fix(avatar): follow language changes each turn

Reviewed-on: #19

backend/app/api/endpoints/ai_models.py

@@ -1,8 +1,11 @@
 """AI模型配置接口"""
-from fastapi import APIRouter, Depends, HTTPException
+import secrets
+
+from fastapi import APIRouter, Depends, Header, HTTPException
 from sqlalchemy import select, update
 
 from app.core.database import get_db
+from app.core.config import settings
 from app.schemas import ApiResponse, AIModelCreateRequest, AIModelUpdateRequest, AIModelTestRequest
 from app.models import AIModelConfig
 from app.utils.crypto import encrypt, decrypt
@@ -22,13 +25,20 @@ async def list_models(db=Depends(get_db)):
 @router.post("")
 async def create_model(req: AIModelCreateRequest, db=Depends(get_db)):
     if req.is_default:
-        await db.execute(update(AIModelConfig).values(is_default=0))
+        await db.execute(
+            update(AIModelConfig)
+            .where(AIModelConfig.usage_scope == req.usage_scope)
+            .values(is_default=0)
+        )
     model = AIModelConfig(
         model_name=req.model_name,
         provider=req.provider,
+        usage_scope=req.usage_scope,
         api_base_url=req.api_base_url,
         api_key_enc=encrypt(req.api_key) if req.api_key else None,
         model_version=req.model_version,
+        vision_model_version=req.vision_model_version,
+        ocr_model_version=req.ocr_model_version,
         temperature=req.temperature,
         max_tokens=req.max_tokens,
         timeout_seconds=req.timeout_seconds,
@@ -47,8 +57,16 @@ async def update_model(model_id: int, req: AIModelUpdateRequest, db=Depends(get_
     model = result.scalar_one_or_none()
     if not model:
         raise HTTPException(status_code=404, detail="模型不存在")
-    if req.is_default:
-        await db.execute(update(AIModelConfig).where(AIModelConfig.id != model_id).values(is_default=0))
+    target_scope = req.usage_scope or model.usage_scope
+    if req.is_default or (req.usage_scope and model.is_default):
+        await db.execute(
+            update(AIModelConfig)
+            .where(
+                AIModelConfig.id != model_id,
+                AIModelConfig.usage_scope == target_scope,
+            )
+            .values(is_default=0)
+        )
     for field, val in req.model_dump(exclude_none=True).items():
         if field == "api_key":
             model.api_key_enc = encrypt(val) if val else None
@@ -59,6 +77,39 @@ async def update_model(model_id: int, req: AIModelUpdateRequest, db=Depends(get_
     return ApiResponse(data=_format_model(model), message="更新成功")
 
 
+@router.get("/runtime/digital-avatar")
+async def get_digital_avatar_runtime_model(
+    x_avatar_config_token: str | None = Header(default=None),
+    db=Depends(get_db),
+):
+    expected = settings.AVATAR_MODEL_CONFIG_TOKEN
+    if not expected:
+        raise HTTPException(status_code=503, detail="数字分身模型配置服务未启用")
+    if not x_avatar_config_token or not secrets.compare_digest(x_avatar_config_token, expected):
+        raise HTTPException(status_code=401, detail="无权读取数字分身模型配置")
+
+    result = await db.execute(
+        select(AIModelConfig).where(
+            AIModelConfig.usage_scope == "digital_avatar",
+            AIModelConfig.is_default == 1,
+            AIModelConfig.is_enabled == 1,
+        )
+    )
+    model = result.scalar_one_or_none()
+    if not model:
+        raise HTTPException(status_code=404, detail="尚未配置启用的数字分身专用模型")
+    return ApiResponse(data={
+        "api_base_url": model.api_base_url or "https://api.openai.com/v1",
+        "api_key": decrypt(model.api_key_enc) if model.api_key_enc else "",
+        "model": model.model_version or model.model_name,
+        "vision_model": model.vision_model_version or "qwen3.6-flash",
+        "ocr_model": model.ocr_model_version or "qwen-vl-ocr",
+        "temperature": model.temperature,
+        "max_tokens": model.max_tokens,
+        "timeout_seconds": model.timeout_seconds,
+    })
+
+
 @router.delete("/{model_id}")
 async def delete_model(model_id: int, db=Depends(get_db)):
     result = await db.execute(select(AIModelConfig).where(AIModelConfig.id == model_id))
@@ -79,8 +130,11 @@ async def test_model(req: AIModelTestRequest, db=Depends(get_db)):
 def _format_model(m: AIModelConfig) -> dict:
     return {
         "id": m.id, "model_name": m.model_name, "provider": m.provider,
+        "usage_scope": m.usage_scope,
         "api_base_url": m.api_base_url, "has_api_key": bool(m.api_key_enc),
         "model_version": m.model_version, "temperature": m.temperature,
+        "vision_model_version": m.vision_model_version,
+        "ocr_model_version": m.ocr_model_version,
         "max_tokens": m.max_tokens, "timeout_seconds": m.timeout_seconds,
         "is_default": m.is_default, "is_enabled": m.is_enabled,
         "created_at": m.created_at.isoformat(),

backend/app/core/config.py

@@ -19,6 +19,7 @@ class Settings(BaseSettings):
     # 安全
     SECRET_KEY: str = os.getenv("SECRET_KEY", "dev-secret-key-change-in-prod")
     AES_KEY: str = os.getenv("AES_KEY", "your-aes-key-32-chars-change-now!")
+    AVATAR_MODEL_CONFIG_TOKEN: str = os.getenv("AVATAR_MODEL_CONFIG_TOKEN", "")
 
