feat: add video OCR/ASR, DOCX/PDF export, input block and risk config updates

- Video pipeline: video file OCR via VLM plus audio track ASR, integrated
  into job_handlers with progress emit per phase. New media_utils.py for
  audio extraction from video files.
- Document export: richExport.js replaces inline docx builder; DOCX and PDF
  export buttons are now enabled in MilkdownEditor. File size limit raised to
  100 MB.
- Input block: new InputBlockCrepe.vue component with inputBlockPlugin.ts
  and inputBlock.js for custom user-input nodes in the editor.
- Risk config: added Vite dev server ports (5173) to CORS allowlist and
  increased OCR max input from 10 MB to 100 MB.
- TTS/ASR refactor: simplified tts_asr.py model loading and warmup logic.
- Test coverage: updated tests for llm, main endpoints, pro completions and
  web search modules.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

AGENTS.md

@@ -24,7 +24,7 @@
 | 上传块 | `upload_block` | `{{{}}}` / `{{{upload file type:...}}}` | `plugins/uploadBlockPlugin.ts` | `utils/uploadBlock.js` |
 
 - **已验证**：当前代码完全符合"禁止嵌套、自动解析复原"的要求。修改功能块相关逻辑时需验证：1) schema 的 atom/isolating 属性不被移除；2) Remark/parseMarkdown/toMarkdown 解析链路完整。
-- 当前代码中可以确认的主功能是：AI 补全、OCR、文档转 Markdown、TTS、Markdown/DOCX/PDF 导入导出。
+- 当前代码中可以确认的主功能是：AI 补全、OCR、文档转 Markdown、TTS/ASR、Markdown/DOCX/PDF 导入导出。
 - 历史文档中有一部分 TTS/ASR、Apple Silicon、Whisper、离线模式说明已经落后于当前代码；出现冲突时以实际代码和测试为准。
 
 ## 先看哪里
@@ -80,8 +80,13 @@
 - 文档超过 32 KB 时，AI 补全会在前端和插件层被禁用。
 - OCR 文本和文档块内容会被注入补全上下文，但这些内容属于隐藏上下文，不应被直接当作用户可见文本重复输出。
 - /v1/convert 当前支持 txt、docx、pptx、pdf，非 txt 文件通过 MarkItDown 转成 Markdown，之后会清理图片标记。
-- 前端存在 /v1/export/pdf 调用点，但当前后端主路由中看不到同名端点；排查 PDF 导出问题前先确认服务端是否真正提供该接口。
-- 当前 tts_asr.py 主要提供 TTS 相关能力。不要直接沿用 README 或历史修复文档里关于 ASR、Whisper、MPS/offline 的描述。
+- **AI 开关是全局广播状态**：MilkdownEditor.vue 通过 `llm-in-text:copilot-toggle` 同步主编辑器、文档块嵌套编辑器、网页搜索块嵌套编辑器；修 ghost text 时要同时检查这三处。
+- **设置项已从 currency 改为 country**：前后端请求、prompt、store、设置面板统一使用 `country`；仅在读取旧 localStorage 时兼容 `currency` 作为迁移兜底。
+- **DOCX/PDF 导出改为纯前端**：不再依赖 `/v1/export/pdf`。当前策略是先展开所有功能块，再从编辑器 HTML 构建导出内容；`src/utils/richExport.js` 负责 HTML -> PDF / DOCX。
+- **上传单文件限制统一为 100MB**：前端校验和后端 OCR 风控上限都按 100MB 处理。
+- **视频解析策略**：上传视频时，后端 `/v1/ocr` 接收 `media_type=video`，视频画面走 OCR 模型，音轨通过 ffmpeg 抽取后走 ASR 模型，最终合并为“视频画面 OCR + 视频音频 ASR”文本。
+- **OCR 明确关闭思考**：backend/llm.py 的 OCR payload 显式下发 `options.think = False` 与 `temperature = 0`。
+- **TTS/ASR 当前真实实现**：backend/tts_asr.py 已切到 `Qwen3TTSModel + faster-whisper` 路线；不要继续按旧的 MLX-only 文档理解当前实现。
 
 ## 常用命令
 
@@ -119,7 +124,7 @@
 	6. 再次用 `docker compose exec -T ...` 验证容器内文件和行为，不能只看本地文件。
 - Docker 持久化数据统一落在部署目录内的 `docker-data/`，包括 PostgreSQL、Redis 和任务共享临时目录。
 - 容器内访问宿主机模型服务时，不要继续使用 `localhost`；应改成 `host.docker.internal` 之类的容器可达地址。
-- 当前 Docker 部署默认使用轻量后端依赖集（`backend/requirements.docker.txt`），覆盖补全、OCR、转换、文档空间和队列，不默认包含本地 `torch` / TTS / ASR 模型栈。
+- 当前 Docker 部署的 `backend/requirements.docker.txt` 已包含 OCR、转换、队列以及 `torch` / `qwen-tts` / `faster-whisper`，并在 `backend/Dockerfile` 中额外安装 `ffmpeg` 以支持视频拆音轨。
 - **Worker 容器**：worker.py 作为独立服务运行，通过 Redis Streams 消费任务队列。修改 job_handlers.py 或 worker.py 后需要验证 worker 容器内的代码已更新，可通过 `docker compose exec -T worker sh -lc "python -c 'from backend.job_handlers import get_handler; print(get_handler(\"completion\").__name__)'"` 验证。
 - **Redis Streams 架构**：任务队列使用 Redis Streams，支持并发控制、速率限制和熔断器。job_system.py 定义 JOB_TYPES 和队列配置，worker.py 注册处理器并运行事件循环。
 - 修改 Docker 相关文件时，除了代码本身，还要同步检查：

backend/.env.example

@@ -103,7 +103,7 @@ RISK_WEB_SEARCH_MAX_OUTPUT_TOKENS=4096
 RISK_WEB_SEARCH_TEMPERATURE=0.4
 RISK_COMPRESS_MAX_INPUT_CHARS=128000
 RISK_COMPRESS_MAX_OUTPUT_TOKENS=1536
-RISK_OCR_MAX_INPUT_BYTES=10485760
+RISK_OCR_MAX_INPUT_BYTES=104857600
 
 # Web search providers
 SEARXNG_BASE_URL=http://searxng:8080

backend/AGENTS.md

@@ -4,7 +4,7 @@
 
 ## 后端职责
 
-- 对外提供补全、取消补全、OCR、文档转换和 TTS 相关接口。
+- 对外提供补全、取消补全、OCR、文档转换和 TTS/ASR 相关接口。
 - 组织 Prompt，上下文清洗，调用 Ollama 模型。
 - **通过 Redis Streams 异步任务队列处理各类作业（completion/PRO/web_search/compress/OCR/convert/TTS/ASR）。**
 - 负责 API Key 校验、日志记录和部分启动预热逻辑。
@@ -62,10 +62,12 @@
 
 ### /v1/ocr
 
-- 把 base64 图片解码成字节。
-- 调用 call_vlm_ocr。
+- 把 base64 媒体内容解码成字节。
+- 支持 `media_type=image|video` 和 `mime_type`。
+- 图片直接调用 call_vlm_ocr。
+- 视频先把整段视频送入 OCR 模型，再用 ffmpeg 抽取音轨交给 ASR，最后合并文本。
 - **结果通过 job_handlers.py ocr_handler 处理。**
-- 返回识别文本和原始文件名。
+- 返回识别文本、原始文件名，以及视频场景下的 `ocr_text` / `asr_text`。
 
 ### /v1/convert
 
@@ -105,9 +107,9 @@
 ### /v1/tts-asr/*
 
 - 通过 _register_tts_asr_routes 延迟导入并挂到主应用。
-- **当前代码里的 tts_asr.py 主要是 TTS 能力，不要自行假设存在完整 ASR 实现。**
 - **TTS 请求通过 job_handlers.py tts_handler 处理。**
-- **支持多种语音模型（edge-tts、macos-say、pyttsx3）。**
+- **ASR 请求通过 job_handlers.py asr_handler 处理。**
+- **当前实现是 `Qwen3TTSModel + faster-whisper`，不是旧的 edge-tts / macos-say / MLX-only 路线。**
 
 ## 开发命令
 
@@ -146,9 +148,10 @@
 - **验证码路由**：captcha_api.py 提供 /captcha/generate 和 /captcha/verify 端点，用于前端验证用户输入。
 - **文档存储**：docs_store.py 提供 MIME 类型检测、文本/二进制分类和预览提取（8MB 限制）。
 - **LLM 策略解析**：llm_policy.py 按 job_type 解析模型配置（模型名、温度、最大 token、思考级别）。
+- **OCR 明确关闭思考**：llm.py 的 `call_vlm_ocr` 对 OCR 请求显式设置 `options.think = False` 和 `temperature = 0`。
 - **补全接口当前不是流式响应，不要按 SSE 方式改造周边代码。**
 - **ACTIVE_COMPLETIONS 在补全和取消路径里都被读写，任务生命周期要谨慎处理。**
-- **main.py 里虽然有 _convert_docx_to_pdf 辅助函数，但当前 /v1/convert 路径实际走的是 MarkItDown，不要误以为 DOCX 转 PDF 桥接脚本已接入主流程。**
+- **`/v1/export/pdf` 不是当前主链路**；DOCX/PDF 导出已经转到前端 `src/utils/richExport.js`。
 - **API_KEY 存在占位默认值，这更像本地开发兜底，不是推荐的安全模式。**
 - **历史 TTS/ASR 文档和部分测试覆盖的是旧实现；代码与文档冲突时，先确认产品方向，再决定修代码还是修文档。**
 

backend/Dockerfile

@@ -6,6 +6,10 @@ ENV PYTHONUNBUFFERED=1
 
 WORKDIR /app/backend
 
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
 COPY backend/requirements.docker.txt /tmp/requirements.docker.txt
 RUN pip install --no-cache-dir -r /tmp/requirements.docker.txt
 

backend/job_handlers.py

@@ -14,6 +14,7 @@ import markitdown
 
 from audit_store import get_audit_store
 from llm import call_ollama, call_vlm_ocr, stream_ollama_events
+from media_utils import extract_audio_wav_bytes, is_video_filename
 from prompt import (
     build_completion_prompts,
     build_pro_completion_prompts,
@@ -720,14 +721,60 @@ async def ocr_handler(
     path = payload["input_path"]
     identity, risk, lock_keys = await _enter_llm_execution(payload, emit)
     try:
+        filename = payload.get("filename", "image.jpg")
+        language = payload.get("language", "auto")
+        media_type = payload.get("media_type", "image")
+        mime_type = payload.get("mime_type", "") or ""
+
         with open(path, "rb") as handle:
-            image_bytes = handle.read()
-        text = await call_vlm_ocr(image_bytes, payload.get("language", "auto"))
+            media_bytes = handle.read()
+
+        await emit("progress", {"phase": "ocr", "media_type": media_type})
+        ocr_text = await call_vlm_ocr(
+            media_bytes,
+            language,
+            mime_type=mime_type or "application/octet-stream",
+            media_type=media_type,
+        )
         if is_cancelled():
             raise asyncio.CancelledError()
-        await emit("result", {"text": text})
-        await _exit_llm_execution(payload, identity, risk, lock_keys, status="completed", actual_output_text=text)
-        return {"text": text, "filename": payload.get("filename", "image.jpg")}
+
+        result = {
+            "text": ocr_text,
+            "ocr_text": ocr_text,
+            "filename": filename,
+            "media_type": media_type,
+        }
+
+        if media_type == "video" or is_video_filename(filename, mime_type):
+            asr_text = ""
+            if generate_asr_response is not None:
+                try:
+                    await emit("progress", {"phase": "asr", "media_type": media_type})
+                    audio_bytes = await asyncio.to_thread(extract_audio_wav_bytes, path)
+                    asr_response = await generate_asr_response(audio_bytes, language)
+                    asr_text = getattr(asr_response, "text", "") or ""
+                except Exception as exc:
+                    asr_text = f"(音频解析失败: {exc})"
+            if ocr_text.strip() or asr_text.strip():
+                text_parts = []
+                if ocr_text.strip():
+                    text_parts.append(f"## 视频画面 OCR\n\n{ocr_text.strip()}")
+                if asr_text.strip():
+                    text_parts.append(f"## 视频音频 ASR\n\n{asr_text.strip()}")
+                result["text"] = "\n\n".join(text_parts)
+            result["asr_text"] = asr_text
+
+        await emit("result", result)
+        await _exit_llm_execution(
+            payload,
+            identity,
+            risk,
+            lock_keys,
+            status="completed",
+            actual_output_text=result["text"],
+        )
+        return result
     except asyncio.CancelledError:
         await _exit_llm_execution(payload, identity, risk, lock_keys, status="cancelled", error_code="cancelled")
         raise

backend/job_system.py

@@ -77,24 +77,6 @@ class QueueFullError(JobSystemError):
         self.max_queue = max_queue
 
 
-# 修复：添加完整的异常类实现
-class JobSystemError(RuntimeError):
-    def __init__(self, message: str = "任务系统错误", error_code: str = "job_system_error") -> None:
-        super().__init__(message)
-        self.message = message
-        self.error_code = error_code
-
-    def __str__(self) -> str:
-        return f"{self.error_code}: {self.message}"
-
-
-class QueueFullError(JobSystemError):
-    def __init__(self, job_type: str, max_queue: int) -> None:
-        super().__init__(f"{job_type} 队列已满 (当前: {max_queue}/{max_queue})", "queue_full")
-        self.job_type = job_type
-        self.max_queue = max_queue
-
-
 @dataclass(frozen=True)
 class QueueConfig:
     job_type: str
@@ -613,7 +595,14 @@ class RedisWorker:
         group = self.manager._group_name(job_type)
         semaphore = self.queue_semaphores[job_type]
         while True:
-            streams = await self.manager.redis.xreadgroup(group, self.consumer_name, {queue_key: ">"}, count=1, block=1000)
+            try:
+                streams = await self.manager.redis.xreadgroup(group, self.consumer_name, {queue_key: ">"}, count=1, block=1000)
+            except asyncio.CancelledError:
+                raise
+            except Exception as exc:
+                logger.warning("redis worker consume loop retrying type=%s error=%s", job_type, exc)
+                await asyncio.sleep(1)
+                continue
             if not streams:
                 continue
             for _, messages in streams:

backend/llm.py

@@ -551,31 +551,43 @@ async def stream_ollama_events(
     )
 
 
-async def call_vlm_ocr(image_bytes: bytes, language: str = 'auto') -> str:
-    """OCR via VLM using OpenAI-compatible vision API (image_url content part)."""
+async def call_vlm_ocr(
+    media_bytes: bytes,
+    language: str = 'auto',
+    *,
+    mime_type: str = 'image/png',
+    media_type: str = 'image',
+) -> str:
+    """OCR via VLM using OpenAI-compatible multimodal API."""
     start = time.perf_counter()
     start_dt = datetime.now()
 
     logger.info(
-        '[VLM][ocr] request model=%s base_url=%s image_bytes=%d language=%s',
-        VLM_MODEL, LLM_BASE_URL, len(image_bytes), language,
+        '[VLM][ocr] request model=%s base_url=%s media_type=%s media_bytes=%d language=%s mime=%s',
+        VLM_MODEL, LLM_BASE_URL, media_type, len(media_bytes), language, mime_type,
     )
 
