+
+    def _invoke(prompt_value: BaseMessage | list[BaseMessage] | str) -> str:
+        messages = _lc_to_openai_messages(prompt_value)
+        resp = client.chat.completions.create(
+            model=CHAT_MODEL,
+            messages=messages,
+            temperature=0.3,
+        )
+        return resp.choices[0].message.content
+
+    return RunnableLambda(_invoke)
+
+
+# RAG_PROMPT = ChatPromptTemplate.from_template(
+#     """你是严谨的检索增强助手。结合<已检索上下文>回答用户问题。
+# - 如果答案不在上下文里，请明确说明“不确定”并给出你能确认的线索。
+# - 用中文输出，尽量给出引用的原句摘要，并在末尾标注引用编号（如 [1][3]）。
+
+# <已检索上下文>
+# {context}
+# </已检索上下文>
+
+# 用户问题：{question}
+# """
+# )
+
+RAG_PROMPT = ChatPromptTemplate.from_template(
+    """你是严谨的检索增强助手。结合<已检索上下文>回答用户问题。
+    -- 如果答案不在上下文里，请明确说明“不确定”并给出你能确认的线索。
+    -- 用中文输出，尽量给出引用的原句摘要，并在末尾标注引用编号（如 [1][3]）。
+        你是严谨的检索增强助手。请**用你自己的话**综合回答，禁止大段原文粘贴。
+    +规则：
+    +1) 先总结，再给出处；答案主体必须是**你自己的表述**。
+    +2) 如需引用原句，每处引用≤50字，并用引号与编号标注，如 “……”[1]。
+    +3) 如果上下文没有明确答案，请说“不确定”，并给出可验证的线索。
+    +4) 输出中文、结构化要点，并在末尾列出参考编号。
+    
+    <已检索上下文>
+    {context}
+    </已检索上下文>
+    
+    用户问题：{question}
+    """
+)
+
+
+
+def format_docs(docs):
+    out = []
+    for i, d in enumerate(docs, 1):
+        meta = d.metadata or {}
+        src = meta.get("source", "unknown")
+        # 摘要最多 500 字符，避免过长提示上下文
+        out.append(f"[{i}] ({src}) {d.page_content[:500]}")
+    return "\n\n".join(out)
+
+
+def main():
+    retriever = load_retriever()
+    llm_runnable = build_llm_runnable()
+
+    # RAG 链：检索 → 拼接上下文 → 提示词 → Qwen(兼容端点) → 解析文本
+    chain = (
+        RunnableParallel(context=retriever | format_docs, question=RunnablePassthrough())
+        | RAG_PROMPT
+        | llm_runnable
+        | StrOutputParser()
+    )
+
+    print("💬 输入你的问题（Ctrl+C 退出）")
+    while True:
+        try:
+            q = input("> ").strip()
+            if not q:
+                continue
+            ans = chain.invoke(q)
+            print("\n" + ans + "\n")
+        except (EOFError, KeyboardInterrupt):
+            print("\n再见～")
+            break
+        except Exception as e:
+            print("❌ 出错：", e)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/beisi_rag/src/ingest_qwen_faiss.py b/beisi_rag/src/ingest_qwen_faiss.py
new file mode 100644
index 00000000..f7919b16
--- /dev/null
+++ b/beisi_rag/src/ingest_qwen_faiss.py
@@ -0,0 +1,75 @@
+# /Users/profighted/beisi-tech/docs/beisi_rag/ingest_qwen_faiss.py
+import os
+from pathlib import Path
+from dotenv import load_dotenv
+
+from langchain_community.document_loaders import TextLoader, DirectoryLoader
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import DashScopeEmbeddings
+
+load_dotenv()
+
+DATA_DIR = Path(__file__).parent.parent / "my-website" / "docs"
+INDEX_DIR = Path(__file__).parent / "vectordb"
+INDEX_DIR.mkdir(exist_ok=True)
+
+MAX_BATCH = 10  # DashScope 限制：一次 <= 10
+
+def load_docs():
+    loader = DirectoryLoader(
+        str(DATA_DIR),
+        glob="**/*",
+        loader_cls=TextLoader,
+        show_progress=True
+    )
+    return loader.load()
+
+def chunked(iterable, n):
+    for i in range(0, len(iterable), n):
+        yield iterable[i:i + n]
+
+def main():
+    # 1) 加载与切分
+    docs = load_docs()
+    splitter = RecursiveCharacterTextSplitter(
+        chunk_size=800,
+        chunk_overlap=120,
+        separators=["\n\n", "\n", "。", "！", "？", "，", " ", ""]
+    )
+    splits = splitter.split_documents(docs)
+    texts = [d.page_content for d in splits]
+    metas = [d.metadata for d in splits]
+
+    # 2) Qwen（DashScope）嵌入
+    api_key = os.environ.get("DASHSCOPE_API_KEY")