     # 新闻平台
     NEWS_PLATFORM_BASE_URL: str = os.getenv(

backend/app/core/database.py

@@ -1,6 +1,7 @@
 """数据库连接管理"""
 import asyncio
 from sqlalchemy.ext.asyncio import create_async_engine, AsyncSession, async_sessionmaker
+from sqlalchemy import text
 from sqlalchemy.orm import DeclarativeBase
 from app.core.config import settings
 from app.core.logger import logger
@@ -64,6 +65,37 @@ async def init_db():
             VirtualUser, UserPersonality, InteractionRecord,
             PendingReplyTask, TokenStat, AIModelConfig, SystemConfig, LoginLog
         )
+        async with engine.begin() as conn:
+            await conn.execute(text("SELECT GET_LOCK('ai_model_config_migration', 30)"))
+            try:
+                columns = (
+                    (
+                        "usage_scope",
+                        "ALTER TABLE ai_model_configs ADD COLUMN usage_scope "
+                        "VARCHAR(16) NOT NULL DEFAULT 'general' AFTER provider",
+                    ),
+                    (
+                        "vision_model_version",
+                        "ALTER TABLE ai_model_configs ADD COLUMN vision_model_version "
+                        "VARCHAR(64) NULL AFTER model_version",
+                    ),
+                    (
+                        "ocr_model_version",
+                        "ALTER TABLE ai_model_configs ADD COLUMN ocr_model_version "
+                        "VARCHAR(64) NULL AFTER vision_model_version",
+                    ),
+                )
+                for column_name, ddl in columns:
+                    result = await conn.execute(text(
+                        "SELECT COUNT(*) FROM information_schema.COLUMNS "
+                        "WHERE TABLE_SCHEMA = DATABASE() AND TABLE_NAME = 'ai_model_configs' "
+                        "AND COLUMN_NAME = :column_name"
+                    ), {"column_name": column_name})
+                    if result.scalar_one() == 0:
+                        await conn.execute(text(ddl))
+                        logger.info("AI模型配置表已增加 %s 字段", column_name)
+            finally:
+                await conn.execute(text("SELECT RELEASE_LOCK('ai_model_config_migration')"))
         logger.info("✅ 数据库模型注册成功")
         logger.info("✅ 数据库初始化完成")
 

backend/app/models/__init__.py

@@ -122,9 +122,12 @@ class AIModelConfig(Base):
     id: Mapped[int] = mapped_column(BigInteger, primary_key=True, autoincrement=True)
     model_name: Mapped[str] = mapped_column(String(64), nullable=False)
     provider: Mapped[str] = mapped_column(String(32), nullable=False)
+    usage_scope: Mapped[str] = mapped_column(String(16), nullable=False, default="general")
     api_base_url: Mapped[str | None] = mapped_column(String(256))
     api_key_enc: Mapped[str | None] = mapped_column(String(512))
     model_version: Mapped[str | None] = mapped_column(String(64))
+    vision_model_version: Mapped[str | None] = mapped_column(String(64))
+    ocr_model_version: Mapped[str | None] = mapped_column(String(64))
     temperature: Mapped[float] = mapped_column(Float, default=0.7)
     max_tokens: Mapped[int] = mapped_column(Integer, default=1000)
     timeout_seconds: Mapped[int] = mapped_column(Integer, default=30)

backend/app/schemas/__init__.py

@@ -154,9 +154,12 @@ class InteractionResponse(BaseModel):
 class AIModelCreateRequest(BaseModel):
     model_name: str = Field(..., min_length=1, max_length=64)
     provider: str = Field(..., pattern="^(openai|zhipu|wenxin|qianwen|local)$")
+    usage_scope: str = Field(default="general", pattern="^(general|digital_avatar)$")
     api_base_url: Optional[str] = None
     api_key: Optional[str] = None
     model_version: Optional[str] = None
+    vision_model_version: Optional[str] = Field(None, max_length=64)
+    ocr_model_version: Optional[str] = Field(None, max_length=64)
     temperature: float = Field(default=0.7, ge=0.0, le=2.0)
     max_tokens: int = Field(default=1000, ge=1, le=32000)
     timeout_seconds: int = Field(default=30, ge=5, le=300)
@@ -165,9 +168,13 @@ class AIModelCreateRequest(BaseModel):
 
 class AIModelUpdateRequest(BaseModel):
     model_name: Optional[str] = None
+    provider: Optional[str] = Field(None, pattern="^(openai|zhipu|wenxin|qianwen|local)$")
+    usage_scope: Optional[str] = Field(None, pattern="^(general|digital_avatar)$")
     api_base_url: Optional[str] = None
     api_key: Optional[str] = None
     model_version: Optional[str] = None
+    vision_model_version: Optional[str] = Field(None, max_length=64)
+    ocr_model_version: Optional[str] = Field(None, max_length=64)
     temperature: Optional[float] = Field(None, ge=0.0, le=2.0)
     max_tokens: Optional[int] = Field(None, ge=1, le=32000)
     timeout_seconds: Optional[int] = Field(None, ge=5, le=300)
@@ -179,9 +186,12 @@ class AIModelResponse(BaseModel):
     id: int
     model_name: str
     provider: str
+    usage_scope: str
     api_base_url: Optional[str]
     has_api_key: bool
     model_version: Optional[str]
+    vision_model_version: Optional[str]
+    ocr_model_version: Optional[str]
     temperature: float
     max_tokens: int
     timeout_seconds: int

backend/app/services/ai_service.py

@@ -28,7 +28,9 @@ class AIService:
     async def _get_default_model(self, db: AsyncSession) -> Optional[AIModelConfig]:
         result = await db.execute(
             select(AIModelConfig).where(
-                AIModelConfig.is_default == 1, AIModelConfig.is_enabled == 1
+                AIModelConfig.usage_scope == "general",
+                AIModelConfig.is_default == 1,
+                AIModelConfig.is_enabled == 1,
             )
         )
         return result.scalar_one_or_none()

digital-avatar-app/backend/database.py

@@ -1,16 +1,30 @@
 import os
 
-from sqlalchemy import create_engine
+from sqlalchemy import create_engine, event
 from sqlalchemy.orm import sessionmaker, declarative_base, Session
 
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 DB_FILE = os.path.join(BASE_DIR, "avatar.db")
 DATABASE_URL = os.getenv("DATABASE_URL", f"sqlite:///{DB_FILE}")
 