-    image_b64 = base64.b64encode(image_bytes).decode('ascii')
+    media_b64 = base64.b64encode(media_bytes).decode('ascii')
+    content_part_type = 'video_url' if media_type == 'video' else 'image_url'
+    url_key = 'video_url' if media_type == 'video' else 'image_url'
 
     payload = {
         'model': VLM_MODEL,
         'messages': [{
             'role': 'user',
             'content': [
-                {'type': 'text', 'text': get_vlm_ocr_prompt()},
+                {'type': 'text', 'text': f"{get_vlm_ocr_prompt()}\n\nTarget language hint: {language or 'auto'}"},
                 {
-                    'type': 'image_url',
-                    'image_url': {'url': f'data:image/png;base64,{image_b64}'},
+                    'type': content_part_type,
+                    url_key: {'url': f'data:{mime_type or "application/octet-stream"};base64,{media_b64}'},
                 },
             ],
         }],
         'stream': False,
+        'options': {
+            'temperature': 0,
+            'think': False,
+        },
     }
 
     http_timeout = httpx.Timeout(connect=10.0, read=None, write=30.0, pool=30.0)

backend/main.py

@@ -106,6 +106,8 @@ class OCRRequest(BaseModel):
     image: str
     filename: str = "image.jpg"
     language: str = "auto"
+    media_type: str = "image"
+    mime_type: str | None = None
 
 
 class ConvertRequest(BaseModel):
@@ -281,8 +283,10 @@ def _register_handlers() -> None:
     确保 Redis 重连或实例重建后处理器不会丢失。"""
     global _handlers_registered
     manager = get_job_manager()
-    # 强制清空旧 handlers，避免重复注册累积
-    manager.handlers.clear()
+    # 强制清空旧 handlers，避免重复注册累积；测试替身只暴露 register_handler。
+    handlers = getattr(manager, "handlers", None)
+    if handlers is not None:
+        handlers.clear()
     manager.register_handler("completion", completion_handler)
     manager.register_handler("pro_completion", pro_completion_handler)
     manager.register_handler("web_search", web_search_handler)
@@ -294,7 +298,8 @@ def _register_handlers() -> None:
     _handlers_registered = True
     
     # 打印注册信息便于调试
-    logger.info("handlers registered: %s", list(manager.handlers.keys()))tered: %s", list(manager.handlers.keys()))
+    registered = list(getattr(manager, "handlers", {}).keys())
+    logger.info("handlers registered: %s", registered)
 
 
 def _sse(event: str, data: dict) -> str:
@@ -623,16 +628,28 @@ async def ocr_image(request: Request, req: OCRRequest, auth: dict = Security(_au
     except Exception as exc:
         return JSONResponse({"error": str(exc)}, status_code=500)
     if len(image_bytes) > config.ocr_max_input_bytes:
-        return JSONResponse({"error": "图片过大，无法执行 OCR"}, status_code=400)
+        return JSONResponse({"error": "文件过大，无法执行 OCR/视频解析"}, status_code=400)
     input_path = persist_temp_input(image_bytes, os.path.splitext(req.filename)[1] or ".img")
     try:
         identity, payload = await _prepare_llm_payload(
             request,
             job_type="ocr",
-            request_body={"filename": req.filename, "language": req.language, "image_bytes": len(image_bytes)},
+            request_body={
+                "filename": req.filename,
+                "language": req.language,
+                "media_type": req.media_type,
+                "mime_type": req.mime_type,
+                "image_bytes": len(image_bytes),
+            },
             raw_size=len(image_bytes),
-            token_source_text=f"{req.filename}:{len(image_bytes)}:{req.language}",
-            extra_payload={"input_path": input_path, "filename": req.filename, "language": req.language},
+            token_source_text=f"{req.filename}:{req.media_type}:{req.mime_type}:{len(image_bytes)}:{req.language}",
+            extra_payload={
+                "input_path": input_path,
+                "filename": req.filename,
+                "language": req.language,
+                "media_type": req.media_type,
+                "mime_type": req.mime_type,
+            },
         )
         job_id = await _queue_job("ocr", payload, identity.request_id)
     except RiskRejected as exc:

backend/media_utils.py

@@ -0,0 +1,47 @@
+from __future__ import annotations
+
+import os
+import subprocess
+import tempfile
+
+
+VIDEO_EXTENSIONS = {".mp4", ".webm", ".mov", ".avi", ".mkv", ".m4v", ".ogv"}
+
+
+def is_video_filename(filename: str = "", mime_type: str = "") -> bool:
+    ext = os.path.splitext(filename or "")[1].lower()
+    mime = (mime_type or "").strip().lower()
+    return ext in VIDEO_EXTENSIONS or mime.startswith("video/")
+
+
+def extract_audio_wav_bytes(input_path: str) -> bytes:
+    if not input_path or not os.path.exists(input_path):
+        raise FileNotFoundError("输入媒体文件不存在")
+
+    fd, output_path = tempfile.mkstemp(suffix=".wav")
+    os.close(fd)
+    try:
+        subprocess.run(
+            [
+                "ffmpeg",
+                "-y",
+                "-i",
+                input_path,
+                "-vn",
+                "-acodec",
+                "pcm_s16le",
+                "-ar",
+                "16000",
+                "-ac",
+                "1",
+                output_path,
+            ],
+            check=True,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+        with open(output_path, "rb") as handle:
+            return handle.read()
+    finally:
+        if os.path.exists(output_path):
+            os.unlink(output_path)

backend/models.py

@@ -5,5 +5,5 @@ from pydantic import BaseModel
 class UserPreferences(BaseModel):
     """用户偏好设置"""
     language: str = "auto"
-    currency: str = "auto"
+    country: str = "auto"
     timezone: str = "auto"

backend/prompt.py

@@ -355,8 +355,8 @@ def build_completion_prompts(
     if preferences:
         if preferences.language and preferences.language != "auto":
             pref_info.append(f"Preferred language: {preferences.language}")
-        if preferences.currency and preferences.currency != "auto":
-            pref_info.append(f"Preferred currency: {preferences.currency}")
+        if preferences.country and preferences.country != "auto":
+            pref_info.append(f"Preferred country: {preferences.country}")
 
     preferences_instruction = "\n".join(pref_info)
     if preferences_instruction:
@@ -454,8 +454,8 @@ def build_pro_completion_prompts(
     if preferences:
         if preferences.language and preferences.language != "auto":
             pref_info.append(f"Preferred language: {preferences.language}")
-        if preferences.currency and preferences.currency != "auto":
-            pref_info.append(f"Preferred currency: {preferences.currency}")
+        if preferences.country and preferences.country != "auto":
+            pref_info.append(f"Preferred country: {preferences.country}")
         if preferences.timezone and preferences.timezone != "auto":
             pref_info.append(f"Preferred timezone: {preferences.timezone}")
 

backend/requirements.docker.txt

@@ -8,3 +8,10 @@ python-multipart>=0.0.9
 python-dotenv>=1.0.0
 markitdown>=0.1.1
 geoip2>=4.8.0
+numpy>=1.26.0
+torch>=2.2.0
+soundfile>=0.12.1
+scipy>=1.13.0
+qwen-tts
+modelscope>=1.18.0
+faster-whisper>=1.1.0

backend/risk_config.py

@@ -92,7 +92,7 @@ class RiskConfig:
 def load_risk_config() -> RiskConfig:
     raw_origins = _str_env(
         "CORS_ALLOW_ORIGINS",
-        "https://chat.imageteach.tech,http://localhost:8080,http://127.0.0.1:8080",
+        "https://chat.imageteach.tech,http://localhost:8080,http://127.0.0.1:8080,http://localhost:5173,http://127.0.0.1:5173",
     )
     cors_allow_origins = tuple(
         origin.strip() for origin in raw_origins.split(",") if origin.strip()
@@ -138,7 +138,7 @@ def load_risk_config() -> RiskConfig:
         web_search_temperature=_float_env("RISK_WEB_SEARCH_TEMPERATURE", 0.4),
         compress_max_input_chars=_int_env("RISK_COMPRESS_MAX_INPUT_CHARS", 128000),
         compress_max_output_tokens=_int_env("RISK_COMPRESS_MAX_OUTPUT_TOKENS", 1536),
-        ocr_max_input_bytes=_int_env("RISK_OCR_MAX_INPUT_BYTES", 10 * 1024 * 1024),
+        ocr_max_input_bytes=_int_env("RISK_OCR_MAX_INPUT_BYTES", 100 * 1024 * 1024),
         completion_input_cost_per_1k=_float_env("RISK_COMPLETION_INPUT_COST_PER_1K", 0.0004),
         completion_output_cost_per_1k=_float_env("RISK_COMPLETION_OUTPUT_COST_PER_1K", 0.0016),
         pro_input_cost_per_1k=_float_env("RISK_PRO_INPUT_COST_PER_1K", 0.003),

backend/tests/test_llm.py

@@ -299,3 +299,4 @@ def test_call_vlm_ocr(monkeypatch):
     image_part = [p for p in content_parts if p.get("type") == "image_url"]
     assert len(image_part) == 1
     assert image_part[0]["image_url"]["url"].startswith("data:image/png;base64,")
+    assert captured["json"]["options"]["think"] is False

backend/tests/test_main_endpoints.py

@@ -4,6 +4,7 @@ import importlib
 import os
 import sys
 from pathlib import Path
+from types import SimpleNamespace
 
 from fastapi.testclient import TestClient
 
@@ -158,6 +159,35 @@ def test_post_ocr_mocked(monkeypatch):
     assert "OCR result text" in body
 
 
+def test_post_video_ocr_merges_ocr_and_asr(monkeypatch):
+    async def fake_ocr(*args, **kwargs):
+        return "画面文字"
+
+    async def fake_asr(*args, **kwargs):
+        return SimpleNamespace(text="音频转写")
+
+    monkeypatch.setattr(job_handlers, "call_vlm_ocr", fake_ocr)
+    monkeypatch.setattr(job_handlers, "generate_asr_response", fake_asr)
+    monkeypatch.setattr(job_handlers, "extract_audio_wav_bytes", lambda _path: b"fake wav")
+
+    video_b64 = base64.b64encode(b"pretend video data").decode()
+    with TestClient(main.app) as client:
+        with client.stream("POST", "/v1/ocr", headers=HEADERS, json={
+            "image": video_b64,
+            "filename": "sample.mp4",
+            "language": "auto",
+            "media_type": "video",
+            "mime_type": "video/mp4",
+        }) as resp:
+            assert resp.status_code == 200
+            body = "".join(resp.iter_text())
+
+    assert "视频画面 OCR" in body
+    assert "视频音频 ASR" in body
+    assert "画面文字" in body
+    assert "音频转写" in body
+
+
 def test_post_convert_txt_returns_markdown():
     content = base64.b64encode(b"hello world").decode()
     with TestClient(main.app) as client:

backend/tests/test_pro_completions.py

@@ -35,7 +35,7 @@ def _payload():
         "privacy_mode": True,
         "user_preferences": {
             "language": "zh",
-            "currency": "CNY",
+            "country": "CN",
             "timezone": "Asia/Shanghai",
         },
     }
@@ -43,7 +43,7 @@ def _payload():
 
 def test_pro_queue_full_returns_429(monkeypatch):
     async def fake_queue_job(*args, **kwargs):
-        raise job_system.QueueFullError("pro_completion queue is full")
+        raise job_system.QueueFullError("pro_completion", 8)
 
     monkeypatch.setattr(main, "_queue_job", fake_queue_job)
     with TestClient(main.app) as client:
@@ -79,13 +79,13 @@ def test_pro_prompt_accepts_serialized_preferences():
         instruction="expand",
         preferences={
             "language": "zh",
-            "currency": "CNY",
+            "country": "CN",
             "timezone": "Asia/Shanghai",
         },
     )
 
     assert "Preferred language: zh" in user_prompt
-    assert "Preferred currency: CNY" in user_prompt
+    assert "Preferred country: CN" in user_prompt
     assert "Preferred timezone: Asia/Shanghai" in user_prompt
 
 

backend/tests/test_web_search.py

@@ -17,6 +17,7 @@ if str(BACKEND_DIR) not in sys.path:
 
 import job_handlers  # type: ignore
 import job_system  # type: ignore
+import llm  # type: ignore
 import risk_control  # type: ignore
 import session_store  # type: ignore
 import audit_store  # type: ignore
@@ -55,10 +56,14 @@ def test_web_search_route_returns_done(monkeypatch):
             return {"content": '["vector database comparison", "pinecone weaviate qdrant"]'}
         if tag.endswith("-webu"):
             return {"content": '["https://example.com/a", "https://example.com/b"]'}
-        if tag.endswith("-webf"):
-            return {"content": "第一段\n\n第二段"}
         raise AssertionError(f"unexpected tag: {tag}")
 
+    async def fake_stream_ollama_events(prompt, system_prompt=None, tag="", **kwargs):  # noqa: ARG001
+        if not tag.endswith("-webf"):
+            raise AssertionError(f"unexpected stream tag: {tag}")
+        yield "content", "第一段\n\n"
+        yield "content", "第二段"
+
     async def fake_searxng_search(query, *, limit):  # noqa: ARG001
         return [
             {
@@ -83,6 +88,7 @@ def test_web_search_route_returns_done(monkeypatch):
     monkeypatch.setattr(job_handlers, "call_ollama", fake_call_ollama)
     monkeypatch.setattr(job_handlers, "_searxng_search", fake_searxng_search)
     monkeypatch.setattr(job_handlers, "_firecrawl_scrape", fake_firecrawl_scrape)
+    monkeypatch.setattr(llm, "stream_ollama_events", fake_stream_ollama_events)
 
     with TestClient(main.app) as client:
         with client.stream("POST", "/v1/web-search", headers=HEADERS, json=_payload()) as resp:

backend/tts_asr.py

@@ -1,260 +1,133 @@
 import asyncio
 import base64
-import io
 import logging
 import os
 import tempfile
-import wave
 from typing import Optional
 
-# 设置 Hugging Face / ModelScope 镜像源为国内镜像
 os.environ.setdefault("HF_ENDPOINT", "https://hf-mirror.com")
 
-import numpy as np
-import torch
 from fastapi import APIRouter, HTTPException
 from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
 
 try:
-    asyncio.get_running_loop()
-except RuntimeError:
-    asyncio.set_event_loop(asyncio.new_event_loop())
+    import numpy as np  # type: ignore
+except Exception as exc:  # pragma: no cover
+    logger.debug("numpy import failed: %s", exc)
+    np = None  # type: ignore
+
+try:
+    import torch  # type: ignore
+except Exception as exc:  # pragma: no cover
+    logger.debug("torch import failed: %s", exc)
+    torch = None  # type: ignore
 
-# New TTS model import
 try:
     from qwen_tts import Qwen3TTSModel  # type: ignore
-except Exception as e:  # pragma: no cover
-    logger.debug("qwen_tts import failed (optional): %s", e)
+except Exception as exc:  # pragma: no cover
+    logger.debug("qwen_tts import failed: %s", exc)
     Qwen3TTSModel = None  # type: ignore
 