+    if not api_key:
+        raise RuntimeError("请先 export DASHSCOPE_API_KEY=你的key")
+
+    embeddings = DashScopeEmbeddings(
+        dashscope_api_key=api_key,
+        model="text-embedding-v3",  # 按你账号可用模型名调整
+    )
+
+    # 3) 建立 FAISS，并以“<=10条/批”的方式喂入
+    if len(texts) == 0:
+        raise RuntimeError("没有可用文本分片，请检查 data/ 是否有可读文件。")
+
+    # 先用第一批初始化索引
+    first_texts = texts[:MAX_BATCH]
+    first_metas = metas[:MAX_BATCH]
+    store = FAISS.from_texts(first_texts, embeddings, metadatas=first_metas)
+
+    # 追加其余批次
+    for batch_texts, batch_metas in zip(
+        chunked(texts[MAX_BATCH:], MAX_BATCH),
+        chunked(metas[MAX_BATCH:], MAX_BATCH),
+    ):
+        store.add_texts(batch_texts, metadatas=batch_metas)
+
+    # 4) 保存索引
+    store.save_local(str(INDEX_DIR))
+    print(f"✅ 完成向量化，共 {len(texts)} 段，索引保存在: {INDEX_DIR}")
+
+if __name__ == "__main__":
+    main()
diff --git a/beisi_rag/src/test.py b/beisi_rag/src/test.py
new file mode 100644
index 00000000..7d306061
--- /dev/null
+++ b/beisi_rag/src/test.py
@@ -0,0 +1,25 @@
+import os
+from openai import OpenAI
+from dotenv import load_dotenv
+load_dotenv()
+
+try:
+    client = OpenAI(
+        # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key="sk-xxx",
+        # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key
+        api_key=os.getenv("DASHSCOPE_API_KEY"),
+        # 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://dashscope-intl.aliyuncs.com/compatible-mode/v1
+        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+    )
+
+    completion = client.chat.completions.create(
+        model="qwen-plus",  # 模型列表：https://help.aliyun.com/zh/model-studio/getting-started/models
+        messages=[
+            {'role': 'system', 'content': 'You are a helpful assistant.'},
+            {'role': 'user', 'content': '你是谁？'}
+            ]
+    )
+    print(completion.choices[0].message.content)
+except Exception as e:
+    print(f"错误信息：{e}")
+    print("请参考文档：https://help.aliyun.com/zh/model-studio/developer-reference/error-code")
\ No newline at end of file
diff --git a/beisi_rag/vectordb/index.faiss b/beisi_rag/vectordb/index.faiss
new file mode 100644
index 00000000..61e97ef1
Binary files /dev/null and b/beisi_rag/vectordb/index.faiss differ
diff --git a/beisi_rag/vectordb/index.pkl b/beisi_rag/vectordb/index.pkl
new file mode 100644
index 00000000..ae08d1a5
Binary files /dev/null and b/beisi_rag/vectordb/index.pkl differ
diff --git a/my-website/src/components/ChatWidget/chatwidget.module.css b/my-website/src/components/ChatWidget/chatwidget.module.css
new file mode 100644
index 00000000..74e328d3
--- /dev/null
+++ b/my-website/src/components/ChatWidget/chatwidget.module.css
@@ -0,0 +1,36 @@
+.fab{
+  position: fixed; right: 20px; bottom: 20px;
+  width: 56px; height: 56px; border-radius: 50%;
+  border: none; background: #3b82f6; color:#fff; font-weight:700;
+  box-shadow: 0 8px 20px rgba(0,0,0,.15); cursor: pointer; z-index: 9999;
+}
+.panel{
+  position: fixed; right: 20px; bottom: 90px;
+  width: 340px; height: 460px; background:#fff; border-radius: 12px;
+  box-shadow: 0 16px 40px rgba(0,0,0,.2); display:flex; flex-direction:column;
+  overflow:hidden; z-index: 9999;
+}
+.header{
+  padding:12px 14px; font-weight:700; background:#f8fafc;
+  border-bottom:1px solid #e5e7eb; display:flex; justify-content:space-between; align-items:center;
+}
+.switch{ font-size:12px; display:flex; gap:6px; align-items:center; }
+.messages{ flex:1; padding:12px; overflow:auto; background:#fafafa; }
+.msg{ display:flex; margin-bottom:8px; }