+IS_SQLITE = DATABASE_URL.startswith("sqlite:")
 engine = create_engine(
     DATABASE_URL,
-    connect_args={"check_same_thread": False} if DATABASE_URL.startswith("sqlite:") else {},
+    connect_args={"check_same_thread": False, "timeout": 30} if IS_SQLITE else {},
 )
+
+
+if IS_SQLITE:
+    @event.listens_for(engine, "connect")
+    def _configure_sqlite_connection(dbapi_connection, _connection_record):
+        cursor = dbapi_connection.cursor()
+        try:
+            cursor.execute("PRAGMA synchronous=NORMAL")
+            cursor.execute("PRAGMA busy_timeout=30000")
+        finally:
+            cursor.close()
+
+
 SessionLocal = sessionmaker(bind=engine, autoflush=False, expire_on_commit=False)
 Base = declarative_base()
 
@@ -26,6 +40,10 @@ def get_db():
 def init_db():
     import models
 
+    if IS_SQLITE:
+        with engine.connect() as conn:
+            conn.exec_driver_sql("PRAGMA journal_mode=WAL")
+            conn.commit()
     Base.metadata.create_all(bind=engine)
 
     # 轻量迁移：为已存在的表补充新列（SQLite 不支持自动 ALTER，逐列尝试）
@@ -35,13 +53,24 @@ def init_db():
         ("knowledge_docs", "embedding_model", "VARCHAR DEFAULT ''"),
         ("knowledge_docs", "chunk_count", "INTEGER DEFAULT 0"),
         ("knowledge_docs", "vectorized_at", "TIMESTAMP"),
+        ("knowledge_docs", "error_message", "VARCHAR DEFAULT ''"),
+        ("knowledge_docs", "index_stage", "VARCHAR DEFAULT ''"),
+        ("knowledge_docs", "index_progress", "INTEGER DEFAULT 0"),
         ("avatars", "owner_id", "VARCHAR DEFAULT ''"),
         ("authorizations", "takeover_enabled", "BOOLEAN DEFAULT 0"),
         ("authorizations", "takeover_mode", "VARCHAR DEFAULT 'immediate'"),
-        ("authorizations", "takeover_delay_seconds", "INTEGER DEFAULT 30"),
+        ("authorizations", "takeover_delay_seconds", "INTEGER DEFAULT 180"),
         ("avatars", "share_token", "VARCHAR DEFAULT NULL"),
+        ("token_account", "user_id", "VARCHAR DEFAULT ''"),
+        ("token_account", "total_granted", "BIGINT DEFAULT 0"),
+        ("token_account", "total_consumed", "BIGINT DEFAULT 0"),
+        ("token_account", "created_at", "TIMESTAMP"),
+        ("token_account", "updated_at", "TIMESTAMP"),
+        ("takeover_messages", "attachment_id", "VARCHAR DEFAULT NULL"),
     )
     _normalize_optional_unique_values()
+    _normalize_takeover_delays()
+    _create_token_indexes()
 
 
 def _try_add_columns(*cols):
@@ -58,3 +87,20 @@ def _try_add_columns(*cols):
 def _normalize_optional_unique_values():
     with engine.begin() as conn:
         conn.exec_driver_sql("UPDATE avatars SET share_token = NULL WHERE share_token = ''")
+
+
+def _normalize_takeover_delays():
+    with engine.begin() as conn:
+        # The old 30-second column default was never wired into the scheduler.
+        conn.exec_driver_sql(
+            "UPDATE authorizations SET takeover_delay_seconds = 180 "
+            "WHERE takeover_delay_seconds IS NULL OR takeover_delay_seconds = 30"
+        )
+
+
+def _create_token_indexes():
+    with engine.begin() as conn:
+        conn.exec_driver_sql(
+            "CREATE UNIQUE INDEX IF NOT EXISTS ux_token_account_user_id "
+            "ON token_account(user_id) WHERE user_id <> ''"
+        )

digital-avatar-app/backend/embeddings.py

@@ -18,6 +18,14 @@ EMBED_DIM = 256
 MODEL = os.getenv("EMBEDDING_MODEL", "mock-hash-embed-v1")
 
 
+def _embedding_endpoint(api_url):
+    """Accept either an OpenAI-compatible base URL or its full endpoint."""
+    api_url = (api_url or "").strip().rstrip("/")
+    if not api_url or api_url.endswith("/embeddings"):
+        return api_url
+    return f"{api_url}/embeddings"
+
+
 def _tokenize(text):
     text = (text or "").lower()
     # 英文/数字按词，CJK 逐字（中文无空格，需拆到字级才能命中子词）
@@ -43,11 +51,11 @@ def _hash_embedding(texts, dim=EMBED_DIM):
     return vecs
 
 
-def embed(texts):
+def embed(texts, on_progress=None):
     """返回 list[list[float]]，与输入顺序一致。"""
     if not texts:
         return []
-    api_url = os.getenv("EMBEDDING_API_URL")
+    api_url = _embedding_endpoint(os.getenv("EMBEDDING_API_URL"))
     if api_url:
         api_key = os.getenv("EMBEDDING_API_KEY", "")
         model = os.getenv("EMBEDDING_MODEL", "text-embedding-3-small")
@@ -56,6 +64,7 @@ def embed(texts):
         except ValueError:
             batch_size = 10
         embeddings = []
+        total = len(texts)
         for start in range(0, len(texts), batch_size):
             batch = texts[start:start + batch_size]
             payload = json.dumps({"input": batch, "model": model}).encode("utf-8")
@@ -76,8 +85,13 @@ def embed(texts):
             if len(items) != len(batch):
                 raise ValueError("embedding response count does not match request")
             embeddings.extend(item["embedding"] for item in items)
+            if on_progress:
+                on_progress(len(embeddings), total)
         return embeddings
-    return _hash_embedding(texts)
+    vectors = _hash_embedding(texts)
+    if on_progress:
+        on_progress(len(vectors), len(texts))
+    return vectors
 
 
 def cosine(a, b):