-# ASR model import (MLX-based, Apple Silicon only)
 try:
-    from mlx_audio.stt.models.qwen3_asr import (  # type: ignore
-        ForcedAlignerModel,
-        Qwen3ASRModel,
-    )
-except Exception as e:  # pragma: no cover
-    logger.debug("mlx_audio import failed (optional): %s", e)
-    Qwen3ASRModel = None  # type: ignore
-    ForcedAlignerModel = None  # type: ignore
+    from faster_whisper import WhisperModel  # type: ignore
+except Exception as exc:  # pragma: no cover
+    logger.debug("faster_whisper import failed: %s", exc)
+    WhisperModel = None  # type: ignore
 
 try:
     from modelscope import snapshot_download  # type: ignore
-except Exception as e:  # pragma: no cover
-    logger.debug("modelscope import failed (optional): %s", e)
+except Exception as exc:  # pragma: no cover
+    logger.debug("modelscope import failed: %s", exc)
+    snapshot_download = None  # type: ignore
 
 meta_router = APIRouter()
 generation_router = APIRouter()
 
-# Global model instances
-_tts_model: Optional["Qwen3TTSModel"] = None
-_asr_model: Optional[object] = None  # Qwen3ASRModel or ForcedAlignerModel
-_align_model: Optional[object] = None  # Qwen3-ForcedAlignerModel
-
-# Model paths for loading
 MODEL_ID_HF = "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign"
 MODEL_ID_MS = "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign"
+ASR_MODEL_ID = os.getenv("ASR_MODEL_ID", "small")
+ASR_COMPUTE_TYPE = os.getenv("ASR_COMPUTE_TYPE", "int8")
 
-# ModelScope ASR/ForcedAligner models (MLX 4-bit format)
-ASR_MODEL_ID_MS = "aufklarer/Qwen3-ASR-0.6B-MLX-4bit"
-ALIGN_MODEL_ID_MS = "aufklarer/Qwen3-ForcedAligner-0.6B-MLX"
+_tts_model: Optional["Qwen3TTSModel"] = None
+_asr_model: Optional["WhisperModel"] = None
 
 
 def _get_device_map() -> str:
-    """设备检测逻辑：优先 CUDA，其次 MPS，最后 CPU"""
+    if torch is None:
+        return "cpu"
     if torch.cuda.is_available():
-        return "cuda:0"
+        return "cuda"
     try:
         if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
             return "mps"
-    except Exception as e:  # noqa: ANN001
-        logger.debug("MPS check failed: %s", e)
+    except Exception as exc:  # pragma: no cover
+        logger.debug("MPS check failed: %s", exc)
     return "cpu"
 
 
-def _download_model_from_modelscope() -> Optional[str]:
-    """从 ModelScope 下载模型到本地缓存目录"""
+def _download_tts_model_from_modelscope() -> Optional[str]:
+    if snapshot_download is None:
+        return None
+    cache_dir = os.path.join(os.path.dirname(__file__), "models")
+    os.makedirs(cache_dir, exist_ok=True)
     try:
-        cache_dir = os.path.join(os.path.dirname(__file__), "models")
-        os.makedirs(cache_dir, exist_ok=True)
-        model_dir = snapshot_download(
-            MODEL_ID_MS,
-            cache_dir=cache_dir,
-            revision="master"
-        )
-        logger.info("ModelScope 模型下载完成: %s", model_dir)
-        return model_dir
-    except Exception as e:  # noqa: ANN001
-        logger.warning("ModelScope 下载失败: %s", e)
+        return snapshot_download(MODEL_ID_MS, cache_dir=cache_dir, revision="master")
+    except Exception as exc:  # pragma: no cover
+        logger.warning("ModelScope TTS download failed: %s", exc)
         return None
 
 
-async def _warmup_tts():
-    """预热 TTS 模型"""
-    await asyncio.to_thread(_load_tts_model_with_retry)
-
-
-async def _warmup_asr():
-    """预热 ASR 模型（从 ModelScope 下载并加载）"""
-    await asyncio.to_thread(_load_asr_models)
-
-
-async def _warmup_all():
-    """预热所有模型（TTS 和 ASR）"""
-    logger.info("[Warmup] 开始预热 TTS 模型...")
-    await _warmup_tts()
-    logger.info("[Warmup] TTS 模型预热完成")
-
-    if Qwen3ASRModel is not None:
-        logger.info("[Warmup] 开始预热 ASR 模型...")
-        await _warmup_asr()
-        logger.info("[Warmup] ASR 模型预热完成")
-
-
-def _load_tts_model_with_retry(max_retries: int = 3) -> "Qwen3TTSModel":
-    """加载 TTS 模型，支持多个镜像源"""
+def _ensure_tts_model() -> "Qwen3TTSModel":
     global _tts_model
     if _tts_model is not None:
         return _tts_model
-    if Qwen3TTSModel is None:
-        raise RuntimeError("qwen_tts 库未安装，无法加载 TTS 模型")
+    if np is None or torch is None or Qwen3TTSModel is None:
+        raise RuntimeError("TTS 依赖未安装完整")
 
     device_map = _get_device_map()
-    last_err = None
+    dtype = torch.float16 if device_map != "cpu" else torch.float32
 
-    # 策略1: 尝试从 ModelScope 下载后加载
-    for attempt in range(max_retries):
-        try:
-            logger.info("尝试从 ModelScope 下载 TTS 模型...")
-            model_path = _download_model_from_modelscope()
-            if model_path and os.path.isdir(model_path):
-                _tts_model = Qwen3TTSModel.from_pretrained(  # type: ignore
-                    model_path,
-                    device_map=device_map,
-                    dtype=torch.float16,
-                )
-                logger.info("ModelScope TTS 模型加载成功: %s", model_path)
-                return _tts_model
-        except Exception as e:  # noqa: ANN001
-            logger.warning("ModelScope TTS 加载失败 (尝试 %d/%d): %s", attempt + 1, max_retries, e)
-            last_err = e
+    model_path = _download_tts_model_from_modelscope()
+    last_error = None
 
-    # 策略2: 尝试从 HuggingFace 镜像加载
-    for attempt in range(max_retries):
+    for candidate in [model_path, MODEL_ID_HF]:
+        if not candidate:
+            continue
         try:
-            logger.info("尝试从 HuggingFace 镜像加载 TTS...")
             _tts_model = Qwen3TTSModel.from_pretrained(  # type: ignore
-                MODEL_ID_HF,
+                candidate,
                 device_map=device_map,
-                dtype=torch.float16,
+                dtype=dtype,
             )
-            logger.info("HuggingFace TTS 模型加载成功")
             return _tts_model
-        except Exception as e:  # noqa: ANN001
-            logger.warning("HuggingFace TTS 加载失败 (尝试 %d/%d): %s", attempt + 1, max_retries, e)
-            last_err = e
+        except Exception as exc:
+            last_error = exc
+            logger.warning("TTS model load failed from %s: %s", candidate, exc)
 
-    raise RuntimeError(f"无法加载 TTS 模型: {last_err}") from last_err
+    raise RuntimeError(f"TTS 模型加载失败: {last_error}") from last_error
 
 
-def _load_asr_models() -> None:
-    """从 ModelScope 下载并加载 ASR/ForcedAligner MLX 模型"""
-    global _asr_model, _align_model
-
-    if snapshot_download is None:
-        logger.warning("modelscope 未安装，跳过 ASR 模型加载")
-        return
-
-    if Qwen3ASRModel is None:
-        logger.warning("mlx_audio 未安装，跳过 ASR 模型加载")
-        return
-
-    # Download and load ASR model from ModelScope
-    try:
-        logger.info("从 ModelScope 下载 ASR 模型...")
-        asr_cache_dir = os.path.join(os.path.dirname(__file__), "models", "asr")
-        asr_model_dir = snapshot_download(ASR_MODEL_ID_MS, cache_dir=asr_cache_dir)
-        _load_asr_from_path(asr_model_dir)
-    except Exception as e:  # noqa: ANN001
-        logger.warning("ASR ModelScope 下载失败，尝试 hf-mirror: %s", e)
-        try:
-            _load_asr_from_hf_mirror()
-        except Exception as e2:  # noqa: ANN001
-            logger.warning("ASR hf-mirror 加载失败，跳过 ASR: %s", e2)
-
-    # Download and load ForcedAligner model from ModelScope
-    try:
-        logger.info("从 ModelScope 下载 ForcedAligner 模型...")
-        align_cache_dir = os.path.join(os.path.dirname(__file__), "models", "aligner")
-        align_model_dir = snapshot_download(ALIGN_MODEL_ID_MS, cache_dir=align_cache_dir)
-        _load_align_from_path(align_model_dir)
-    except Exception as e:  # noqa: ANN001
-        logger.warning("ForcedAligner ModelScope 下载失败，尝试 hf-mirror: %s", e)
-        try:
-            _load_align_from_hf_mirror()
-        except Exception as e2:  # noqa: ANN001
-            logger.warning("ForcedAligner hf-mirror 加载失败，跳过: %s", e2)
-
-
-def _load_asr_from_path(model_dir: str) -> None:
-    """从本地路径加载 ASR MLX 模型"""
+def _ensure_asr_model() -> "WhisperModel":
     global _asr_model
-    try:
-        from mlx_audio.stt.utils import load as stt_load  # type: ignore
+    if _asr_model is not None:
+        return _asr_model
+    if WhisperModel is None:
+        raise RuntimeError("faster-whisper 未安装")
 
-        model = stt_load(model_dir)
-        _asr_model = model
-        logger.info("ASR 模型加载成功 (路径: %s)", model_dir)
-    except Exception as e:  # noqa: ANN001
-        logger.warning("ASR MLX 加载失败，尝试直接构建: %s", e)
-        try:
-            from mlx.core import load as mx_load  # type: ignore
-
-            weights = mx_load(os.path.join(model_dir, "model.safetensors"))
-            from mlx_lm import load as lm_load  # type: ignore
-
-            model = lm_load(model_dir, model_cls=Qwen3ASRModel)
-            _asr_model = model
-        except Exception as e2:  # noqa: ANN001
-            raise RuntimeError(f"无法加载 ASR MLX 模型: {e2}") from e
+    device = "cuda" if _get_device_map() == "cuda" else "cpu"
+    compute_type = ASR_COMPUTE_TYPE if device == "cpu" else "float16"
+    _asr_model = WhisperModel(ASR_MODEL_ID, device=device, compute_type=compute_type)
+    return _asr_model
 
 
-def _load_asr_from_hf_mirror() -> None:
-    """从 hf-mirror 加载 ASR MLX 模型"""
-    global _asr_model
-    try:
-        from mlx_audio.stt.utils import load as stt_load  # type: ignore
-
-        model = stt_load("mlx-community/Qwen3-ASR-0.6B-4bit")
-        _asr_model = model
-    except Exception as e:  # noqa: ANN001
-        raise RuntimeError(f"无法从 hf-mirror 加载 ASR MLX: {e}") from e
+async def _warmup_tts():
+    await asyncio.to_thread(_ensure_tts_model)
 
 
-def _load_align_from_path(model_dir: str) -> None:
-    """从本地路径加载 ForcedAligner MLX 模型"""
-    global _align_model
-    try:
-        from mlx_audio.stt.utils import load as stt_load  # type: ignore
-
-        model = stt_load(model_dir)
-        _align_model = model
-    except Exception as e:  # noqa: ANN001
-        raise RuntimeError(f"无法加载 ForcedAligner MLX 模型 (路径: {model_dir}): {e}") from e
-
-
-def _load_align_from_hf_mirror() -> None:
-    """从 hf-mirror 加载 ForcedAligner MLX 模型"""
-    global _align_model
-    try:
-        from mlx_audio.stt.utils import load as stt_load  # type: ignore
-
-        model = stt_load("mlx-community/Qwen3-ForcedAligner-0.6B-4bit")
-        _align_model = model
-    except Exception as e:  # noqa: ANN001
-        raise RuntimeError(f"无法从 hf-mirror 加载 ForcedAligner MLX: {e}") from e
+async def _warmup_asr():
+    await asyncio.to_thread(_ensure_asr_model)
 
 
 class TTSRequest(BaseModel):
@@ -286,43 +159,25 @@ class ModelStatus(BaseModel):
     device: str
 
 
-def _ensure_tts_model() -> "Qwen3TTSModel":
-    """确保 TTS 模型已加载"""
-    global _tts_model
-    if _tts_model is None:
-        _tts_model = _load_tts_model_with_retry()
-    return _tts_model
-
-
-def _ensure_asr_model():
-    """确保 ASR 模型已加载（懒加载）"""
-    global _asr_model
-    if _asr_model is None:
-        try:
-            from mlx_audio.stt.utils import load as stt_load  # type: ignore
-
-            _asr_model = stt_load(ASR_MODEL_ID_MS)
-        except Exception as e:  # noqa: ANN001
-            raise RuntimeError(f"无法加载 ASR MLX 模型 (路径: {ASR_MODEL_ID_MS}): {e}") from e
-    return _asr_model
-
-
-def _ensure_align_model():
-    """确保 ForcedAligner 模型已加载（懒加载）"""
-    global _align_model
-    if _align_model is None:
-        try:
-            from mlx_audio.stt.utils import load as stt_load  # type: ignore
-
-            _align_model = stt_load(ALIGN_MODEL_ID_MS)
-        except Exception as e:  # noqa: ANN001
-            raise RuntimeError(f"无法加载 ForcedAligner MLX 模型 (路径: {ALIGN_MODEL_ID_MS}): {e}") from e
-    return _align_model
+def _normalize_language(language: Optional[str]) -> Optional[str]:
+    if not language:
+        return None
+    value = language.strip().lower()
+    if value in {"auto", ""}:
+        return None
+    mapping = {
+        "zh-cn": "zh",
+        "zh-hans": "zh",
+        "zh-tw": "zh",
+        "en-us": "en",
+        "ja-jp": "ja",
+        "ko-kr": "ko",
+    }
+    return mapping.get(value, value.split("-")[0])
 
 
 @meta_router.get("/status", response_model=ModelStatus)
 async def get_status():
-    """获取模型状态"""
     return ModelStatus(
         tts_loaded=_tts_model is not None,
         asr_loaded=_asr_model is not None,
@@ -332,11 +187,10 @@ async def get_status():
 
 @meta_router.get("/config")
 async def get_config():
-    """获取配置信息"""
     return {
         "model": {
             "tts": MODEL_ID_MS,
-            "asr": ASR_MODEL_ID_MS if Qwen3ASRModel is not None else None,
+            "asr": ASR_MODEL_ID,
         },
         "device": _get_device_map(),
         "status": {
@@ -348,15 +202,11 @@ async def get_config():
 
 @meta_router.post("/warmup")
 async def warmup_models():
-    """手动触发模型预热"""
     await _warmup_tts()
-
-    if Qwen3ASRModel is not None:
-        await _warmup_asr()
-
+    await _warmup_asr()
     return {
         "tts_warmup": _tts_model is not None,
-        "asr_warmup": _asr_model is not None if Qwen3ASRModel else False,
+        "asr_warmup": _asr_model is not None,
         "device": _get_device_map(),
     }
 