+.user{ justify-content:flex-end; }
+.assistant{ justify-content:flex-start; }
+.bubble{
+  max-width: 78%; padding:8px 10px; border-radius:10px; line-height:1.4;
+  background:#e5e7eb; color:#111827; white-space:pre-wrap;
+}
+.user .bubble{ background:#3b82f6; color:#fff; }
+.inputBar{
+  display:flex; gap:8px; padding:10px; border-top:1px solid #e5e7eb; background:#fff;
+}
+.inputBar input{
+  flex:1; height:36px; padding:6px 10px; border:1px solid #d1d5db; border-radius:8px; outline:none;
+}
+.inputBar button{
+  height:36px; padding:0 12px; border:none; border-radius:8px;
+  background:#111827; color:#fff; cursor:pointer;
+}
diff --git a/my-website/src/components/ChatWidget/chatwidget.tsx b/my-website/src/components/ChatWidget/chatwidget.tsx
new file mode 100644
index 00000000..de55e4a2
--- /dev/null
+++ b/my-website/src/components/ChatWidget/chatwidget.tsx
@@ -0,0 +1,168 @@
+import React, {useMemo, useRef, useState} from 'react';
+import useDocusaurusContext from '@docusaurus/useDocusaurusContext';
+import styles from './chatwidget.module.css';
+
+type Role = 'user' | 'assistant';
+interface Msg { role: Role; text: string; }
+
+const CHAT_ID = 'default';
+
+export default function ChatWidget() {
+  const {siteConfig} = useDocusaurusContext();
+  const API_BASE =
+    (siteConfig as any)?.customFields?.CHAT_API_BASE ?? 'http://115.190.200.222:8000';
+
+  const [open, setOpen] = useState(false);
+  const [msgs, setMsgs] = useState<Msg[]>([
+    {role: 'assistant', text: '嗨，我是你的智能助手～'},
+  ]);
+  const [userInput, setUserInput] = useState('');
+  const [loading, setLoading] = useState(false);
+  const [useStream, setUseStream] = useState(false);
+  const listRef = useRef<HTMLDivElement>(null);
+
+  const chatUrl = useMemo(() => `${API_BASE}/chat`, [API_BASE]);
+  const streamUrl = useMemo(() => `${API_BASE}/chat/stream`, [API_BASE]);
+
+  const scrollToBottom = () => {
+    requestAnimationFrame(() => {
+      const el = listRef.current;
+      if (el) el.scrollTop = el.scrollHeight;
+    });
+  };
+
+  const send = async () => {
+    const q = userInput.trim();
+    if (!q || loading) return;
+    setMsgs((m) => [...m, {role: 'user', text: q}]);
+    setUserInput('');
+    setLoading(true);
+
+    try {
+      if (!useStream || !('ReadableStream' in globalThis)) {
+        // --- 非流式 ---
+        const res = await fetch(chatUrl, {
+          method: 'POST',
+          headers: {'Content-Type': 'application/json'},
+          body: JSON.stringify({
+            question: q,
+            top_k: 4,
+            model: 'qwen-plus',
+            temperature: 0.3,
+          }),
+          // credentials: 'include',
+        });
+        if (!res.ok) throw new Error(`${res.status} ${res.statusText}`);
+        const data = await res.json();
+        const answer = data?.answer ?? data?.data?.answer ?? '';
+        setMsgs((m) => [...m, {role: 'assistant', text: String(answer)}]);
+      } else {
+        // 流式
+        const res = await fetch(streamUrl, {
+          method: 'POST',
+          headers: {'Content-Type': 'application/json'},
+          body: JSON.stringify({
+            question: q,
+            top_k: 4,
+            model: 'qwen-plus',
+            temperature: 0.3,
+          }),
+          // credentials: 'include',
+        });
+        if (!res.ok || !res.body) throw new Error('SSE 不可用');
+
+        const reader = res.body.getReader();
+        const decoder = new TextDecoder('utf-8');
+        let buf = '';
+
+        while (true) {
+          const {done, value} = await reader.read();