digital-avatar-app/backend/main.py

@@ -8,7 +8,7 @@ from apscheduler.schedulers.asyncio import AsyncIOScheduler
 from apscheduler.triggers.interval import IntervalTrigger
 
 from database import init_db, SessionLocal
-from models import Avatar, Authorization, Organization, TokenAccount, TokenPlan
+from models import Avatar, Authorization, Organization, TokenAccount, TokenPlan, User
 from fastapi.staticfiles import StaticFiles
 import routers.avatars
 import routers.tokens
@@ -19,10 +19,14 @@ import routers.huihui_auth
 import routers.chat
 import routers.takeover
 from responses import ok
+from services.chat_attachment_service import purge_expired_chat_attachments
+from services.knowledge_vectorizer import knowledge_vectorizer
+from services.token_billing import DEFAULT_TOKEN_GRANT, release_stale_reservations
 
 logger = logging.getLogger(__name__)
 
 takeover_scheduler = None
+maintenance_scheduler = None
 
 app = FastAPI(title="会会数字分身 API", version="1.0.0")
 
@@ -56,17 +60,29 @@ def health():
 def seed():
     db = SessionLocal()
     try:
-        if db.query(TokenAccount).first() is None:
-            db.add(TokenAccount(balance=1250))
-
-        if db.query(TokenPlan).count() == 0:
-            plans = [
-                TokenPlan(id="1", name="新手体验", amount=1000, price=9.9, desc="新手体验"),
-                TokenPlan(id="2", name="热门套餐", amount=5000, price=39.9, badge="热门"),
-                TokenPlan(id="3", name="超值套餐", amount=12000, price=89.9, badge="超值"),
-                TokenPlan(id="4", name="企业推荐", amount=30000, price=199, badge="企业推荐", desc="适合高频使用"),
+        plan_specs = [
+            {"id": "1", "name": "基础套餐", "amount": 2_000_000, "price": 10, "badge": "", "desc": "2M 积分"},
+            {"id": "2", "name": "标准套餐", "amount": 20_000_000, "price": 100, "badge": "常用", "desc": "20M 积分"},
+            {"id": "3", "name": "专业套餐", "amount": 250_000_000, "price": 1000, "badge": "加赠25%", "desc": "250M 积分"},
+            {"id": "4", "name": "企业套餐", "amount": 2_500_000_000, "price": 10000, "badge": "企业推荐", "desc": "2500M 积分"},
         ]
-            db.add_all(plans)
+        for spec in plan_specs:
+            plan = db.query(TokenPlan).filter(TokenPlan.id == spec["id"]).first()
+            if plan is None:
+                db.add(TokenPlan(**spec))
+            else:
+                for key, value in spec.items():
+                    setattr(plan, key, value)
+
+        for user in db.query(User).all():
+            account = db.query(TokenAccount).filter(TokenAccount.user_id == user.id).first()
+            if account is None:
+                db.add(TokenAccount(
+                    user_id=user.id,
+                    balance=DEFAULT_TOKEN_GRANT,
+                    total_granted=DEFAULT_TOKEN_GRANT,
+                    total_consumed=0,
+                ))
 
         if db.query(Avatar).count() == 0:
             avatar = Avatar(
@@ -105,6 +121,7 @@ def seed():
             db.add_all(orgs)
 
         db.commit()
+        release_stale_reservations(db)
     finally:
         db.close()
 
@@ -115,9 +132,19 @@ def on_startup():
 
     init_db()
     seed()
+    knowledge_vectorizer.start()
 
     # Release stale resources when startup is invoked again by a reload/test.
     stop_takeover_scheduler()
+    stop_maintenance_scheduler()
+    try:
+        start_maintenance_scheduler()
+    except Exception as exc:
+        stop_maintenance_scheduler()
+        logger.warning(
+            "Failed to initialize chat attachment cleanup, app will continue: %s",
+            exc,
+        )
 
     # --- Takeover scheduler ---
     try:
@@ -138,7 +165,14 @@ def on_startup():
         boxim_client = BoxIMClient(boxim_config)
 
         from services.takeover_service import TakeoverService
-        takeover_service = TakeoverService(SessionLocal, boxim_client)
+        takeover_service = TakeoverService(
+            SessionLocal,
+            boxim_client,
+            poll_concurrency=int(os.getenv("BOXIM_POLL_CONCURRENCY", "8")),
+            max_message_age_seconds=int(
+                os.getenv("BOXIM_MAX_MESSAGE_AGE_SECONDS", "600")
+            ),
+        )
 
         poll_interval = max(0.5, float(os.getenv("BOXIM_POLL_INTERVAL_SECONDS", "1")))
         takeover_scheduler = AsyncIOScheduler()
@@ -182,6 +216,51 @@ def stop_takeover_scheduler():
         finally:
             takeover_scheduler = None
 
+
+def purge_expired_chat_attachments_job():
+    db = SessionLocal()
+    try:
+        count = purge_expired_chat_attachments(db)
+        if count:
+            logger.info("Purged %s expired chat image attachment(s)", count)
+    except Exception as exc:
+        db.rollback()
+        logger.warning("Failed to purge expired chat image attachments: %s", exc)
+    finally:
+        db.close()
+
+
+def start_maintenance_scheduler():
+    global maintenance_scheduler
+
+    purge_expired_chat_attachments_job()
+    interval_minutes = max(
+        5, min(1440, int(os.getenv("CHAT_ATTACHMENT_CLEANUP_MINUTES", "60")))
+    )
+    maintenance_scheduler = AsyncIOScheduler()
+    maintenance_scheduler.add_job(
+        purge_expired_chat_attachments_job,
+        trigger=IntervalTrigger(minutes=interval_minutes),
+        id="chat_attachment_cleanup",
+        max_instances=1,
+        coalesce=True,
+    )
+    maintenance_scheduler.start()
+
+
+def stop_maintenance_scheduler():
+    global maintenance_scheduler
+
+    if maintenance_scheduler is not None:
+        try:
+            if maintenance_scheduler.running:
+                maintenance_scheduler.shutdown(wait=False)
+        except Exception as exc:
+            logger.warning("Failed to stop maintenance scheduler cleanly: %s", exc)
+        finally:
+            maintenance_scheduler = None
+
 @app.on_event("shutdown")
 def on_shutdown():
     stop_takeover_scheduler()
+    stop_maintenance_scheduler()