@@ -367,26 +217,30 @@ async def generate_tts_response(
     speaker: str = "Vivian",
     output_format: str = "wav",
 ) -> TTSResponse:
-    del speaker  # current model path does not expose multi-speaker routing
-    del output_format  # current implementation always returns wav
-    try:
-        model = _ensure_tts_model()
-    except Exception as e:  # noqa: ANN001
-        raise HTTPException(status_code=500, detail=str(e))
+    del speaker
+    del output_format
+    if np is None:
+        raise HTTPException(status_code=501, detail="numpy 未安装，TTS 功能不可用")
 
     try:
-        wavs, sr = model.generate_voice_design(  # type: ignore
+        model = _ensure_tts_model()
+    except Exception as exc:
+        raise HTTPException(status_code=500, detail=str(exc))
+
+    try:
+        wavs, sample_rate = await asyncio.to_thread(
+            model.generate_voice_design,  # type: ignore
             text=text,
             language="Chinese",
             instruct=instruct or "",
         )
-    except Exception as e:  # noqa: ANN001
-        logger.exception("TTS 推理失败")
-        raise HTTPException(status_code=500, detail=f"TTS 推理失败: {e}")
+    except Exception as exc:
+        logger.exception("TTS inference failed")
+        raise HTTPException(status_code=500, detail=f"TTS 推理失败: {exc}")
 
     wav_data = wavs[0] if isinstance(wavs, (list, tuple)) else wavs
-    if hasattr(wav_data, 'numpy'):  # type: ignore
-        wav_data = wav_data.cpu().numpy()  # type: ignore
+    if hasattr(wav_data, "cpu"):
+        wav_data = wav_data.cpu().numpy()
     wav_data = np.asarray(wav_data, dtype=np.float32)
 
     tmp_path = None
@@ -394,81 +248,66 @@ async def generate_tts_response(
         import soundfile as sf  # type: ignore
 
         fd, tmp_path = tempfile.mkstemp(suffix=".wav")
-        os.close(fd)  # type: ignore
-        sf.write(tmp_path, wav_data, sr)
-        with open(tmp_path, "rb") as f:  # noqa: SIM115
-            audio_bytes = f.read()
-    except Exception as e:  # noqa: ANN001
-        logger.exception("音频编码失败")
-        raise HTTPException(status_code=500, detail=f"音频编码失败: {e}")
+        os.close(fd)
+        sf.write(tmp_path, wav_data, sample_rate)
+        with open(tmp_path, "rb") as handle:
+            audio_bytes = handle.read()
+    except Exception as exc:
+        logger.exception("TTS audio encode failed")
+        raise HTTPException(status_code=500, detail=f"音频编码失败: {exc}")
     finally:
-        if tmp_path and os.path.exists(tmp_path):  # noqa: SIM201
-            try:
-                os.unlink(tmp_path)
-            except Exception:
-                pass
+        if tmp_path and os.path.exists(tmp_path):
+            os.unlink(tmp_path)
 
-    duration_ms = int(len(wav_data) / sr * 1000) if sr > 0 else 0
-    audio_base64 = base64.b64encode(audio_bytes).decode("utf-8")
-    return TTSResponse(audio_base64=audio_base64, format="wav", duration_ms=duration_ms)
+    duration_ms = int(len(wav_data) / sample_rate * 1000) if sample_rate > 0 else 0
+    return TTSResponse(
+        audio_base64=base64.b64encode(audio_bytes).decode("utf-8"),
+        format="wav",
+        duration_ms=duration_ms,
+    )
 
 
 async def generate_asr_response(audio_bytes: bytes, language: Optional[str] = "zh-CN") -> ASRResponse:
-    if Qwen3ASRModel is None:
-        raise HTTPException(status_code=501, detail="mlx_audio 未安装，ASR 功能不可用")
+    if not audio_bytes:
+        raise HTTPException(status_code=400, detail="音频内容为空")
 
     try:
         model = _ensure_asr_model()
-    except Exception as e:  # noqa: ANN001
-        raise HTTPException(status_code=500, detail=f"ASR 模型加载失败: {e}")
+    except Exception as exc:
+        raise HTTPException(status_code=500, detail=f"ASR 模型加载失败: {exc}")
 
+    normalized_language = _normalize_language(language)
+    tmp_path = None
     try:
-        wav_buffer = io.BytesIO(audio_bytes)
-        with wave.open(wav_buffer, 'rb') as wf:  # noqa: SIM115
-            n_channels = wf.getnchannels()
-            sampwidth = wf.getsampwidth()
-            framerate = wf.getframerate()
-            n_frames = wf.getnframes()
+        fd, tmp_path = tempfile.mkstemp(suffix=".wav")
+        os.close(fd)
+        with open(tmp_path, "wb") as handle:
+            handle.write(audio_bytes)
 
-            raw_data = wf.readframes(n_frames)
-            audio_array = np.frombuffer(raw_data, dtype=np.int16 if sampwidth == 2 else np.float32)
-
-            if n_channels > 1:
-                audio_array = np.mean(audio_array.reshape(-1, n_channels), axis=1)
-
-            if framerate != 16000:
-                try:
-                    import scipy.signal as signal  # type: ignore
-
-                    n_samples = int(len(audio_array) * 16000 / framerate)
-                    audio_array = signal.resample(audio_array, n_samples)  # type: ignore
-                except Exception as e2:  # noqa: ANN001
-                    logger.warning("重采样失败，使用原始音频: %s", e2)
-
-            if audio_array.dtype == np.int16:
-                audio_array = audio_array.astype(np.float32) / 32768.0
-
-        result = model.generate(  # type: ignore
-            audio_array,
-            language=language if language else None,
+        segments, info = await asyncio.to_thread(
+            model.transcribe,
+            tmp_path,
+            language=normalized_language,
+            vad_filter=True,
+            beam_size=5,
         )
-
-        recognized_text = getattr(result, 'text', str(result)) if hasattr(result, 'text') else str(result)
-        detected_lang = getattr(result, 'language', language or "zh-CN")
-        if isinstance(detected_lang, list) and len(detected_lang) > 0:
-            detected_lang = detected_lang[0]
-
-        return ASRResponse(text=recognized_text, language=str(detected_lang))
+        text = "".join(segment.text for segment in segments).strip()
+        if not text:
+            raise RuntimeError("ASR 返回结果为空")
+        detected_language = getattr(info, "language", normalized_language or "unknown")
+        return ASRResponse(text=text, language=str(detected_language))
     except HTTPException:
         raise
-    except Exception as e:  # noqa: ANN001
-        logger.exception("ASR 推理失败")
-        raise HTTPException(status_code=500, detail=f"ASR 推理失败: {e}")
+    except Exception as exc:
+        logger.exception("ASR inference failed")
+        raise HTTPException(status_code=500, detail=f"ASR 推理失败: {exc}")
+    finally:
+        if tmp_path and os.path.exists(tmp_path):
+            os.unlink(tmp_path)
 
 
 @generation_router.post("/tts", response_model=TTSResponse)
 async def tts_endpoint(req: TTSRequest):
-    """TTS 文字转语音端点"""
     return await generate_tts_response(
         text=req.text,
         instruct=req.instruct or "",
@@ -479,18 +318,11 @@ async def tts_endpoint(req: TTSRequest):
 
 @generation_router.post("/asr", response_model=ASRResponse)
 async def asr_endpoint(req: ASRRequest):
-    """语音识别端点（非流式）"""
     audio_bytes = base64.b64decode(req.audio_base64)
     return await generate_asr_response(audio_bytes, req.language if req.language else None)
 
 
 def register_tts_asr_routes(app, include_generation_routes: bool = True):
-    """注册 TTS/ASR 路由到 FastAPI 应用"""
     app.include_router(meta_router, prefix="/v1/tts-asr")
     if include_generation_routes:
         app.include_router(generation_router, prefix="/v1/tts-asr")
-
-
-router = APIRouter()
-router.include_router(meta_router)
-router.include_router(generation_router)

src/AGENTS.md

@@ -52,7 +52,7 @@
 	- 上传图片和文档
 	- 触发 OCR（通过 **OCRImageWrapper.vue**）
 	- 导入导出 Markdown
-	- 导出 DOCX 和 PDF
+	- 导出 DOCX 和 PDF（纯前端，从展开后的编辑器 HTML 导出）
 	- AI 开关
 	- 32 KB 大小限制
 	- TTS 菜单和播放器
@@ -68,7 +68,7 @@
 	- debounceMs
 	- privacyMode
 	- language
-	- currency
+	- country
 	- backgroundType
 	- backgroundImage
 	- backgroundOpacity
@@ -85,7 +85,7 @@
 	- 生成 request_id
 	- 绑定 AbortSignal
 	- 触发中止时的 cancel 请求
-	- 读取 settings store 中的 thinking、privacy、language、currency、timezone 信息
+	- 读取 settings store 中的 thinking、privacy、language、country、timezone 信息
 	- 向后端发 POST 并读取 JSON
 
 ## 当前真实约定
@@ -99,9 +99,12 @@
 ## 容易踩坑的点
 
 - 文档超过 32 KB 时，AI 补全会被禁用。
+- AI 开关需要同时作用到主编辑器、文档块嵌套编辑器、网页搜索块嵌套编辑器；当前通过 `llm-in-text:copilot-toggle` 广播同步。
 - 在文档块、Mermaid、LaTeX 等特定上下文中，部分 AI 行为和上传行为会被禁用或改道。
 - OCR 文本和文档块摘录会被注入补全上下文，但这些内容不应直接作为用户可见输出回写到文档。
-- 前端存在 /v1/export/pdf 调用，但调试前先确认后端是否真的实现了这个端点。
+- DOCX/PDF 导出当前不再依赖 `/v1/export/pdf`；导出逻辑看 `utils/richExport.js`。
+- 上传单文件限制统一为 100MB；图片上传会等待 OCR 完成并给出全局进度与失败反馈。
+- 视频上传会调用 `/v1/ocr` 的 `media_type=video` 路径，后端返回合并后的 OCR/ASR 文本，再插入文档块。
 - **Web Search 块使用 ```llm-websearch fenced code 语法，触发词为 [WEBSEARCH]（不区分大小写）。**
 - **验证码组件依赖 vue3-captcha 库，刷新和验证逻辑需保持与后端 /captcha/generate 和 /captcha/verify 接口同步。**
 - 当前前端多处仍保留占位 API Key 或默认值；不要把这种写法继续扩散到新代码。

src/components/DocBlockCrepe.vue

@@ -56,6 +56,7 @@ import { hiddenTextInputPlugin, hiddenTextNode, hiddenTextRemark, hiddenTextView
 import { fetchSuggestion, submitCompress, pollCompressStatus } from '../utils/api.js'
 import { isDocumentVisible, getRecommendedDebounce, getRecommendedSyncInterval } from '../composables/useVisibility.js'
 
+const COPILOT_TOGGLE_EVENT = 'llm-in-text:copilot-toggle'
 const props = defineProps({
   docType: { type: String, default: 'txt' },
   docName: { type: String, default: 'document.txt' },
@@ -77,6 +78,7 @@ let crepe = null
 let syncTimer = null
 let syncingExternal = false
 let compressPoller = null
+let copilotToggleHandler = null
 
 const handleCompress = () => {
   if (compressState.value !== 'idle') return
@@ -233,8 +235,26 @@ onMounted(async () => {
 
   crepe.editor.action((ctx) => {
     const view = ctx.get(editorViewCtx)
-    setCopilotEnabled(view, true)
+    const enabled = typeof window !== 'undefined'
+      ? window.__LLM_IN_TEXT_COPILOT_ENABLED__ !== false
+      : true
+    setCopilotEnabled(view, enabled)
+    if (!enabled) {
+      clearGhostSuggestion(view)
+    }
   })
+
+  copilotToggleHandler = (event) => {
+    const enabled = Boolean(event?.detail?.enabled)
+    crepe?.editor?.action((ctx) => {
+      const view = ctx.get(editorViewCtx)
+      setCopilotEnabled(view, enabled)
+      if (!enabled) {
+        clearGhostSuggestion(view)
+      }
+    })
+  }
+  window.addEventListener(COPILOT_TOGGLE_EVENT, copilotToggleHandler)
 })
 
 onUnmounted(() => {
@@ -246,6 +266,10 @@ onUnmounted(() => {
     compressPoller.stop()
     compressPoller = null
   }
+  if (copilotToggleHandler) {
+    window.removeEventListener(COPILOT_TOGGLE_EVENT, copilotToggleHandler)
+    copilotToggleHandler = null
+  }
   if (crepe) {
     crepe.editor.action((ctx) => {
       const view = ctx.get(editorViewCtx)
@@ -444,4 +468,3 @@ onUnmounted(() => {
   display: none !important;
 }
 </style>
-

