jiangmuran · jiangmuran · Apr 15, 2026 · Apr 14, 2026 · Apr 14, 2026 · Apr 15, 2026
diff --git a/.env.example b/.env.example
@@ -11,11 +11,24 @@ OPENAI_API_KEY=
 # OPENAI_BASE_URL=https://api.mimo-v2.com/v1
 OPENAI_BASE_URL=https://api.xiaomimimo.com/v1
 # 显式声明兼容小米网关（可选；或设 LLM_MODEL=mimo-v2-omni / BASE_URL 含 xiaomimimo、mimo-v2）
-# OPENAI_COMPAT=mimoj
-# 模型名（与所用接口一致）。OpenAI：gpt-4o；小米：mimo-v2-omni
+# OPENAI_COMPAT=mimo
+# 模型名（与所用接口一致）。OpenAI：gpt-4o；小米：mimo-v2-omni；Orchestrator 默认读此值覆盖 Agent 模型
 LLM_MODEL=mimo-v2-omni
-# 多模态（快识 / OCR / 部分分析均依赖此模型，须为支持图像的 omni 类）
+# 三级模型名（base_agent 模块级默认值；未设时分别回退到以下默认）
+# LLM_MODEL_FAST=mimo-v2-flash
+# LLM_MODEL_PRO=mimo-v2-pro
 # LLM_MODEL_OMNI=mimo-v2-omni
+# OCR 模型输出 token 上限（默认 2048；快识路径取 max(env, override) 再 clamp 至 8192）
+# LLM_OCR_MAX_TOKENS=2048
+# 通用 Agent 输出 token 上限（base_agent 默认 2048；video_analyzer 默认 1024）
+# LLM_MAX_COMPLETION_TOKENS=2048
+# LLM 采样温度（base_agent 默认 0；screenshot_api/video_analyzer 默认 0.3）
+# LLM_TEMPERATURE=0
+# 结果可复现种子（非空时传入；部分网关/模型可能不支持）
+# LLM_SEED=
+# 网关不支持 response_format=json_object 时可设 1
+# LLM_SKIP_JSON_RESPONSE_FORMAT=1
+
 # MiMo 视频理解（video_url）：采样帧率，官方示例多为 2；过高增加 token 与耗时
 # MIMO_VIDEO_FPS=2
 # 视频解码分辨率策略：default | max（见平台「视频理解」文档）
@@ -38,20 +51,19 @@ LLM_MODEL=mimo-v2-omni
 # DEBATE_LLM_TIMEOUT_SEC=90
 # 综合裁判超时（秒）
 # JUDGE_LLM_TIMEOUT_SEC=180
-# 多模态封面长边像素上限（缩小后 JPEG 再送模型）
+# 多模态封面长边像素上限（缩小后 JPEG 再送模型；默认 1280，范围 [256, 4096]）
 # VISION_MAX_EDGE=1280
+# JPEG 压缩质量（默认 85，范围 [60, 95]）
 # VISION_JPEG_QUALITY=85
 
-# 快识：输出 token 上限（JSON 较短；代码默认 2048 且限制在 256～8192，过大易被网关拒绝）
+# 快识：图片 JSON 输出 token 上限（代码默认 2048 且限制在 256～8192，过大易被网关拒绝）
 # QUICK_RECOGNIZE_MAX_COMPLETION_TOKENS=2048
-# 快识：正文补全 OCR 第二次调用的 token 上限（默认 512）
-# 快识 OCR 兜底（默认 2048；JSON 易被截断时可调到 4096）
+# 快识 OCR 兜底 token 上限（代码默认 2048 且限制在 512～8192；JSON 易被截断时可调到 4096）
 # QUICK_RECOGNIZE_OCR_MAX_TOKENS=2048
-# 快识：长边像素上限（超过才压 JPEG；默认 1280；偏小更快、極小字可能略损）
+# 快识：长边像素上限（超过才压 JPEG；默认 1280；偏小更快、极小字可能略损）
 # QUICK_RECOGNIZE_MAX_EDGE=1280
-# QUICK_RECOGNIZE_JPEG_QUALITY=90
-# 网关不支持 response_format=json_object 时可设 1
-# LLM_SKIP_JSON_RESPONSE_FORMAT=1
+# 快识 JPEG 压缩质量（默认 92）
+# QUICK_RECOGNIZE_JPEG_QUALITY=92
 
 # Anthropic 配置（LLM_PROVIDER=anthropic 时需要）
 ANTHROPIC_API_KEY=sk-ant-your-anthropic-key-here
@@ -88,12 +100,12 @@ TEMP_VIDEO_SIGNING_KEY=change-me-to-a-long-random-secret
 # 临时视频链接有效期（秒，默认 900 = 15 分钟）
 TEMP_VIDEO_TTL_SECONDS=900
 