digital-avatar-app/backend/models.py

@@ -1,6 +1,7 @@
 import uuid
 
 from sqlalchemy import (
+    BigInteger,
     Boolean,
     Column,
     DateTime,
@@ -66,7 +67,7 @@ class Authorization(Base):
     status = Column(String, default="active")  # active | inactive
     takeover_enabled = Column(Boolean, default=False)  # 是否开启分身接管
     takeover_mode = Column(String, default="immediate")  # immediate | delayed
-    takeover_delay_seconds = Column(Integer, default=30)  # 延迟秒数
+    takeover_delay_seconds = Column(Integer, default=180)  # 延迟秒数，默认 3 分钟
     created_at = Column(DateTime, server_default=func.now())
 
     def to_dict(self):
@@ -119,13 +120,14 @@ class TakeoverMessage(Base):
     direction = Column(String, nullable=False)  # incoming | outgoing
     message_type = Column(Integer, default=0)
     content = Column(Text, default="")
+    attachment_id = Column(String, nullable=True)
     is_avatar = Column(Boolean, default=False)
     send_time = Column(DateTime, nullable=False)
     created_at = Column(DateTime, server_default=func.now())
 
 
 class TakeoverReplyTask(Base):
-    """Restart-safe three-second BOXIM reply task."""
+    """Restart-safe delayed BOXIM reply task."""
 
     __tablename__ = "takeover_reply_tasks"
     __table_args__ = (
@@ -187,7 +189,10 @@ class KnowledgeDoc(Base):
     file_type = Column(String, default="")  # pdf | doc | docx | xlsx
     file_size = Column(Integer, default=0)
     file_url = Column(String, default="")
-    status = Column(String, default="uploaded")  # uploaded | parsing | ready
+    status = Column(String, default="uploaded")  # uploaded | parsing | ready | failed
+    error_message = Column(String, default="")  # 建立索引失败原因
+    index_stage = Column(String, default="")  # queued | extracting | chunking | embedding | ready | failed
+    index_progress = Column(Integer, default=0)  # 0-100
     vectorized = Column(Boolean, default=False)  # 是否已向量化
     embedding_model = Column(String, default="")  # 向量模型标识
     chunk_count = Column(Integer, default=0)  # 切片数量
@@ -203,6 +208,9 @@ class KnowledgeDoc(Base):
             "fileSize": self.file_size,
             "fileUrl": self.file_url,
             "status": self.status,
+            "errorMessage": self.error_message or "",
+            "indexStage": self.index_stage or "",
+            "indexProgress": int(self.index_progress or 0),
             "vectorized": bool(self.vectorized),
             "embeddingModel": self.embedding_model,
             "chunkCount": self.chunk_count,
@@ -256,17 +264,83 @@ class KnowledgeChunk(Base):
         }
 
 
+class ChatAttachment(Base):
+    """Private, avatar-scoped result of one chat image analysis."""
+
+    __tablename__ = "chat_attachments"
+    id = Column(String, primary_key=True, default=lambda: uuid.uuid4().hex)
+    avatar_id = Column(String, nullable=False, default="", index=True)
+    uploader_kind = Column(String, default="owner")  # owner | public | boxim
+    filename = Column(String, default="")
+    mime_type = Column(String, default="")
+    file_size = Column(Integer, default=0)
+    status = Column(String, default="processing")  # processing | ready | failed
+    category = Column(String, default="general_image")
+    summary = Column(Text, default="")
+    extracted_text = Column(Text, default="")
+    structured_data = Column(JSON, default=dict)
+    warning = Column(Text, default="")
+    vision_model = Column(String, default="")
+    ocr_model = Column(String, default="")
+    used_at = Column(DateTime)
+    expires_at = Column(DateTime, nullable=False)
+    created_at = Column(DateTime, server_default=func.now())
+
+    def to_dict(self):
+        return {
+            "id": self.id,
+            "avatarId": self.avatar_id,
+            "filename": self.filename,
+            "mimeType": self.mime_type,
+            "fileSize": self.file_size,
+            "status": self.status,
+            "category": self.category,
+            "summary": self.summary,
+            "warning": self.warning,
+            "expiresAt": _iso(self.expires_at),
+            "createdAt": _iso(self.created_at),
+        }
+
+
 class TokenAccount(Base):
     __tablename__ = "token_account"
     id = Column(Integer, primary_key=True)
-    balance = Column(Integer, default=1250)
+    user_id = Column(String, nullable=False, default="", index=True)
+    balance = Column(BigInteger, default=1_000_000)
+    total_granted = Column(BigInteger, default=1_000_000)
+    total_consumed = Column(BigInteger, default=0)
+    created_at = Column(DateTime, server_default=func.now())
+    updated_at = Column(DateTime, server_default=func.now(), onupdate=func.now())
+
+
+class TokenUsage(Base):
+    __tablename__ = "token_usage"
+    __table_args__ = (
+        Index("ix_token_usage_user_created", "user_id", "created_at"),
+        Index("ix_token_usage_avatar_created", "avatar_id", "created_at"),
+    )
+
+    id = Column(String, primary_key=True, default=lambda: uuid.uuid4().hex)
+    user_id = Column(String, nullable=False, index=True)
+    avatar_id = Column(String, nullable=False, default="", index=True)
+    source = Column(String, nullable=False, default="chat")
+    model = Column(String, default="")
+    status = Column(String, nullable=False, default="reserved")
+    reserved_tokens = Column(BigInteger, default=0)
+    prompt_tokens = Column(BigInteger, default=0)
+    completion_tokens = Column(BigInteger, default=0)
+    total_tokens = Column(BigInteger, default=0)
+    balance_after = Column(BigInteger, default=0)
+    failure_reason = Column(String, default="")
+    created_at = Column(DateTime, server_default=func.now())
+    settled_at = Column(DateTime)
 
 
 class TokenPlan(Base):
     __tablename__ = "token_plans"
     id = Column(String, primary_key=True)
     name = Column(String, default="")
-    amount = Column(Integer, default=0)
+    amount = Column(BigInteger, default=0)
     price = Column(Float, default=0)
     badge = Column(String, default="")
     desc = Column(String, default="")
@@ -282,6 +356,47 @@ class TokenPlan(Base):
         }
 