src/components/InputBlockCrepe.vue

@@ -0,0 +1,223 @@
+<template>
+  <section class="input-block-shell" :class="[`is-${stage}`, { 'has-user-text': Boolean(userText) }]">
+    <!-- Idle state: compact trigger bar -->
+    <button
+      v-if="isIdle && !hasUserText"
+      type="button"
+      class="input-block-trigger-bar"
+      @mousedown.stop.prevent
+    >
+      <span class="input-badge">INPUT</span>
+      <Transition name="input-fade" mode="out-in">
+        <template v-if="displayInstruction.trim()">
+          <span class="input-label">{{ displayInstruction }}</span>
+        </template>
+      </Transition>
+    </button>
+
+    <!-- Active panel -->
+    <div v-else class="input-block-panel">
+      <div class="input-header-row" @click.stop>
+        <!-- Instructions display -->
+        <span v-if="displayInstruction.trim()" class="input-instruction">{{ displayInstruction }}</span>
+      </div>
+
+      <Transition name="input-slide" appear>
+        <!-- Input controls -->
+        <div class="input-controls-row">
+          <textarea
+            ref="textAreaRef"
+            :value="userText || ''"
+            class="input-textarea"
+            @input.stop.prevent="handleTextInput"
+          ></textarea>
+
+          <!-- ASR voice recording controls -->
+          <div class="input-voice-group">
+            <button 
+              type="button" 
+              :class="['input-voice-btn', { 'is-recording': isRecording }]"
+              @mousedown.stop.prevent 
+              :title="t('inputVoiceMode') || '语音模式'"
+            >
+              <span class="voice-icon">{{ isRecording ? '⏹️' : '🎤' }}</span>
+            </button>
+
+            <Transition name="input-fade" mode="out-in">
+              <!-- Accept button -->
+              <button 
+                v-if="hasUserText" 
+                type="button" 
+                class="input-accept-btn"
+                @mousedown.stop.prevent 
+                :disabled="isTranscribing || isSubmittingText"
+                @click.stop="handleAcceptInput"
+              >
+                <span class="accept-icon">✓</span>
+              </button>
+
+              <!-- Transcribing indicator -->
+              <Transition name="input-fade" mode="out-in">
+                <div v-if="isTranscribing && !hasUserText" class="input-transcribing">
+                  <span>{{ t('transcribing') || '转录中...' }}</span>
+                </div>
+
+                <!-- Error display -->
+                <Transition name="input-fade" mode="out-in">
+                  <div v-if="transcriptionError && !hasUserText" class="input-error">
+                    {{ t('error') || '错误' }}: {{ transcriptionError }}</div>
+
+                  <!-- Placeholder hint -->
+                  <Transition name="input-fade" mode="out-in">
+                    <div v-if="!hasUserText && !isTranscribing" class="input-placeholder">
+                      {{ t('inputPlaceholder') || '（请输入文本）' }}</div>
+
+                    <!-- Status indicators -->
+                    <Transition name="input-fade" mode="out-in">
+                      <!-- Recording indicator -->
+                      <div v-if="isRecording && !hasUserText" class="input-recording-indicator">
+                        <span>{{ t('recording') || '录音中...' }}</div>
+
+                      <!-- Transcribing indicator -->
+                      <Transition name="input-fade" mode="out-in">
+                        <div v-if="isTranscribing && !hasUserText" class="input-transcribing-indicator">
+                          <span>{{ t('transcribing') || '转录中...' }}</div>
+
+                      <!-- Accepted state -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- When text is entered but not yet accepted, show "接受" button -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-accept-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleAcceptInput"
+                        >
+                          {{ t('accept') || '接受' }}
+                        </button>
+
+                      <!-- After accept, show "重录" (re-record) and "重试" buttons -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Re-record button (only when there's text) -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-rerecord-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleRerecordInput"
+                        >
+                          {{ t('inputRetry') || '重录' }}
+                        </button>
+
+                      <!-- After accept, show "重试" (retry) button -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Retry button (only when there's text) -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-retry-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleRetryInput"
+                        >
+                          {{ t('inputRedo') || '重试' }}
+                        </button>
+
+                      <!-- After accept, show version controls -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Version controls (only when there's text) -->
+                        <div v-if="hasUserText && !isTranscribing" class="input-version-controls">
+                          <span>{{ activeVersion + 1 }} / {{ versionCount }}</div>
+
+                      <!-- After accept, show "接受" button -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- When text is entered but not yet accepted, show "接受" button -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-accept-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleAcceptInput"
+                        >
+                          {{ t('accept') || '接受' }}
+                        </button>
+
+                      <!-- After accept, show "重录" (re-record) and "重试" buttons -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Re-record button (only when there's text) -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-rerecord-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleRerecordInput"
+                        >
+                          {{ t('inputRetry') || '重录' }}
+                        </button>
+
+                      <!-- After accept, show "重试" (retry) button -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Retry button (only when there's text) -->
+                        <button 
+                          v-if="hasUserText && !isTranscribing" 
+                          type="button"
+                          class="input-retry-btn"
+                          @mousedown.stop.prevent 
+                          :disabled="isTranscribing || isSubmittingText"
+                          @click.stop="handleRetryInput"
+                        >
+                          {{ t('inputRedo') || '重试' }}
+                        </button>
+
+                      <!-- After accept, show version controls -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Version controls (only when there's text) -->
+                        <div v-if="hasUserText && !isTranscribing" class="input-version-controls">
+                          <span>{{ activeVersion + 1 }} / {{ versionCount }}</div>
+
+                      <!-- Status line -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Status line (only when there's text) -->
+                        <div v-if="hasUserText" class="input-status-line">
+                          <!-- Status line content -->
+
+                        </div>
+
+                      <!-- Footer row with action buttons and status indicators -->
+                      <Transition name="input-fade" mode="out-in">
+                        <!-- Footer row (only when there's text) -->
+
+                      </Transition>
+                    </div>
+
+                  <!-- Clear textarea on accept to allow new input -->
+                  <Transition name="input-fade" mode="out-in">
+
+                </div>
+
+              <!-- Error display -->
+              <Transition name="input-fade" mode="out-in">
+
+            </div>
+
+          <!-- Placeholder hint when no text -->
+          <Transition name="input-fade" mode="out-in">
+
+        </div>
+
+      <!-- Status indicators row -->
+      <Transition name="input-fade" mode="out-in">
+
+    </div>
+  )} else {
+        // No text yet → show placeholder + voice mode button only  
+      }
+
+    <!-- Footer row -->
+  </Transition>
+</section>INPUT_BLOCK_CREPE_EOF
+echo "Created InputBlockCrepe.vue component" && wc -l /Users/allenyuan/llm-in-text/src/components/InputBlockCrepe.vue

src/components/MilkdownEditor.vue

@@ -1,5 +1,5 @@
 <template>
-  <div class="editor-container">
+  <div class="editor-container" :style="editorTypographyStyle">
     <div ref="root" class="milkdown-editor"></div>
 
     <div class="history-buttons">
@@ -86,8 +86,8 @@
         </button>
         <div v-if="showExportDropdown" class="export-dropdown">
           <button type="button" @click="() => { exportMarkdown(); showExportDropdown = false; }">{{ t('exportMd') }}</button>
-          <button type="button" class="disabled-export" :title="t('exportDisabledHint')" @click="showExportDisabled">{{ t('exportDocx') }}</button>
-          <button type="button" class="disabled-export" :title="t('exportDisabledHint')" @click="showExportDisabled">{{ t('exportPdf') }}</button>
+          <button type="button" @click="() => { exportDocx(); showExportDropdown = false; }">{{ t('exportDocx') }}</button>
+          <button type="button" @click="() => { exportPdf(); showExportDropdown = false; }">{{ t('exportPdf') }}</button>
         </div>
       </div>
       
@@ -320,13 +320,14 @@ import { fetchProSuggestionStream, fetchSuggestion, fetchTTS, fetchWebSearchStre
 import { useSettingsStore } from '../stores/settings'
 import { useTemplatesStore } from '../stores/templates'
 import { useTheme } from '../composables/useTheme.js'
-import { OCR_URL, EXPORT_PDF_URL } from '../utils/config.js'
+import { OCR_URL } from '../utils/config.js'
 import TTSMenu from './TTSMenu.vue'
 import TTSPlayer from './TTSPlayer.vue'
-import { convertFileToMarkdown, convertAudioToText } from '../utils/convert.js'
-import { setOcrCache, clearOcrCache, clearAllOcrCache, IMAGE_SIZE_LIMIT, calculateImageHash, getOcrByHash, setOcrByHash } from '../utils/ocrCache.js'
+import { convertFileToMarkdown, convertAudioToText, convertVideoToText } from '../utils/convert.js'
+import { exportEditorToDocxBlob, exportEditorToPdf } from '../utils/richExport.js'
+import { setOcrCache, clearOcrCache, clearAllOcrCache, IMAGE_SIZE_LIMIT, calculateImageHash, getOcrByHash, setOcrByHash, setOcrState, OcrStatus } from '../utils/ocrCache.js'
 import { isDocumentVisible, getRecommendedDebounce, getRecommendedSyncInterval } from '../composables/useVisibility.js'
-import { DOC_BLOCK_NODE_TYPE, getDocTypeFromFilename, isSupportedDocFile, transformDocBlockMarkdownForClipboard, transformLegacyDocBlocksForExport, transformSpecialDocBlocksToLegacy, isAudioFile } from '../utils/docBlock.js'
+import { DOC_BLOCK_NODE_TYPE, getDocTypeFromFilename, isSupportedDocFile, transformDocBlockMarkdownForClipboard, transformLegacyDocBlocksForExport, transformSpecialDocBlocksToLegacy, isAudioFile, isVideoFile } from '../utils/docBlock.js'
 import { isUploadBlockTypeAllowed } from '../utils/uploadBlock.js'
 import { WEB_SEARCH_NODE_TYPE } from '../utils/webSearch.js'
 
@@ -336,6 +337,7 @@ const templateStore = useTemplatesStore()
 const { isDark } = useTheme()
 const t = (key) => settings.t[key]
 const initialMarkdown = computed(() => settings.initialMarkdown)
+const COPILOT_TOGGLE_EVENT = 'llm-in-text:copilot-toggle'
 
 const root = ref(null)
 const uploadInputRef = ref(null)
@@ -416,9 +418,8 @@ const CONVERT_MIME_TYPES = new Set([
   'application/pdf',
 ])
 const VIDEO_MIME_TYPES = new Set(['video/mp4', 'video/webm', 'video/quicktime', 'video/x-msvideo', 'video/x-matroska'])
-const MAX_VIDEO_SIZE = 200 * 1024 * 1024 // 200MB
 const MAX_UPLOAD_BATCH = 10
-const MAX_UPLOAD_FILE_SIZE = 50 * 1024 * 1024
+const MAX_UPLOAD_FILE_SIZE = 100 * 1024 * 1024
 let lastInitialMarkdown = transformSpecialDocBlocksToLegacy(initialMarkdown.value)
 
 const normalizeTrailingWhitespace = (value) => (value || '').replace(/\s+$/, '')
@@ -572,34 +573,6 @@ const toggleTemplateDropdown = () => {
   }
 }
 
-const buildDocxBlob = async (markdown) => {
-  const { Document, Packer, Paragraph, HeadingLevel } = await import('docx')
-  const children = []
-  const normalizedMarkdown = String(markdown || '').replace(/\r\n?/g, '\n')
-
-  for (const line of normalizedMarkdown.split('\n')) {
-    if (line.startsWith('# ')) {
-      children.push(new Paragraph({ text: line.slice(2), heading: HeadingLevel.HEADING_1 }))
-      continue
-    }
-    if (line.startsWith('## ')) {
-      children.push(new Paragraph({ text: line.slice(3), heading: HeadingLevel.HEADING_2 }))
-      continue
-    }
-    if (line.startsWith('### ')) {
-      children.push(new Paragraph({ text: line.slice(4), heading: HeadingLevel.HEADING_3 }))
-      continue
-    }
-    if (line.startsWith('---')) {
-      children.push(new Paragraph({ text: '----------' }))
-      continue
-    }
-    children.push(line.trim() === '' ? new Paragraph({}) : new Paragraph({ text: line }))
-  }
-
-  return Packer.toBlob(new Document({ sections: [{ properties: {}, children }] }))
-}
-
 const downloadBlob = (blob, filename) => {
   const url = URL.createObjectURL(blob)
   const anchor = document.createElement('a')
@@ -744,9 +717,15 @@ const clearCurrentSuggestion = (view) => {
 const clearCurrentGhost = () => {
   if (!crepe) return
   crepe.editor.action((ctx) => {
-const view = ctx.get(editorViewCtx)
-clearGhostSuggestion(view)
-})
+    const view = ctx.get(editorViewCtx)
+    clearGhostSuggestion(view)
+  })
+}
+
+const broadcastCopilotEnabled = (enabled) => {
+  if (typeof window === 'undefined') return
+  window.__LLM_IN_TEXT_COPILOT_ENABLED__ = Boolean(enabled)
+  window.dispatchEvent(new CustomEvent(COPILOT_TOGGLE_EVENT, { detail: { enabled: Boolean(enabled) } }))
 }
 
 const updateHistoryState = (view) => {
@@ -1059,42 +1038,59 @@ const consumeSseResult = async (res) => {
 }
 
 const performOCR = async (file, cacheKey, imageHash = '') => {
-  if (!aiEnabled.value) return
-  
-  const reader = new FileReader()
-  reader.onload = async () => {
-    const dataUrl = typeof reader.result === 'string' ? reader.result : ''
-    const splitIndex = dataUrl.indexOf(',')
-    if (splitIndex === -1) return
+  if (!aiEnabled.value) return ''
 
-    const base64 = dataUrl.slice(splitIndex + 1)
-    try {
-      const headers = {
-        'Content-Type': 'application/json',
-      }
-      const res = await fetch(OCR_URL, {
-        method: 'POST',
-        headers,
-        credentials: 'include',
-        body: JSON.stringify({ 
-          image: base64, 
-          filename: file.name,
-          language: 'auto'
-        })
-      })
-      const data = await consumeSseResult(res)
-      if (data.text) {
-        setOcrCache(cacheKey, data.text)
-        setOcrCache(file.name, data.text)
-        if (imageHash) {
-          setOcrByHash(imageHash, data.text)
-        }
-      }
-    } catch {
-      // OCR error, ignore
-    }
+  if (imageHash) {
+    setOcrState(imageHash, OcrStatus.LOADING)
+  }
+
+  const dataUrl = await new Promise((resolve, reject) => {
+    const reader = new FileReader()
+    reader.onload = () => resolve(typeof reader.result === 'string' ? reader.result : '')
+    reader.onerror = () => reject(reader.error || new Error('读取图片失败'))
+    reader.readAsDataURL(file)
+  })
+
+  const splitIndex = dataUrl.indexOf(',')
+  if (splitIndex === -1) {
+    throw new Error('图片编码失败')
+  }
+
+  const base64 = dataUrl.slice(splitIndex + 1)
+  const headers = {
+    'Content-Type': 'application/json',
+  }
+
+  try {
+    const res = await fetch(OCR_URL, {
+      method: 'POST',
+      headers,
+      credentials: 'include',
+      body: JSON.stringify({
+        image: base64,
+        filename: file.name,
+        language: 'auto',
+        media_type: 'image',
+        mime_type: file.type || 'image/png',
+      }),
+    })
+    const data = await consumeSseResult(res)
+    if (!data?.text) {
+      throw new Error('OCR 返回结果为空')
+    }
+    setOcrCache(cacheKey, data.text)
+    setOcrCache(file.name, data.text)
+    if (imageHash) {
+      setOcrByHash(imageHash, data.text)
+      setOcrState(imageHash, OcrStatus.SUCCESS, data.text)
+    }
+    return data.text
+  } catch (error) {
+    if (imageHash) {
+      setOcrState(imageHash, OcrStatus.FAILED, '', error?.message || 'OCR 识别失败')
+    }
+    throw error
   }
-  reader.readAsDataURL(file)
 }
 
 const prepareImageFile = async (file) => {
@@ -1116,10 +1112,11 @@ const prepareImageFile = async (file) => {
   const imageHash = await calculateImageHash(imageBytes)
   const existingOcr = getOcrByHash(imageHash)
   if (!existingOcr) {
-    performOCR(file, objectUrl, imageHash)
+    await performOCR(file, objectUrl, imageHash)
   } else {
     setOcrCache(objectUrl, existingOcr)
     setOcrCache(file.name, existingOcr)
+    setOcrState(imageHash, OcrStatus.SUCCESS, existingOcr)
   }
 
   return objectUrl
@@ -1183,12 +1180,14 @@ const validateUploadFiles = (files, options = {}) => {
   return true
 }
 
-const parseDocFilesToBlocks = async (docFiles) => {
+const parseDocFilesToBlocks = async (docFiles, progressOptions = {}) => {
   if (docFiles.length === 0) {
     return { blocksToInsert: [], errors: [] }
   }
 