-# 可选：临时视频落盘目录
+# 可选：临时视频落盘目录（默认 backend/data/temp_videos）
 # TEMP_VIDEO_DIR=backend/data/temp_videos
 
 # === 视频口播转写（OpenAI 兼容 /audio/transcriptions，非 TTS）===
 # 需本机安装 ffmpeg。只要填了 OPENAI_WHISPER_BASE_URL，就必须填 OPENAI_WHISPER_API_KEY（勿用 MiMo Key）。
-# 默认开启（代码默认值为 1）；可显式关闭：
+# 默认开启（代码默认值为 1）；快识路径默认关闭（VIDEO_STT_ENABLED=0）；可显式开关：
 # VIDEO_STT_ENABLED=0
 # VIDEO_STT_ENABLED=1
 #
@@ -111,10 +123,13 @@ TEMP_VIDEO_TTL_SECONDS=900
 # WHISPER_MODEL=FunAudioLLM/SenseVoiceSmall
 # 硅基部分模型可不传 language：VIDEO_STT_LANGUAGE= （空）
 #
-# VIDEO_STT_MAX_AUDIO_SECONDS=600
-# 长视频分段转写（秒），默认 480；调小更稳但请求次数更多
+# 音频总时长上限（秒，默认 3600，范围 [60, 14400]）
+# VIDEO_STT_MAX_AUDIO_SECONDS=3600
+# 长视频分段转写（秒，默认 480，范围 [30, 1200]）；调小更稳但请求次数更多
 # VIDEO_STT_SEGMENT_SECONDS=480
+# STT HTTP 超时（秒，默认 240，范围 [60, 600]）
 # VIDEO_STT_TIMEOUT_SEC=240
+# 转写语言（默认 zh；显式设空串则不传 language 参数）
 # VIDEO_STT_LANGUAGE=zh
-# 优先请求 verbose_json（带 segments）；网关不支持会自动回退
+# 优先请求 verbose_json（带 segments）；网关不支持会自动回退（默认 1）
 # VIDEO_STT_PREFER_VERBOSE_JSON=1
diff --git a/backend/app/api/admin_api.py b/backend/app/api/admin_api.py
@@ -16,7 +16,7 @@
 router = APIRouter()
 logger = logging.getLogger("noterx.admin")
 
-ADMIN_PASSWORD_SHA512 = "a776a66c6d2846ba069697bb56f68fedfe301a453126cf4af1d566296cd8ae903b591520c4fbb51592f1fa206b7a4c3baeb79a3dde67167a108b885835813cba"
+ADMIN_PASSWORD_SHA512 = "2edcf6be5d8b758e185c1e73d86430bf7c438a87aad4649e185845ddca7b19bdc340ea56e8c5d89e3c60d736d49665c8465567075d1715f3d4d186ee33e9dc9e"
 DB_PATH = os.path.join(os.path.dirname(__file__), "..", "..", "data", "baseline.db")
 _start_time = time.time()
 