 
+class TokenPaymentOrder(Base):
+    __tablename__ = "token_payment_orders"
+
+    id = Column(String, primary_key=True, default=lambda: uuid.uuid4().hex)
+    order_no = Column(String, nullable=False, unique=True, index=True)
+    user_id = Column(String, nullable=False, index=True)
+    plan_id = Column(String, nullable=False)
+    payment_method = Column(String, nullable=False)
+    pay_type = Column(String, nullable=False)
+    pay_way = Column(String, nullable=False)
+    points_amount = Column(BigInteger, nullable=False)
+    price_cents = Column(Integer, nullable=False)
+    status = Column(String, nullable=False, default="pending", index=True)
+    provider_order_id = Column(String, default="")
+    provider_order_no = Column(String, default="")
+    provider_status = Column(String, default="")
+    pay_message = Column(Text, default="")
+    failure_reason = Column(String, default="")
+    created_at = Column(DateTime, server_default=func.now())
+    updated_at = Column(DateTime, server_default=func.now(), onupdate=func.now())
+    paid_at = Column(DateTime)
+
+    def to_dict(self):
+        return {
+            "id": self.id,
+            "orderNo": self.order_no,
+            "planId": self.plan_id,
+            "paymentMethod": self.payment_method,
+            "payType": self.pay_type,
+            "payWay": self.pay_way,
+            "pointsAmount": self.points_amount,
+            "price": self.price_cents / 100,
+            "status": self.status,
+            "providerStatus": self.provider_status,
+            "payMessage": self.pay_message,
+            "failureReason": self.failure_reason,
+            "createdAt": _iso(self.created_at),
+            "paidAt": _iso(self.paid_at),
+        }
+
+
 class User(Base):
     """会会用户 ↔ 本地用户体系映射（短信验证码登录落库）"""
 

digital-avatar-app/backend/requirements.txt

@@ -8,3 +8,4 @@ pypdf
 python-docx
 openpyxl
 apscheduler>=3.10
+Pillow>=10.4

digital-avatar-app/backend/routers/authorizations.py

@@ -2,7 +2,7 @@ from fastapi import APIRouter, Body, Depends, Header, HTTPException
 from sqlalchemy.orm import Session
 
 from database import get_db
-from models import Authorization, TakeoverCursor, TakeoverReplyTask
+from models import Authorization, Avatar, TakeoverCursor, TakeoverReplyTask
 from responses import fail, ok
 from routers.avatars import _require_owned_avatar
 
@@ -14,6 +14,10 @@ ALLOWED_PERMISSIONS = set(PERMISSION_ORDER)
 AVATAR_PERMISSION_ORDER = PERMISSION_ORDER
 AVATAR_PERMISSION_KEY = "authorizationPermissions"
 DEFAULT_AVATAR_PERMISSIONS = ["friend", "chat"]
+TAKEOVER_DELAY_KEY = "takeoverReplyDelaySeconds"
+DEFAULT_TAKEOVER_DELAY_SECONDS = 180
+MIN_TAKEOVER_DELAY_SECONDS = 3
+MAX_TAKEOVER_DELAY_SECONDS = 86_400
 LEGACY_PERMISSION_MAP = {
     "read": "browse",
     "reply": "chat",
@@ -91,9 +95,62 @@ def _permission_settings_payload(avatar) -> dict:
     return {
         "avatarId": avatar.id,
         "permissions": _stored_avatar_permissions(avatar),
+        "takeoverReplyDelaySeconds": _stored_takeover_delay(avatar),
     }
 
 
+def _stored_takeover_delay(avatar) -> int:
+    raw = (avatar.config or {}).get(TAKEOVER_DELAY_KEY, DEFAULT_TAKEOVER_DELAY_SECONDS)
+    if isinstance(raw, bool):
+        return DEFAULT_TAKEOVER_DELAY_SECONDS
+    try:
+        delay = int(raw)
+    except (TypeError, ValueError):
+        return DEFAULT_TAKEOVER_DELAY_SECONDS
+    if not MIN_TAKEOVER_DELAY_SECONDS <= delay <= MAX_TAKEOVER_DELAY_SECONDS:
+        return DEFAULT_TAKEOVER_DELAY_SECONDS
+    return delay
+
+
+def _validate_takeover_delay(value) -> int:
+    if isinstance(value, bool) or not isinstance(value, int):
+        raise ValueError("自动回复等待时间必须是整数秒")
+    if not MIN_TAKEOVER_DELAY_SECONDS <= value <= MAX_TAKEOVER_DELAY_SECONDS:
+        raise ValueError("自动回复等待时间需在 3 秒到 24 小时之间")
+    return value
+
+
+def _disable_other_takeovers(db: Session, avatar) -> list[str]:
+    disabled_ids = []
+    others = (
+        db.query(Avatar)
+        .filter(Avatar.owner_id == avatar.owner_id, Avatar.id != avatar.id)
+        .all()
+    )
+    for other in others:
+        permissions = _stored_avatar_permissions(other)
+        if "takeover" not in permissions:
+            continue
+        other.config = {
+            **(other.config or {}),
+            AVATAR_PERMISSION_KEY: [item for item in permissions if item != "takeover"],
+        }
+        disabled_ids.append(other.id)
+        tasks = (
+            db.query(TakeoverReplyTask)
+            .filter(
+                TakeoverReplyTask.avatar_id == other.id,
+                TakeoverReplyTask.status.in_(("pending", "generating", "ready", "sending")),
+            )
+            .all()
+        )
+        for task in tasks:
+            task.status = "cancelled"
+            task.cancel_reason = "another_avatar_takeover_enabled"
+            task.locked_at = None
+    return disabled_ids
+
+
 def _require_authorization(db: Session, avatar_id: str, authorization_id: str) -> Authorization:
     authorization = (
         db.query(Authorization)
@@ -144,10 +201,19 @@ def update_permission_settings(
     db: Session = Depends(get_db),
 ):
     avatar = _require_owned_avatar(db, avatar_id, authorization)
-    if "permissions" not in payload:
-        return fail("缺少 permissions", 400)
+    if "permissions" not in payload and TAKEOVER_DELAY_KEY not in payload:
+        return fail("缺少授权设置", 400)
     try:
-        permissions = _normalize_avatar_permissions(payload["permissions"])
+        permissions = (
+            _normalize_avatar_permissions(payload["permissions"])
+            if "permissions" in payload
+            else _stored_avatar_permissions(avatar)
+        )
+        takeover_delay = (
+            _validate_takeover_delay(payload[TAKEOVER_DELAY_KEY])
+            if TAKEOVER_DELAY_KEY in payload
+            else _stored_takeover_delay(avatar)
+        )
     except ValueError as exc:
         return fail(str(exc), 400)
 