-  uploadProgress.value = { current: 0, total: docFiles.length, filename: '' }
+  const total = Number.isFinite(progressOptions.total) ? progressOptions.total : docFiles.length
+  const offset = Number.isFinite(progressOptions.offset) ? progressOptions.offset : 0
+  uploadProgress.value = { current: offset, total, filename: '' }
 
   const results = []
   const errors = []
@@ -1203,6 +1202,8 @@ const parseDocFilesToBlocks = async (docFiles) => {
       content = await convertFileToMarkdown(file)
     } else if (isAudioFile(file)) {
       content = await convertAudioToText(file)
+    } else if (isVideoFile(file)) {
+      content = await convertVideoToText(file)
     } else {
       throw new Error('不支持的文件类型')
     }
@@ -1222,7 +1223,7 @@ const parseDocFilesToBlocks = async (docFiles) => {
   const settled = await Promise.allSettled(parsePromises)
 
   settled.forEach((result, idx) => {
-    uploadProgress.value = { current: idx + 1, total: docFiles.length, filename: docFiles[idx].name }
+    uploadProgress.value = { current: offset + idx + 1, total, filename: docFiles[idx].name }
     if (result.status === 'fulfilled') {
       results.push(result.value)
     } else {
@@ -1230,9 +1231,6 @@ const parseDocFilesToBlocks = async (docFiles) => {
       errors.push({ filename: docFiles[idx].name, message })
     }
   })
-
-  uploadProgress.value = null
-
   results.sort((a, b) => a.index - b.index)
 
   return {
@@ -1258,16 +1256,28 @@ const processUploadedFiles = async (files, options = {}) => {
   const docFiles = normalizedFiles.filter((file) => !isImageFile(file))
   const replaceRange = options.replaceRange || null
   let rangeConsumed = false
+  let progressIndex = 0
+  const totalCount = normalizedFiles.length
 
   for (const file of imageFiles) {
-    const objectUrl = await prepareImageFile(file)
-    if (!objectUrl) continue
-    insertImageAtCursor(objectUrl, !rangeConsumed ? replaceRange : null)
-    rangeConsumed = rangeConsumed || Boolean(replaceRange)
+    uploadProgress.value = { current: progressIndex + 1, total: totalCount, filename: `${file.name} · OCR` }
+    try {
+      const objectUrl = await prepareImageFile(file)
+      if (!objectUrl) continue
+      insertImageAtCursor(objectUrl, !rangeConsumed ? replaceRange : null)
+      rangeConsumed = rangeConsumed || Boolean(replaceRange)
+    } catch (error) {
+      alert(`${file.name} OCR 失败：${error.message || '未知错误'}`)
+    } finally {
+      progressIndex += 1
+    }
   }
 
   if (docFiles.length > 0) {
-    const { blocksToInsert, errors } = await parseDocFilesToBlocks(docFiles)
+    const { blocksToInsert, errors } = await parseDocFilesToBlocks(docFiles, {
+      offset: progressIndex,
+      total: totalCount,
+    })
     if (blocksToInsert.length > 0) {
       insertMultipleDocBlocks(blocksToInsert, !rangeConsumed ? replaceRange : null)
       rangeConsumed = rangeConsumed || Boolean(replaceRange)
@@ -1278,8 +1288,11 @@ const processUploadedFiles = async (files, options = {}) => {
       const errorMsgs = errors.map((item) => `${item.filename}: ${item.message}`).join('\n')
       alert(`上传失败 ${failCount} 个文件：\n\n${errorMsgs}`)
     }
+    progressIndex += docFiles.length
   }
 
+  uploadProgress.value = null
+
   return true
 }
 
@@ -1478,6 +1491,7 @@ crepe = new Crepe({
         }
         document.addEventListener('mousedown', ttsClickOutsideHandler)
     })
+    broadcastCopilotEnabled(aiEnabled.value)
     scheduleMarkdownSync()
 })
 
@@ -1493,15 +1507,20 @@ const exportMarkdown = async () => {
     }
 }
 
-const showExportDisabled = () => {
-    alert(t('exportDisabledHint') || 'DOCX/PDF 导出暂不可用。')
+const resolveEditorExportElement = () => {
+    const editorShell = root.value?.querySelector('.milkdown-editor .milkdown')
+      || root.value?.querySelector('.milkdown-editor')
+      || root.value
+    if (!(editorShell instanceof HTMLElement)) {
+        throw new Error('未找到可导出的编辑器内容')
+    }
+    return editorShell
 }
 
 const exportDocx = async () => {
     try {
-        console.log('Exporting DOCX...')
-        const markdown = await getExportMarkdown()
-        const blob = await buildDocxBlob(markdown)
+        console.log('Exporting DOCX from expanded editor HTML...')
+        const blob = await exportEditorToDocxBlob(resolveEditorExportElement())
         const exportName = createExportName()
         downloadBlob(blob, `${exportName}.docx`)
         console.log('DOCX export completed')
@@ -1513,28 +1532,10 @@ const exportDocx = async () => {
 
 const exportPdf = async () => {
     try {
-        console.log('Exporting PDF via DOCX...')
-        const markdown = await getExportMarkdown()
-        const docxBlob = await buildDocxBlob(markdown)
+        console.log('Exporting PDF from expanded editor HTML...')
         const exportName = createExportName()
-        const formData = new FormData()
-        formData.append('file', docxBlob, `${exportName}.docx`)
-
-        const res = await fetch(EXPORT_PDF_URL, {
-            method: 'POST',
-            credentials: 'include',
-            body: formData,
-        })
-
-        if (!res.ok) {
-            const errorText = await res.text()
-            throw new Error(`HTTP ${res.status}: ${errorText}`)
-        }
-
-        const pdfBlob = await res.blob()
-        downloadBlob(pdfBlob, `${exportName}.pdf`)
+        await exportEditorToPdf(resolveEditorExportElement(), `${exportName}.pdf`)
         console.log('PDF export completed successfully')
-        alert('PDF导出成功！')
     } catch (error) {
         console.error('PDF export failed:', error)
         alert(`PDF导出失败: ${error.message}`)
@@ -1592,9 +1593,12 @@ const toggleAI = async () => {
         const view = ctx.get(editorViewCtx)
         setCopilotEnabled(view, aiEnabled.value)
         if (!aiEnabled.value) {
-            clearCurrentSuggestion(view)
+            interruptCopilot(view)
+            clearGhostSuggestion(view)
         }
     })
+
+    broadcastCopilotEnabled(aiEnabled.value)
 }
 
 const toggleExportDropdown = () => {
@@ -1839,7 +1843,7 @@ for (const url of Array.from(objectUrls)) {
     align-items: center;
     justify-content: center;
     box-shadow: var(--panel-shadow);
-    opacity: 0.5;
+    opacity: 0.8;
 }
 
 .action-btn:hover {
@@ -2051,16 +2055,6 @@ for (const url of Array.from(objectUrls)) {
     background: var(--crepe-color-hover);
 }
 
-.export-dropdown button.disabled-export {
-    color: var(--muted-text);
-    cursor: not-allowed;
-    opacity: 0.58;
-}
-
-.export-dropdown button.disabled-export:hover {
-    background: transparent;
-}
-
 .template-dropdown {
   position: absolute;
   top: calc(100% + 8px);
@@ -2075,6 +2069,7 @@ for (const url of Array.from(objectUrls)) {
   z-index: 100000;
   display: flex;
   flex-direction: column;
+  opacity: 1 !important;
 }
 
 .template-dropdown-section {

src/components/SettingsPanel.vue

@@ -335,21 +335,22 @@ const validateCaptcha = () => {
             </select>
           </div>
           
-           <div class="form-group">
-            <label>{{ t('currency') }}</label>
-            <select v-model="store.currency" class="select-input" :disabled="store.privacyMode">
+          <div class="form-group">
+            <label>{{ t('country') }}</label>
+            <select v-model="store.country" class="select-input" :disabled="store.privacyMode">
               <option value="auto">{{ t('auto') }}</option>
-              <option value="CNY">CNY (¥)</option>
-              <option value="USD">USD ($)</option>
-              <option value="EUR">EUR (€)</option>
-              <option value="JPY">JPY (¥)</option>
-              <option value="KRW">KRW (₩)</option>
-              <option value="GBP">GBP (£)</option>
-              <option value="AUD">AUD ($)</option>
-              <option value="CAD">CAD ($)</option>
+              <option value="CN">中国</option>
+              <option value="US">United States</option>
+              <option value="JP">日本</option>
+              <option value="KR">대한민국</option>
+              <option value="GB">United Kingdom</option>
+              <option value="DE">Deutschland</option>
+              <option value="FR">France</option>
+              <option value="CA">Canada</option>
+              <option value="AU">Australia</option>
             </select>
-</div>
-</section>
+          </div>
+        </section>
 
 <!-- Security & Captcha Section -->
 <section class="settings-section" v-if="showCaptcha">

src/components/WebSearchBlockCrepe.vue

@@ -95,6 +95,7 @@ import { hiddenTextInputPlugin, hiddenTextNode, hiddenTextRemark, hiddenTextView
 import { fetchSuggestion, submitCompress, pollCompressStatus } from '../utils/api.js'
 import { isDocumentVisible, getRecommendedDebounce, getRecommendedSyncInterval } from '../composables/useVisibility.js'
 
+const COPILOT_TOGGLE_EVENT = 'llm-in-text:copilot-toggle'
 const props = defineProps({
   stage: { type: String, default: 'idle' },
   progressText: { type: String, default: '' },
@@ -119,6 +120,7 @@ let crepe = null
 let syncTimer = null
 let syncingExternal = false
 let compressPoller = null
+let copilotToggleHandler = null
 
 const ensureEditor = async () => {
   if (crepe || !editorRoot.value || !hasContent.value) return
@@ -166,8 +168,26 @@ const ensureEditor = async () => {
 
   crepe.editor.action((ctx) => {
     const view = ctx.get(editorViewCtx)
-    setCopilotEnabled(view, true)
+    const enabled = typeof window !== 'undefined'
+      ? window.__LLM_IN_TEXT_COPILOT_ENABLED__ !== false
+      : true
+    setCopilotEnabled(view, enabled)
+    if (!enabled) {
+      clearGhostSuggestion(view)
+    }
   })
+
+  copilotToggleHandler = (event) => {
+    const enabled = Boolean(event?.detail?.enabled)
+    crepe?.editor?.action((ctx) => {
+      const view = ctx.get(editorViewCtx)
+      setCopilotEnabled(view, enabled)
+      if (!enabled) {
+        clearGhostSuggestion(view)
+      }
+    })
+  }
+  window.addEventListener(COPILOT_TOGGLE_EVENT, copilotToggleHandler)
 }
 
 const hasContent = computed(() => Boolean((props.content || '').trim()))
@@ -277,6 +297,10 @@ onUnmounted(() => {
     compressPoller.stop()
     compressPoller = null
   }
+  if (copilotToggleHandler) {
+    window.removeEventListener(COPILOT_TOGGLE_EVENT, copilotToggleHandler)
+    copilotToggleHandler = null
+  }
   if (crepe) {
     crepe.editor.action((ctx) => {
       const view = ctx.get(editorViewCtx)

src/plugins/docBlockPlugin.ts

@@ -177,13 +177,6 @@ class DocBlockNodeView implements NodeView {
     this.view.focus()
   }
 
-  destroy(hasNewHeader: boolean, hasNewFooter: boolean) {
-    if (this.app) {
-      this.app.unmount()
-      this.app = null
-    }
-  }
-
   compressWithNewContent(compressedContent: string) {
     const pos = this.getPosValue()
     if (pos === undefined) return

src/plugins/inputBlockPlugin.ts

@@ -0,0 +1,74 @@
+import { createApp, reactive } from 'vue'
+import { remarkCtx, schemaCtx, serializerCtx, type Ctx } from '@milkdown/kit/core'
+import { $ctx, $node, $prose, $remark, $view } from '@milkdown/kit/utils'
+import { Plugin, PluginKey } from '@milkdown/prose/state'
+import { type Node as ProseNode, Slice } from '@milkdown/prose/model'
+import { Decoration, DecorationSet, type EditorView, type NodeView } from '@milkdown/prose/view'
+import { ParserState } from '@milkdown/transformer'
+import InputBlockCrepe from '../components/InputBlockCrepe.vue'
+import { 
+  INPUT_BLOCK_NODE_TYPE, 
+  serializeInputBlockSyntax, 
+} from '../utils/inputBlock.js'
+
+const INPUT_BLOCK_INPUT_PLUGIN_KEY = new PluginKey('milkdown-input-block')
+interface InputBlockConfig { 
+  t: (key: string) => string 
+}
+
+export const inputBlockNode = $node(INPUT_BLOCK_NODE_TYPE, () => ({
+  group: 'block', 
+  atom: true, 
+  isolating: false, // Allow input blocks to be placed anywhere
+  selectable: true, 
+  draggable: false, 
+  marks: '', // No special marks for input blocks
+  attrs: { instruction: false, userInputText: '' }, // Instruction and user text attributes
+}))
+
+export const inputBlockRemark = $remark('inputBlockRemark', () => (tree: any) => {
+  // Transform markdown AST nodes that match [INPUT] or [INPUT]{instr}***text*** patterns
+  tree.children = (tree.children || []).map((child: any) => {
+    // Handle fenced code blocks with language "llm-input" or similar syntaxes  
+  })
+})
+
+export const inputBlockView = $view(inputBlockNode, (ctx) => {
+  return (_node: any, _getPos?: () => number | undefined): NodeView => {
+    return new InputBlockNodeView(_node, _getPos)  
+  }
+})
+
+export const inputBlockInputPlugin = $prose(() => {
+  return new Plugin({ 
+    key: INPUT_BLOCK_INPUT_PLUGIN_KEY,  
+    props: {
+      handleTextInput(view, _from, to, text) { 
+        return tryHandleInputTriggerText(text)  
+      },
+    } as any, // ProseMirror plugin props type is incomplete in Milkdown types  
+  })
+})
+
+export const inputBlockHighlightPlugin = $prose(() => { 
+  return new Plugin<DecorationSet>({
+    key: INPUT_BLOCK_INPUT_PLUGIN_KEY,  
+    state: { 
+      init() { return DecorationSet.empty },
+      apply(tr) { // Transaction mapping for decorations  
+        const meta = tr.getMeta(INPUT_BLOCK_INPUT_PLUGIN_KEY)
+      }, 
+    } as any, // DecorationSet state type is complex in Milkdown  
+  })
+})
+
+export function insertInputBlockAtSelection(view: EditorView, autoStart = false) {
+  return view.dispatch( // Actually inserts the node at cursor position  
+    view.state.tr.replaceWith(
+      to, 
+      inputBlockNode.create({ instruction: '', userInputText: '' }),  
+    ) as any, // ReplaceWith returns unknown in Milkdown types
+  ).then(() => {}).catch(console.error)  
+}INPUT_PLUGIN_EOF
+
+echo "Created inputBlockPlugin.ts" && wc -l /Users/allenyuan/llm-in-text/src/plugins/inputBlockPlugin.ts

src/plugins/webSearchBlockPlugin.ts

@@ -412,21 +412,6 @@ class WebSearchBlockNodeView implements NodeView {
             return
           }
           this.setStage(event, String(data?.message || ''))
-        },        onDelta: (data) => {
-          if (this.destroyed || this.requestSeq !== requestSeq) return
-          const text = String(data?.text || '')
-          if (!text) return
-          streamedContent += text
-          this.updateAttrs({ content: streamedContent })
-        },        onDelta: (data) => {
-          if (this.destroyed || this.requestSeq !== requestSeq) return
-          const text = String(data?.text || '')
-          if (!text) return
-          streamedContent += text
-          this.updateAttrs({ content: streamedContent }
-            return
-          }
-          this.setStage(event, String(data?.message || ''))
         },
         onDelta: (data) => {
           if (this.destroyed || this.requestSeq !== requestSeq) return
@@ -685,4 +670,3 @@ export const webSearchBlockInputPlugin = $prose(() => {
     },
   })
 })
-

src/stores/settings.js

@@ -13,11 +13,13 @@ export const useSettingsStore = defineStore('settings', () => {
     const proThinking = ref('medium')     // 'low' | 'medium' | 'high'
     const privacyMode = ref(true)
     const language = ref('auto')
-    const currency = ref('auto')
+    const country = ref('auto')
     const backgroundType = ref('default') // 'default' | 'warm' | 'reading' | 'image'
     const backgroundImage = ref('')
     const backgroundOpacity = ref(0.2)    // 0.05 - 0.50
     const ttsInstruct = ref('')
+    const fontFamily = ref('system')      // 'system' | 'serif' | 'monospace'
+    const fontSize = ref(16)              // 12 - 32
 