+          if (done) break;
+          buf += decoder.decode(value, {stream: true});
+
+          // 按 SSE 块拆分：每块以 \n\n 结束，且行以 "data: " 开头
+          const chunks = buf.split('\n\n');
+          buf = chunks.pop() ?? '';
+
+          for (const chunk of chunks) {
+            if (!chunk.startsWith('data:')) continue;
+            const payload = chunk.replace(/^data:\s*/, '').trim();
+            if (!payload) continue;        // 最后一条空消息
+            if (payload === '[DONE]') continue;
+
+            let obj: any = null;
+            try { obj = JSON.parse(payload); } catch { /* 纯文本 */ }
+
+            if (obj?.type === 'meta') {
+              // 开始事件，忽略
+            } else if (obj?.type === 'reasoning') {
+              setMsgs((m) => [...m, {role: 'assistant', text: `思考：${obj.text || ''}`}]);
+            } else if (obj?.type === 'sep') {
+              setMsgs((m) => [...m, {role: 'assistant', text: '———'}]);
+            } else if (obj?.type === 'answer') {
+              setMsgs((m) => [...m, {role: 'assistant', text: obj.text || ''}]);
+            } else {
+              // 兼容纯文本
+              setMsgs((m) => [...m, {role: 'assistant', text: String(obj ?? payload)}]);
+            }
+            scrollToBottom();
+          }
+        }
+      }
+    } catch (e: any) {
+      setMsgs((m) => [...m, {role: 'assistant', text: `❗请求失败：${e?.message || e}`}]);
+    } finally {
+      setLoading(false);
+      scrollToBottom();
+    }
+  };
+
+  return (
+    <>
+      <button className={styles.fab} onClick={() => setOpen((o) => !o)}>
+        {open ? '×' : 'Chat'}
+      </button>
+
+      {open && (
+        <div className={styles.panel}>
+          <div className={styles.header}>
+            聊天助手
+            <label className={styles.switch}>
+              <input
+                type="checkbox"
+                checked={useStream}
+                onChange={(e) => setUseStream(e.target.checked)}
+              />
+              <span>流式</span>
+            </label>
+          </div>
+
+          <div className={styles.messages} ref={listRef}>
+            {msgs.map((m, i) => (
+              <div key={i} className={`${styles.msg} ${styles[m.role]}`}>
+                <div className={styles.bubble}>{m.text}</div>
+              </div>
+            ))}
+            {loading && (
+              <div className={`${styles.msg} ${styles.assistant}`}>
+                <div className={styles.bubble}>思考中…</div>
+              </div>
+            )}
+          </div>
+
+          <div className={styles.inputBar}>
+            <input
+              value={userInput}
+              onChange={(e) => setUserInput(e.target.value)}
+              onKeyDown={(e) => (e.key === 'Enter' ? send() : null)}
+              placeholder="输入问题并回车"
+            />
+            <button onClick={send} disabled={loading || !userInput.trim()}>
+              发送
+            </button>
+          </div>
+        </div>
+      )}
+    </>
+  );
+}
diff --git a/my-website/src/theme/Root.tsx b/my-website/src/theme/Root.tsx
new file mode 100644
index 00000000..a74c0ed1
--- /dev/null
+++ b/my-website/src/theme/Root.tsx
@@ -0,0 +1,15 @@
+import React from 'react';
+import {useLocation} from '@docusaurus/router';
+import ChatWidget from '../components/ChatWidget/chatwidget';
+
+export default function Root({children}) {
+  const location = useLocation();
+  const isHome = location.pathname === '/' || location.pathname.endsWith('/index.html');
+
+  return (
+    <>
+      {children}
+      {!isHome && <ChatWidget />}
+    </>
+  );
+}