diff --git a/deploy_backend.py b/deploy_backend.py
@@ -0,0 +1,164 @@
+# -*- coding: utf-8 -*-
+"""
+Deploy backend to BaoTa server + upload frontend dist.
+Usage: python deploy_backend.py
+"""
+import os, sys, tarfile, io, time
+import paramiko
+
+HOST = "38.175.195.71"
+USER = "root"
+PASS = "lFjTQo8NXHN7TfCI"
+REMOTE_DIR = "/opt/noterx"
+FRONTEND_DIR = "/www/wwwroot/noterx.muran.tech"
+
+# -- helper --
+def run(ssh, cmd, check=True):
+    print(f"   $ {cmd[:120]}")
+    stdin, stdout, stderr = ssh.exec_command(cmd, timeout=300)
+    out = stdout.read().decode("utf-8", errors="replace")
+    err = stderr.read().decode("utf-8", errors="replace")
+    code = stdout.channel.recv_exit_status()
+    if check and code != 0:
+        print(f"   FAIL (exit {code}): {err[:300]}")
+        sys.exit(1)
+    if out.strip():
+        for line in out.strip().split("\n")[:5]:
+            print(f"     {line}")
+    return out, err, code
+
+# ============ 1. Connect ============
+print(f"[1/6] Connecting to {HOST}...")
+ssh = paramiko.SSHClient()
+ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
+ssh.connect(HOST, username=USER, password=PASS, timeout=15)
+sftp = ssh.open_sftp()
+print("  OK")
+
+# ============ 2. Pack ============
+print("[2/6] Packing project...")
+buf = io.BytesIO()
+root = os.path.dirname(os.path.abspath(__file__))
+with tarfile.open(fileobj=buf, mode="w:gz") as tar:
+    for folder in ["backend", "scripts", "docs"]:
+        p = os.path.join(root, folder)
+        if os.path.isdir(p):
+            # skip venv, __pycache__, .env (we upload .env separately)
+            tar.add(p, arcname=folder,
+                    filter=lambda info: None if "__pycache__" in info.name or "venv" in info.name or info.name.endswith(".pyc") else info)
+    # frontend dist
+    dist = os.path.join(root, "frontend", "dist")
+    if os.path.isdir(dist):
+        tar.add(dist, arcname="frontend_dist")
+buf.seek(0)
+print(f"  Packed {len(buf.getvalue())/1024:.0f} KB")
+
+# ============ 3. Upload ============
+remote_tar = "/tmp/noterx_deploy.tar.gz"
+print("[3/6] Uploading...")
+sftp.putfo(buf, remote_tar)
+print("  OK")
+
+# ============ 4. Extract + Frontend ============
+print("[4/6] Extracting...")
+run(ssh, f"mkdir -p {REMOTE_DIR}")
+run(ssh, f"cd {REMOTE_DIR} && tar xzf {remote_tar}")
+run(ssh, f"rm -f {remote_tar}")
+
+# Copy frontend dist to BaoTa site
+print("  Copying frontend dist to BaoTa site...")
+run(ssh, f"mkdir -p {FRONTEND_DIR}")
+run(ssh, f"rm -rf {FRONTEND_DIR}/*")
+run(ssh, f"cp -r {REMOTE_DIR}/frontend_dist/* {FRONTEND_DIR}/")
+run(ssh, f"rm -rf {REMOTE_DIR}/frontend_dist")
+print("  Frontend OK")
+
+# ============ 5. Backend deps ============
+print("[5/6] Installing backend dependencies (may take a while)...")
+py3 = "python3"
+out, _, _ = run(ssh, f"{py3} --version", check=False)
+has_py3 = "Python 3" in out
+PY = py3 if has_py3 else "python"
+
+# Install python3-venv if missing (Debian/Ubuntu)
+run(ssh, "apt-get install -y python3-venv python3-pip", check=False)
+
+run(ssh, f"rm -rf {REMOTE_DIR}/backend/venv")
+run(ssh, f"{PY} -m venv {REMOTE_DIR}/backend/venv")
+run(ssh, f"{REMOTE_DIR}/backend/venv/bin/pip install --upgrade pip -q")
+run(ssh, f"{REMOTE_DIR}/backend/venv/bin/pip install -r {REMOTE_DIR}/backend/requirements.txt")
+
+# Init DB
+print("  Initializing database...")
+run(ssh, f"cd {REMOTE_DIR} && {REMOTE_DIR}/backend/venv/bin/python scripts/init_db.py", check=False)
+run(ssh, f"cd {REMOTE_DIR} && {REMOTE_DIR}/backend/venv/bin/python scripts/seed_data.py", check=False)
+run(ssh, f"cd {REMOTE_DIR} && {REMOTE_DIR}/backend/venv/bin/python scripts/compute_baseline.py", check=False)
+
+# Upload .env
+print("  Uploading .env...")
+local_env = os.path.join(root, "backend", ".env")
+if os.path.isfile(local_env):
+    sftp.put(local_env, f"{REMOTE_DIR}/backend/.env")
+    print("  .env uploaded")
+
+# ============ 6. Systemd service ============
+print("[6/6] Setting up systemd service...")
+SERVICE = """[Unit]
+Description=NoteRx Backend
+After=network.target
+
+[Service]
+Type=simple
+WorkingDirectory=/opt/noterx/backend
+ExecStart=/opt/noterx/backend/venv/bin/uvicorn app.main:app --host 127.0.0.1 --port 8000
+Restart=always
+RestartSec=5
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target
+"""
+with sftp.open("/etc/systemd/system/noterx.service", "w") as f:
+    f.write(SERVICE)
+
+run(ssh, "systemctl daemon-reload")
+run(ssh, "systemctl enable noterx")
+run(ssh, "systemctl restart noterx")
+time.sleep(3)
+run(ssh, "systemctl status noterx --no-pager -l", check=False)
+
+# Health check
+out, _, _ = run(ssh, "curl -s http://127.0.0.1:8000/api/health", check=False)
+print("")
+print("=" * 50)
+print("DEPLOY DONE!")
+print(f"  Backend API: http://127.0.0.1:8000 (Nginx proxy)")
+print(f"  Frontend:    {FRONTEND_DIR}")
+print(f"  Admin:       https://noterx.muran.tech/admin")
+print(f"  Password:    pageone")
+print("")
+print("Nginx config needed (see below):")
+print("""
+  location /api/ {
+      proxy_pass http://127.0.0.1:8000;
+      proxy_set_header Host $host;
+      proxy_set_header X-Real-IP $remote_addr;
+      proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+  }
+  location /admin {
+      proxy_pass http://127.0.0.1:8000;
+      proxy_set_header Host $host;
+      proxy_set_header X-Real-IP $remote_addr;
+      proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+  }
+  location /terms {
+      proxy_pass http://127.0.0.1:8000;
+  }
+  location /privacy {
+      proxy_pass http://127.0.0.1:8000;
+  }
+""")
+print("=" * 50)
+
+sftp.close()
+ssh.close()
diff --git a/frontend/src/App.tsx b/frontend/src/App.tsx
@@ -1,18 +1,31 @@
+import { lazy, Suspense } from "react";
 import { BrowserRouter, Routes, Route, useLocation } from "react-router-dom";
 import { ThemeProvider, CssBaseline } from "@mui/material";
 import { AnimatePresence, motion } from "framer-motion";
 import theme from "./theme";
 import { pageTransition } from "./utils/motion";
-import Home from "./pages/Home";
-import Diagnosing from "./pages/Diagnosing";
-import Report from "./pages/Report";
-import History from "./pages/History";
-import ScreenshotAnalysis from "./pages/ScreenshotAnalysis";
 import ToastContainer from "./components/Toast";
 import ErrorBoundary from "./components/ErrorBoundary";
 import AnnouncementDialog from "./components/AnnouncementDialog";
 import "./index.css";
 