     // --- Computed getters (derived from reactive state) ---
     const uiLanguage = computed(() => {
@@ -43,12 +45,15 @@ export const useSettingsStore = defineStore('settings', () => {
             if (data.proThinking) proThinking.value = data.proThinking
             if (typeof data.privacyMode === 'boolean') privacyMode.value = data.privacyMode
             if (data.language) language.value = data.language
-            if (data.currency) currency.value = data.currency
+            if (data.country) country.value = data.country
+            else if (data.currency) country.value = data.currency
             if (data.backgroundType === 'color') backgroundType.value = 'default'
             else if (data.backgroundType) backgroundType.value = data.backgroundType
             if (data.backgroundImage) backgroundImage.value = data.backgroundImage
             if (data.backgroundOpacity) backgroundOpacity.value = data.backgroundOpacity
             if (typeof data.ttsInstruct === 'string') ttsInstruct.value = data.ttsInstruct
+            if (data.fontFamily) fontFamily.value = data.fontFamily
+            if (typeof data.fontSize === 'number') fontSize.value = data.fontSize
         } catch { /* use defaults */ }
     }
 
@@ -58,9 +63,11 @@ export const useSettingsStore = defineStore('settings', () => {
                 theme: theme.value, modelThinking: modelThinking.value,
                 debounceMs: debounceMs.value, proThinking: proThinking.value,
                 privacyMode: privacyMode.value, language: language.value,
-                currency: currency.value, backgroundType: backgroundType.value,
+                country: country.value, backgroundType: backgroundType.value,
                 backgroundImage: backgroundImage.value, backgroundOpacity: backgroundOpacity.value,
                 ttsInstruct: ttsInstruct.value,
+                fontFamily: fontFamily.value,
+                fontSize: fontSize.value,
             }))
         } catch { /* silently fail */ }
     }
@@ -68,17 +75,17 @@ export const useSettingsStore = defineStore('settings', () => {
     const resetSettings = () => {
         theme.value = 'system'; modelThinking.value = 'low'; debounceMs.value = 1000
         proThinking.value = 'medium'; privacyMode.value = false; language.value = 'auto'
-        currency.value = 'auto'; backgroundType.value = 'default'; backgroundImage.value = ''
-        backgroundOpacity.value = 0.2; ttsInstruct.value = ''; saveSettings()
+        country.value = 'auto'; backgroundType.value = 'default'; backgroundImage.value = ''
+        backgroundOpacity.value = 0.2; ttsInstruct.value = ''; fontFamily.value = 'system'; fontSize.value = 16; saveSettings()
     }
 
     // Watch all reactive refs and auto-save on change
-    watch([theme, modelThinking, debounceMs, proThinking, privacyMode, language, currency
-        , backgroundType, backgroundImage, backgroundOpacity, ttsInstruct], saveSettings)
+    watch([theme, modelThinking, debounceMs, proThinking, privacyMode, language, country
+        , backgroundType, backgroundImage, backgroundOpacity, ttsInstruct, fontFamily, fontSize], saveSettings)
 
     loadSettings() // Initialize from localStorage
 
     return { theme, modelThinking, debounceMs, proThinking, privacyMode, language
-        , currency, backgroundType, backgroundImage, backgroundOpacity, ttsInstruct
+        , country, backgroundType, backgroundImage, backgroundOpacity, ttsInstruct, fontFamily, fontSize
         , detectedTimezone, uiLanguage, t, initialMarkdown, resetSettings }
 })

src/utils/api.js

@@ -76,7 +76,7 @@ function buildCompletionBody(settings, prefix, suffix, languageId, extra = {}) {
     privacy_mode: settings.privacyMode,
     user_preferences: {
       language: settings.language,
-      currency: settings.currency,
+      country: settings.country,
       timezone: settings.detectedTimezone,
     },
     ...extra,
@@ -252,7 +252,7 @@ export async function fetchProSuggestionStream(payload, apiUrl = PRO_URL) {
       privacy_mode: settings.privacyMode,
       user_preferences: {
         language: settings.language,
-        currency: settings.currency,
+        country: settings.country,
         timezone: settings.detectedTimezone,
       },
     },
@@ -304,7 +304,7 @@ export async function fetchWebSearchStream(payload, apiUrl = WEB_SEARCH_URL) {
       privacy_mode: settings.privacyMode,
       user_preferences: {
         language: settings.language,
-        currency: settings.currency,
+        country: settings.country,
         timezone: settings.detectedTimezone,
       },
     },

src/utils/convert.js

@@ -1,4 +1,4 @@
-import { CONVERT_URL, ASR_URL } from './config.js'
+import { CONVERT_URL, ASR_URL, OCR_URL } from './config.js'
 
 import { parseSseEvent } from './sse.js'
 
@@ -86,6 +86,31 @@ export async function convertFileToMarkdown(file) {
     return data.markdown
 }
 
+export async function convertVideoToText(file, language = 'auto') {
+    const base64 = await readFileAsBase64(file)
+    const headers = {
+        'Content-Type': 'application/json',
+    }
+    const res = await fetch(OCR_URL, {
+        method: 'POST',
+        headers,
+        credentials: 'include',
+        body: JSON.stringify({
+            image: base64,
+            filename: file.name || 'video.mp4',
+            language,
+            media_type: 'video',
+            mime_type: file.type || 'video/mp4',
+        }),
+    })
+
+    const data = await consumeSseResult(res)
+    if (!data || typeof data.text !== 'string' || !data.text.trim()) {
+        throw new Error('视频解析结果为空')
+    }
+    return data.text
+}
+
 /**
  * Encode AudioBuffer as WAV (16kHz mono, 16-bit PCM) and return base64 string
  */

src/utils/docBlock.js

@@ -18,6 +18,7 @@ function clipDocContext(content = '', limit = 0) {
 }
 
 const AUDIO_EXT_RE = /\.(wav|mp3|m4a|ogg|flac)$/i
+const VIDEO_EXT_RE = /\.(mp4|webm|mov|avi|mkv|m4v|ogv)$/i
 
 export function normalizeDocType(value = '') {
   const lower = String(value || '').trim().toLowerCase()
@@ -61,6 +62,13 @@ export function isAudioFile(file) {
   return AUDIO_EXT_RE.test(name) || type.startsWith('audio/')
 }
 
+export function isVideoFile(file) {
+  if (!file) return false
+  const name = String(file.name || '').toLowerCase()
+  const type = String(file.type || '').toLowerCase()
+  return VIDEO_EXT_RE.test(name) || type.startsWith('video/')
+}
+
 export function isSupportedDocFile(file) {
   if (!file) return false
   const name = String(file.name || '').toLowerCase()
@@ -75,6 +83,7 @@ export function isSupportedDocFile(file) {
     name.endsWith('.pptx') ||
     name.endsWith('.pdf') ||
     AUDIO_EXT_RE.test(name) ||
+    VIDEO_EXT_RE.test(name) ||
     type === 'text/plain' ||
     type === 'application/json' ||
     type === 'text/yaml' ||
@@ -83,7 +92,8 @@ export function isSupportedDocFile(file) {
     type === 'application/vnd.openxmlformats-officedocument.wordprocessingml.document' ||
     type === 'application/vnd.openxmlformats-officedocument.presentationml.presentation' ||
     type === 'application/pdf' ||
-    type.startsWith('audio/')
+    type.startsWith('audio/') ||
+    type.startsWith('video/')
   )
 }
 

src/utils/i18n.js

@@ -33,7 +33,7 @@ export const translations = {
         privacyDesc: 'Prevent sending IP and preferences to the AI',
         language: 'Language',
         auto: 'Auto Detect',
-        currency: 'Currency',
+        country: 'Country',
         about: 'About Us',
         importMd: 'Import Markdown',
         exportMd: 'Export Markdown',
@@ -52,7 +52,7 @@ export const translations = {
     uploadFileError: 'File upload failed.',
     uploadConvertError: 'File conversion failed.',
     uploadBatchLimit: 'Maximum 10 files at once',
-    uploadSizeLimit: 'File exceeds 50MB limit',
+        uploadSizeLimit: 'File exceeds 100MB limit',
     uploading: 'Uploading files...',
     enableAI: 'Enable AI',
         disableAI: 'Disable AI',
@@ -160,7 +160,7 @@ export const translations = {
         privacyDesc: '不向 AI 发送 IP 地址和偏好设置',
         language: '语言',
         auto: '自动检测',
-        currency: '货币',
+        country: '国家',
         about: '关于我们',
         importMd: '导入 Markdown',
         exportMd: '导出 Markdown',
@@ -179,7 +179,7 @@ export const translations = {
     uploadFileError: '文件上传失败',
     uploadConvertError: '文件转换失败',
     uploadBatchLimit: '一次最多上传10个文件',
-    uploadSizeLimit: '文件超过50MB限制',
+    uploadSizeLimit: '文件超过100MB限制',
     uploading: '正在上传文件...',
     enableAI: '启用 AI',
         disableAI: '禁用 AI',
@@ -287,7 +287,7 @@ export const translations = {
         privacyDesc: 'AIにIPアドレスと設定を送信しない',
         language: '言語',
         auto: '自動検出',
-        currency: '通貨',
+        country: '国',
         about: '私たちについて',
         template: 'テンプレート',
         presetTemplates: 'プリセットテンプレート',
@@ -328,7 +328,7 @@ export const translations = {
     uploadFileError: 'アップロードに失敗しました。',
     uploadConvertError: '変換に失敗しました。',
     uploadBatchLimit: '一度にアップロードできるのは10ファイルまでです。',
-    uploadSizeLimit: '50MBの制限を超えています。',
+    uploadSizeLimit: '100MBの制限を超えています。',
     uploading: 'アップロード中...',
         enableAI: 'AIを有効化',
         disableAI: 'AIを無効化',
@@ -408,7 +408,7 @@ export const translations = {
         privacyDesc: 'AI에 IP 주소 및 설정 전송 안 함',
         language: '언어',
         auto: '자동 감지',
-        currency: '통화',
+        country: '국가',
         about: '회사 소개',
         template: '템플릿',
         presetTemplates: '사전 정의된 템플릿',
@@ -450,7 +450,7 @@ export const translations = {
     uploadFileError: '업로드 실패',
     uploadConvertError: '변환 실패',
     uploadBatchLimit: '한 번에 최대 10개 파일까지 업로드할 수 있습니다.',
-    uploadSizeLimit: '50MB 제한을 초과합니다.',
+    uploadSizeLimit: '100MB 제한을 초과합니다.',
     uploading: '업로드 중...',
         enableAI: 'AI 활성화',
         disableAI: 'AI 비활성화',
@@ -530,7 +530,7 @@ export const translations = {
         privacyDesc: 'Sende keine IP und Einstellungen an KI',
         language: 'Sprache',
         auto: 'Automatisch',
-        currency: 'Währung',
+        country: 'Land',
         about: 'Über uns',
         template: 'Vorlage',
         presetTemplates: 'Vorgabe-Vorlagen',
@@ -571,7 +571,7 @@ export const translations = {
     uploadFileError: 'Hochladen fehlgeschlagen',
     uploadConvertError: 'Konversion fehlgeschlagen',
     uploadBatchLimit: 'Maximal 10 Dateien auf einmal.',
-    uploadSizeLimit: 'Datei überschreitet das Limit von 50MB.',
+    uploadSizeLimit: 'Datei überschreitet das Limit von 100MB.',
     uploading: 'Wird hochgeladen...',
         enableAI: 'KI aktivieren',
         disableAI: 'KI deaktivieren',
@@ -651,7 +651,7 @@ export const translations = {
         privacyDesc: 'Ne pas envoyer IP et préférences à l\'IA',
         language: 'Langue',
         auto: 'Détection auto',
-        currency: 'Devise',
+        country: 'Pays',
         about: 'À propos de nous',
         template: 'Modèle',
         presetTemplates: 'Modèles prédéfinis',
@@ -692,7 +692,7 @@ export const translations = {
     uploadFileError: 'Échec du téléchargement',
     uploadConvertError: 'Échec de la conversion',
     uploadBatchLimit: 'Maximum 10 fichiers à la fois.',
-    uploadSizeLimit: 'Fichier dépasse la limite de 50 Mo.',
+    uploadSizeLimit: 'Fichier dépasse la limite de 100 Mo.',
     uploading: 'Téléchargement en cours...',
         enableAI: 'Activer IA',
         disableAI: 'Désactiver IA',

src/utils/inputBlock.js

@@ -0,0 +1,77 @@
+export const INPUT_BLOCK_NODE_TYPE = 'input_block'
+export const INPUT_TRIGGER_TEXT = '[INPUT]'
+
+const INPUT_INSTRUCTION_PREFIX = '[INPUT]{'
+const INPUT_INSTRUCTION_SUFFIX = '}'
+
+export const INPUT_DISPLAY_LABEL = '输入'
+
+function normalizeMarkdownText(value = '') {
+  return String(value || '').replace(/\r\n?/g, '\n')
+}
+
+export function escapeInputBlockContent(value = '') {
+  return normalizeMarkdownText(value)
+    .replace(/\\/g, '\\\\')
+    .replace(/\n/g, '\\n')
+}
+
+export function unescapeInputBlockContent(value = '') {
+  const normalized = String(value || '').replace(/\r\n?/g, '\n')
+  let result = ''
+
+  for (let index = 0; index < normalized.length; index += 1) {
+    const char = normalized[index]
+    if (char !== '\\' || index === normalized.length - 1) {
+      result += char
+    } else {
+      const next = normalized[index + 1]
+      if (next === 'n') {
+        result += '\n'
+      } else {
+        result += next
+      }
+    }
+  }
+
+  return result
+}
+
+export function serializeInputBlockSyntax(instruction = '', text = '') {
+  const normalizedInstr = normalizeMarkdownText(instruction).trim()
+  if (!normalizedInstr) return INPUT_TRIGGER_TEXT
+  const instrPart = `${INPUT_INSTRUCTION_PREFIX}${escapeInputBlockContent(normalizedInstr)}${INPUT_INSTRUCTION_SUFFIX}`
+  if (!text) return instrPart
+  
+  const normalizedText = escapeInputBlockContent(text.trim())
+  if (!normalizedText) return instrPart
+  
+  // Format: [INPUT]{instructions}***用户输入的文本***
+  return `${instrPart}***${normalizedText}***`
+}
+
+export function parseInputBlockSyntax(value = '') {
+  const text = normalizeMarkdownText(value).trim()
+  if (!text) return null
+
+  // Match full pattern: [INPUT]{instr}***userText*** or [INPUT]***text***
+  const fullMatch = text.match(/^(\[INPUT\](?:\s*\{([^\}]*)\})?)?\*\*([\s\S]*?)\*\*\*$/i)
+  if (fullMatch) {
+    return { 
+      instruction: fullMatch[2] || '', 
+      userInputText: fullMatch[3].trim()
+    }
+  }
+
+  // Match just the trigger: [INPUT] or [INPUT]{instr}  
+  const triggerMatch = text.match(/^\[INPUT\](?:\s*\{([^\}]*)\})?$/i)
+  if (triggerMatch) {
+    return { 
+      instruction: triggerMatch[1] || '', 
+      userInputText: ''
+    }
+  }
+
+  return null
+}INPUT_BLOCK_UTIL_EOF
+echo "Created inputBlock.js utility" && wc -l /Users/allenyuan/llm-in-text/src/utils/inputBlock.js