@@ -155,7 +221,9 @@ def update_permission_settings(
     avatar.config = {
         **(avatar.config or {}),
         AVATAR_PERMISSION_KEY: permissions,
+        TAKEOVER_DELAY_KEY: takeover_delay,
     }
+    disabled_avatar_ids = _disable_other_takeovers(db, avatar) if "takeover" in permissions else []
     cursor = db.query(TakeoverCursor).filter(TakeoverCursor.avatar_id == avatar.id).first()
     if cursor and "takeover" in permissions and "takeover" not in previous_permissions:
         cursor.initialized = False
@@ -179,7 +247,9 @@ def update_permission_settings(
             task.locked_at = None
     db.commit()
     db.refresh(avatar)
-    return ok(_permission_settings_payload(avatar), "授权设置已保存")
+    response = _permission_settings_payload(avatar)
+    response["disabledAvatarIds"] = disabled_avatar_ids
+    return ok(response, "授权设置已保存")
 
 
 @router.get("/avatar/{avatar_id}/authorizations")
@@ -240,7 +310,7 @@ def create_auth(
         status="active",
         takeover_enabled=False,
         takeover_mode="immediate",
-        takeover_delay_seconds=30,
+        takeover_delay_seconds=DEFAULT_TAKEOVER_DELAY_SECONDS,
     )
     db.add(item)
     db.commit()

digital-avatar-app/backend/routers/avatars.py

@@ -6,7 +6,17 @@ from sqlalchemy.orm import Session
 
 from database import get_db
 from routers.knowledge import UPLOAD_DIR
-from models import Avatar, KnowledgeDoc, KnowledgeChunk, QAPair, Authorization, User
+from models import (
+    Authorization,
+    Avatar,
+    KnowledgeChunk,
+    KnowledgeDoc,
+    QAPair,
+    TakeoverCursor,
+    TakeoverMessage,
+    TakeoverReplyTask,
+    User,
+)
 from responses import ok, fail
 
 router = APIRouter(tags=["分身"])
@@ -74,18 +84,21 @@ def list_avatars(page: int = 1, limit: int = 20, authorization: str = Header(Non
 
 
 @router.get("/avatar/{avatar_id}")
-def get_avatar(avatar_id: str, db: Session = Depends(get_db)):
-    a = db.query(Avatar).filter(Avatar.id == avatar_id).first()
-    if not a:
-        return fail("分身不存在", 404)
-    return ok(a.to_dict())
+def get_avatar(
+    avatar_id: str,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    return ok(_require_owned_avatar(db, avatar_id, authorization).to_dict())
 
 
 @router.post("/avatar")
 def create_avatar(payload: dict = Body(...), authorization: str = Header(None), db: Session = Depends(get_db)):
     user = _resolve_user(authorization, db)
+    if not user:
+        raise HTTPException(status_code=401, detail="未登录")
     a = Avatar(
-        owner_id=user.huihui_user_id if user else "",
+        owner_id=user.huihui_user_id,
         name=payload.get("name", "未命名分身"),
         display_name=payload.get("displayName", "") or payload.get("display_name", ""),
         description=payload.get("description", ""),
@@ -102,10 +115,13 @@ def create_avatar(payload: dict = Body(...), authorization: str = Header(None),
 
 
 @router.put("/avatar/{avatar_id}")
-def update_avatar(avatar_id: str, payload: dict = Body(...), db: Session = Depends(get_db)):
-    a = db.query(Avatar).filter(Avatar.id == avatar_id).first()
-    if not a:
-        return fail("分身不存在", 404)
+def update_avatar(
+    avatar_id: str,
+    payload: dict = Body(...),
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    a = _require_owned_avatar(db, avatar_id, authorization)
     mapping = {
         "displayName": "display_name",
         "photoUrl": "photo_url",
@@ -114,22 +130,32 @@ def update_avatar(avatar_id: str, payload: dict = Body(...), db: Session = Depen
     for key in ("name", "displayName", "description", "photoUrl", "emoji", "status", "tokenBalance", "config"):
         if key in payload:
             col = mapping.get(key, key)
-            setattr(a, col, payload[key])
+            value = payload[key]
+            if key == "config":
+                if not isinstance(value, dict):
+                    return fail("分身配置格式不正确", 400)
+                value = {**(a.config or {}), **value}
+            setattr(a, col, value)
     db.commit()
     db.refresh(a)
     return ok(a.to_dict())
 