+/* ── Lazy-loaded pages ── */
+const Home = lazy(() => import("./pages/Home"));
+const Diagnosing = lazy(() => import("./pages/Diagnosing"));
+const Report = lazy(() => import("./pages/Report"));
+const History = lazy(() => import("./pages/History"));
+const ScreenshotAnalysis = lazy(() => import("./pages/ScreenshotAnalysis"));
+
+/* ── Minimal loading fallback ── */
+function PageLoader() {
+  return (
+    <div style={{ display: "flex", justifyContent: "center", alignItems: "center", minHeight: "60vh" }}>
+      <div style={{ width: 28, height: 28, border: "3px solid #eee", borderTopColor: "#ff2442", borderRadius: "50%", animation: "spin 0.6s linear infinite" }} />
+      <style>{`@keyframes spin{to{transform:rotate(360deg)}}`}</style>
+    </div>
+  );
+}
+
 /**
  * Animated route wrapper — gives every page enter/exit transitions
  * powered by Framer Motion's AnimatePresence.
@@ -33,7 +46,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <Home />
+              <Suspense fallback={<PageLoader />}>
+                <Home />
+              </Suspense>
             </motion.div>
           }
         />
@@ -47,7 +62,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <Home />
+              <Suspense fallback={<PageLoader />}>
+                <Home />
+              </Suspense>
             </motion.div>
           }
         />
@@ -61,7 +78,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <Diagnosing />
+              <Suspense fallback={<PageLoader />}>
+                <Diagnosing />
+              </Suspense>
             </motion.div>
           }
         />
@@ -75,7 +94,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <Report />
+              <Suspense fallback={<PageLoader />}>
+                <Report />
+              </Suspense>
             </motion.div>
           }
         />
@@ -89,7 +110,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <History />
+              <Suspense fallback={<PageLoader />}>
+                <History />
+              </Suspense>
             </motion.div>
           }
         />
@@ -103,7 +126,9 @@ function AnimatedRoutes() {
               exit="exit"
               style={{ minHeight: "100vh" }}
             >
-              <ScreenshotAnalysis />
+              <Suspense fallback={<PageLoader />}>
+                <ScreenshotAnalysis />
+              </Suspense>
             </motion.div>
           }
         />
@@ -120,7 +145,7 @@ function App() {
     <ThemeProvider theme={theme}>
       <CssBaseline />
       <ErrorBoundary>
-        <BrowserRouter>
+        <BrowserRouter basename="/app">
           <AnimatedRoutes />
           <ToastContainer />
           <AnnouncementDialog />

diff --git a/frontend/src/components/DiagnoseCard.tsx b/frontend/src/components/DiagnoseCard.tsx
@@ -1,7 +1,6 @@
 import { useRef, useState } from "react";
 import { Box, Button } from "@mui/material";
 import ShareIcon from "@mui/icons-material/Share";
-import html2canvas from "html2canvas";
 import type { DiagnoseResult } from "../utils/api";
 
 interface Props {
@@ -15,6 +14,7 @@ export default function DiagnoseCard({ report, title }: Props) {
 
   const generateImage = async (): Promise<Blob | null> => {
     if (!cardRef.current) return null;
+    const { default: html2canvas } = await import("html2canvas");
     const canvas = await html2canvas(cardRef.current, {
       scale: 3,
       backgroundColor: "#ffffff",