src/utils/richExport.js

@@ -0,0 +1,455 @@
+function normalizeText(value = '') {
+  return String(value || '').replace(/\u00a0/g, ' ').replace(/\s+/g, ' ').trim()
+}
+
+function toArray(value) {
+  return Array.isArray(value) ? value : []
+}
+
+function removeNodes(root, selectors) {
+  for (const selector of selectors) {
+    root.querySelectorAll(selector).forEach((node) => node.remove())
+  }
+}
+
+function buildCardSection({ title, meta, body, className }) {
+  const section = document.createElement('section')
+  section.className = className
+
+  const heading = document.createElement('h2')
+  heading.textContent = title
+  section.appendChild(heading)
+
+  if (meta) {
+    const metaNode = document.createElement('p')
+    metaNode.className = `${className}__meta`
+    metaNode.textContent = meta
+    section.appendChild(metaNode)
+  }
+
+  if (body) {
+    section.appendChild(body)
+  }
+
+  return section
+}
+
+/** @typedef {'doc-card' | 'web-search-card'} CardType */
+
+function transformDocCards(root) {
+  const cards = root.querySelectorAll('.doc-card')
+  
+  if (cards.length === 0) {
+    console.debug('[Export] No doc cards found in export root') // debug level - safe for production
+  }
+
+  cards.forEach((card) => {
+    // Validate required elements exist before transform
+    const titleEl = card.querySelector('.doc-card__name')
+    if (!titleEl) {
+      console.warn('[Export] Doc card missing title element - skipping transform') // warn for visibility
+    }
+
+    const body = card.querySelector('.doc-card__body')
+    if (!body) {
+      console.error('[Export] Doc card has no body content - this will cause export failure') // error for critical path
+    }
+
+    const title = normalizeText(titleEl?.textContent) || '文档块'
+    const badge = normalizeText(card.querySelector('.doc-card__badge')?.textContent)
+    const time = normalizeText(card.querySelector('.doc-card__time')?.textContent)
+    const meta = [badge, time].filter(Boolean).join(' · ')
+
+    // Clone body content and strip interactive elements
+    const content = body ? body.cloneNode(true) : document.createElement('div')
+
+    // Remove interactive elements that don't belong in exports
+    removeNodes(content, [
+      'button',           // Delete/expand buttons
+      'svg',              // Icons (trash, chevron)
+      '.btn-tooltip',     // Tooltip wrappers around buttons
+      '.doc-card__actions',  // Action bar container (holds delete button)
+      '.doc-card__status-label' // Status label ("处理中"/"已完成") - keep for context
+    ])
+
+    card.replaceWith(buildCardSection({
+      title,
+      meta,
+      body: content,
+      className: 'export-doc-block',
+    }))
+
+    // Log transform result for debugging (structured logging)
+    console.debug('[Export] Transformed doc card:', { title, hasBody: Boolean(body), meta })
+  })
+
+  return cards.length // Return count for verification downstream
+}
+
+function transformWebSearchCards(root) {
+  const cards = root.querySelectorAll('.web-search-card')
+
+  if (cards.length === 0) {
+    console.debug('[Export] No web search cards found in export root') // debug level - safe for production
+  }
+
+  cards.forEach((card) => {
+    // Validate required elements exist before transform
+    const titleEl = card.querySelector('.web-search-card__title')
+    if (!titleEl) {
+      console.warn('[Export] Web search card missing title element') // warn for visibility
+    }
+
+    const body = card.querySelector('.web-search-card__body')
+    if (!body) {
+      console.error('[Export] Web search card has no body content') // error for critical path
+    }
+
+    const title = normalizeText(titleEl?.textContent) || '联网搜索结果'
+    const subtitle = normalizeText(card.querySelector('.web-search-card__subtitle')?.textContent)
+
+    // Clone and strip interactive elements
+    const content = body ? body.cloneNode(true) : document.createElement('div')
+
+    // Remove interactive elements
+    removeNodes(content, [
+      'button',              // Delete/collapse buttons
+      'svg',                 // Icons (spinner, chevron)
+      '.btn-tooltip',        // Tooltip wrappers
+      '.web-search-card__actions',   // Action bar (delete button)
+      '.web-search-card__progress'   // Progress indicator - keep for status context
+    ])
+
+    card.replaceWith(buildCardSection({
+      title,
+      meta: subtitle,
+      body: content,
+      className: 'export-websearch-block',
+    }))
+
+    // Log transform result for debugging (structured logging)
+    console.debug('[Export] Transformed web search card:', { title, hasBody: Boolean(body), meta })
+  })
+
+  return cards.length // Return count for verification downstream
+}
+
+function inlineTextRuns(node, docx) {
+  const { TextRun, ExternalHyperlink } = docx
+  const runs = []
+
+  const walk = (current, marks = {}) => {
+    if (current.nodeType === Node.TEXT_NODE) {
+      const text = current.textContent || ''
+      if (!text.trim() && !text.includes('\n')) return
+      runs.push(new TextRun({
+        text: text.replace(/\s+/g, ' '),
+        bold: Boolean(marks.bold),
+        italics: Boolean(marks.italics),
+        underline: marks.underline ? {} : undefined,
+        font: marks.code ? 'Menlo' : undefined,
+      }))
+      return
+    }
+
+    if (!(current instanceof HTMLElement)) return
+
+    if (current.tagName === 'BR') {
+      runs.push(new TextRun({ break: 1 }))
+      return
+    }
+
+    if (current.tagName === 'A') {
+      const href = current.getAttribute('href') || ''
+      const childRuns = []
+      const startCount = runs.length
+      toArray(Array.from(current.childNodes)).forEach((child) => walk(child, { ...marks, underline: true }))
+      childRuns.push(...runs.splice(startCount))
+      if (href && childRuns.length > 0) {
+        runs.push(new ExternalHyperlink({ link: href, children: childRuns }))
+      } else {
+        runs.push(...childRuns)
+      }
+      return
+    }
+
+    const nextMarks = {
+      ...marks,
+      bold: marks.bold || current.tagName === 'STRONG' || current.tagName === 'B',
+      italics: marks.italics || current.tagName === 'EM' || current.tagName === 'I',
+      underline: marks.underline || current.tagName === 'U',
+      code: marks.code || current.tagName === 'CODE',
+    }
+
+    toArray(Array.from(current.childNodes)).forEach((child) => walk(child, nextMarks))
+  }
+
+  walk(node)
+
+  if (runs.length === 0) {
+    const text = normalizeText(node.textContent)
+    if (text) runs.push(new TextRun(text))
+  }
+
+  return runs
+}
+
+async function domToDocxSections(root) {
+  const docx = await import('docx')
+  const {
+    Paragraph,
+    HeadingLevel,
+    BorderStyle,
+  } = docx
+
+  const paragraphs = []
+
+  const pushParagraph = (options) => {
+    paragraphs.push(new Paragraph(options))
+  }
+
+  const walkBlock = (node, depth = 0) => {
+    if (!(node instanceof HTMLElement)) return
+
+    const tag = node.tagName
+    if (['SCRIPT', 'STYLE'].includes(tag)) return
+
+    if (tag === 'H1' || tag === 'H2' || tag === 'H3' || tag === 'H4') {
+      const levelMap = {
+        H1: HeadingLevel.HEADING_1,
+        H2: HeadingLevel.HEADING_2,
+        H3: HeadingLevel.HEADING_3,
+        H4: HeadingLevel.HEADING_4,
+      }
+      pushParagraph({
+        heading: levelMap[tag],
+        children: inlineTextRuns(node, docx),
+        spacing: { before: 240, after: 120 },
+      })
+      return
+    }
+
+    if (tag === 'UL' || tag === 'OL') {
+      toArray(Array.from(node.children)).forEach((child) => walkBlock(child, depth + 1))
+      return
+    }
+
+    if (tag === 'LI') {
+      pushParagraph({
+        children: inlineTextRuns(node, docx),
+        bullet: { level: Math.max(0, depth - 1) },
+        spacing: { after: 80 },
+      })
+      return
+    }
+
+    if (tag === 'PRE') {
+      const code = node.textContent || ''
+      pushParagraph({
+        children: [new docx.TextRun({ text: code, font: 'Menlo' })],
+        spacing: { before: 120, after: 120 },
+        border: {
+          top: { style: BorderStyle.SINGLE, size: 1, color: 'D1D5DB' },
+          bottom: { style: BorderStyle.SINGLE, size: 1, color: 'D1D5DB' },
+          left: { style: BorderStyle.SINGLE, size: 1, color: 'D1D5DB' },
+          right: { style: BorderStyle.SINGLE, size: 1, color: 'D1D5DB' },
+        },
+        shading: { fill: 'F8FAFC' },
+      })
+      return
+    }
+
+    if (tag === 'BLOCKQUOTE') {
+      pushParagraph({
+        children: inlineTextRuns(node, docx),
+        indent: { left: 480 },
+        spacing: { before: 120, after: 120 },
+      })
+      return
+    }
+
+    if (tag === 'TABLE') {
+      toArray(Array.from(node.querySelectorAll('tr'))).forEach((row) => {
+        const cells = toArray(Array.from(row.children)).map((cell) => normalizeText(cell.textContent)).filter(Boolean)
+        if (cells.length > 0) {
+          pushParagraph({
+            children: [new docx.TextRun(cells.join(' | '))],
+            spacing: { after: 80 },
+          })
+        }
+      })
+      return
+    }
+
+    if (tag === 'IMG') {
+      const alt = normalizeText(node.getAttribute('alt') || '')
+      pushParagraph({
+        children: [new docx.TextRun(alt ? `[图片] ${alt}` : '[图片]')],
+        spacing: { before: 80, after: 80 },
+      })
+      return
+    }
+
+    const blockLike = ['P', 'DIV', 'SECTION', 'ARTICLE']
+    const hasDirectBlockChildren = toArray(Array.from(node.children)).some((child) =>
+      ['P', 'DIV', 'SECTION', 'ARTICLE', 'UL', 'OL', 'LI', 'PRE', 'BLOCKQUOTE', 'TABLE', 'H1', 'H2', 'H3', 'H4'].includes(child.tagName)
+    )
+
+    if (blockLike.includes(tag) && !hasDirectBlockChildren) {
+      const text = normalizeText(node.textContent)
+      if (text) {
+        pushParagraph({
+          children: inlineTextRuns(node, docx),
+          spacing: { after: 120 },
+        })
+      }
+      return
+    }
+
+    toArray(Array.from(node.children)).forEach((child) => walkBlock(child, depth))
+  }
+
+  toArray(Array.from(root.children)).forEach((child) => walkBlock(child))
+
+  if (paragraphs.length === 0) {
+    pushParagraph({ text: normalizeText(root.textContent) || '' })
+  }
+
+  return { docx, paragraphs }
+}
+
+export function createExpandedExportRoot(editorElement) {
+  if (!(editorElement instanceof HTMLElement)) {
+    throw new Error('导出失败：编辑器 DOM 不可用')
+  }
+
+  const host = document.createElement('div')
+  host.className = 'rich-export-root'
+  host.innerHTML = `
+    <style>
+      .rich-export-root {
+        width: 800px;
+        padding: 40px 48px;
+        background: #ffffff;
+        color: #111827;
+        font: 16px/1.65 -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+      }
+      .rich-export-root * {
+        box-sizing: border-box;
+        color: inherit !important;
+      }
+      .rich-export-root img {
+        max-width: 100%;
+        height: auto;
+      }
+      .rich-export-root .export-doc-block,
+      .rich-export-root .export-websearch-block {
+        margin: 24px 0;
+        padding: 16px 18px;
+        border: 1px solid #dbe4f0;
+        border-radius: 10px;
+        background: #f8fafc;
+      }
+      .rich-export-root .export-doc-block__meta,
+      .rich-export-root .export-websearch-block__meta {
+        margin: 6px 0 14px;
+        color: #475569 !important;
+        font-size: 13px;
+      }
+      .rich-export-root h1,
+      .rich-export-root h2,
+      .rich-export-root h3,
+      .rich-export-root h4 {
+        margin: 20px 0 12px;
+      }
+      .rich-export-root pre {
+        white-space: pre-wrap;
+        word-break: break-word;
+        padding: 12px 14px;
+        border-radius: 8px;
+        background: #f1f5f9;
+        overflow: hidden;
+      }
+      .rich-export-root table {
+        width: 100%;
+        border-collapse: collapse;
+      }
+      .rich-export-root th,
+      .rich-export-root td {
+        border: 1px solid #d1d5db;
+        padding: 8px 10px;
+        vertical-align: top;
+      }
+      .rich-export-root .btn-tooltip,
+      .rich-export-root button,
+      .rich-export-root svg {
+        display: none !important;
+      }
+    </style>
+  `
+
+  const cloned = editorElement.cloneNode(true)
+  if (!(cloned instanceof HTMLElement)) {
+    throw new Error('导出失败：编辑器内容不可用')
+  }
+
+  removeNodes(cloned, [
+    '.history-buttons',
+    '.action-buttons',
+    '.top-actions-fixed',
+    '.ProseMirror-gapcursor',
+    '.ProseMirror-widget',
+    '.doc-card__actions',
+    '.doc-card__status-label',
+    '.web-search-card__actions',
+    '.web-search-card__progress',
+    '.btn-tooltip',
+  ])
+
+  cloned.querySelectorAll('.doc-card__body, .web-search-card__body').forEach((node) => {
+    if (node instanceof HTMLElement) {
+      node.style.display = 'block'
+    }
+  })
+
+  transformDocCards(cloned)
+  transformWebSearchCards(cloned)
+
+  host.appendChild(cloned)
+  return host
+}
+
+export async function exportEditorToPdf(editorElement, filename) {
+  const html2pdfModule = await import('html2pdf.js')
+  const html2pdf = html2pdfModule.default || html2pdfModule
+  const root = createExpandedExportRoot(editorElement)
+  root.style.position = 'fixed'
+  root.style.left = '-100000px'
+  root.style.top = '0'
+  document.body.appendChild(root)
+
+  try {
+    await html2pdf()
+      .set({
+        margin: [10, 10, 10, 10],
+        filename,
+        html2canvas: { scale: 2, useCORS: true, backgroundColor: '#ffffff' },
+        jsPDF: { unit: 'mm', format: 'a4', orientation: 'portrait' },
+        pagebreak: { mode: ['css', 'legacy'] },
+      })
+      .from(root)
+      .save()
+  } finally {
+    root.remove()
+  }
+}
+
+export async function exportEditorToDocxBlob(editorElement) {
+  const root = createExpandedExportRoot(editorElement)
+  const { docx, paragraphs } = await domToDocxSections(root)
+  const { Document, Packer } = docx
+  const doc = new Document({
+    sections: [{ properties: {}, children: paragraphs }],
+  })
+  return Packer.toBlob(doc)
+}