 
 @router.delete("/avatar/{avatar_id}")
-def delete_avatar(avatar_id: str, db: Session = Depends(get_db)):
-    a = db.query(Avatar).filter(Avatar.id == avatar_id).first()
-    if not a:
-        return fail("分身不存在", 404)
+def delete_avatar(
+    avatar_id: str,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    a = _require_owned_avatar(db, avatar_id, authorization)
     # 级联清理关联数据，避免孤儿记录
     db.query(KnowledgeDoc).filter(KnowledgeDoc.avatar_id == avatar_id).delete()
     db.query(KnowledgeChunk).filter(KnowledgeChunk.avatar_id == avatar_id).delete()
     db.query(QAPair).filter(QAPair.avatar_id == avatar_id).delete()
     db.query(Authorization).filter(Authorization.avatar_id == avatar_id).delete()
+    db.query(TakeoverReplyTask).filter(TakeoverReplyTask.avatar_id == avatar_id).delete()
+    db.query(TakeoverMessage).filter(TakeoverMessage.avatar_id == avatar_id).delete()
+    db.query(TakeoverCursor).filter(TakeoverCursor.avatar_id == avatar_id).delete()
     db.delete(a)
     db.commit()
     return ok({"success": True})

digital-avatar-app/backend/routers/huihui_auth.py

@@ -29,6 +29,7 @@ _CN_TZ = timezone(timedelta(hours=8))
 from database import get_db
 from models import Avatar, TakeoverCursor, TakeoverMessage, TakeoverReplyTask, User
 from responses import ok, fail
+from services.boxim_client import BoxIMClient, BoxIMError
 
 router = APIRouter(tags=["会会账号"])
 
@@ -111,6 +112,19 @@ def _cfg_ready() -> bool:
     return bool(AUTH_BASE_URL and APP_ID and ACCESS_ID and ACCESS_SECRET)
 
 
+def _create_boxim_client() -> BoxIMClient:
+    return BoxIMClient({
+        "HUIHUI_PLATFORM_BASE_URL": os.getenv(
+            "HUIHUI_PLATFORM_BASE_URL", "https://open.99hui.com/api"
+        ),
+        "BOXIM_API_BASE_URL": os.getenv("BOXIM_API_BASE_URL", "https://im.99hui.com/api"),
+        "HUIHUI_APP_ID": APP_ID,
+        "HUIHUI_ACCESS_ID": ACCESS_ID,
+        "HUIHUI_ACCESS_SECRET": ACCESS_SECRET,
+        "BOXIM_TIMEOUT_SECONDS": os.getenv("BOXIM_TIMEOUT_SECONDS", "20"),
+    })
+
+
 def _call_huihui(path: str, params: dict, as_query: bool = False):
     """调用会会接口，返回 (ok: bool, payload: dict, http_status: int)"""
     url = f"{AUTH_BASE_URL}{path}"
@@ -281,6 +295,47 @@ def pwd_login(body: dict = Body(...), db: Session = Depends(get_db)):
     })
 
 
+@router.post("/huihui/token/login")
+async def token_login(body: dict = Body(...), db: Session = Depends(get_db)):
+    """Validate a production Huihui token through BOXIM and issue an app session."""
+    huihui_token = (body.get("token") or "").strip()
+    if not huihui_token or len(huihui_token) > 8192:
+        return fail("会会登录凭证无效或已过期", 401)
+    if not _cfg_ready():
+        return fail("会会登录服务未配置", 500)
+
+    client = _create_boxim_client()
+    try:
+        token_data = await client.exchange_access_token(huihui_token)
+        profile = await client.get_self(token_data["accessToken"])
+    except BoxIMError as exc:
+        if exc.auth_error:
+            return fail("会会登录凭证无效或已过期", 401)
+        return fail("会会登录服务暂时不可用，请稍后重试", 502)
+
+    # BOXIM's id is its internal IM id. Account ownership must use huihuiUserId.
+    huihui_user_id = str(profile.get("huihuiUserId") or "").strip()
+    if not huihui_user_id:
+        return fail("会会未返回用户标识", 502)
+
+    phone = str(_pick(profile, "mobile", "phone", default="")).strip()
+    nickname = str(_pick(profile, "nickName", "nickname", "name", "userName", default="")).strip()
+    avatar_url = str(
+        _pick(profile, "headImage", "headImageThumb", "avatar", "avatarUrl", default="")
+    ).strip()
+    return _issue_session(
+        db,
+        phone,
+        {
+            "userId": huihui_user_id,
+            "nickname": nickname,
+            "avatarUrl": avatar_url,
+            "token": huihui_token,
+        },
+        reuse_existing_session=True,
+    )
+
+
 def _transfer_avatar_ownership(db: Session, old_owner_id: str, new_owner_id: str) -> int:
     """Move one user's avatar-owned data to a replacement Huihui identity."""
     if not old_owner_id or old_owner_id == new_owner_id:
@@ -332,7 +387,13 @@ def _find_or_link_user(db: Session, phone: str, huihui_user_id: str) -> User:
     return user
 
 
-def _issue_session(db: Session, phone: str, info: dict):
+def _issue_session(
+    db: Session,
+    phone: str,
+    info: dict,
+    *,
+    reuse_existing_session: bool = False,
+):
     """建/链本地用户并签发本系统会话 token"""
     huihui_user_id = info.get("userId", "")
     user = _find_or_link_user(db, phone, huihui_user_id)
@@ -343,12 +404,16 @@ def _issue_session(db: Session, phone: str, info: dict):
     if info.get("avatarUrl"):
         user.avatar_url = info["avatarUrl"]
     user.huihui_token = info.get("token", "")
+    if not reuse_existing_session or not user.app_token:
         user.app_token = uuid.uuid4().hex
     user.last_login_at = datetime.now()
     db.add(user)
     db.commit()
     db.refresh(user)
 
+    from services.token_billing import get_or_create_account
+    get_or_create_account(db, user.id)
+
     return ok({
         "token": user.app_token,
         "user": user.to_dict(),
@@ -356,7 +421,6 @@ def _issue_session(db: Session, phone: str, info: dict):
             "userId": huihui_user_id,
             "nickname": info.get("nickname", ""),
             "avatarUrl": info.get("avatarUrl", ""),
-            "token": info.get("token", ""),
         },
     })
 

digital-avatar-app/backend/routers/knowledge.py

@@ -1,7 +1,8 @@
 import os
 import json
+import shutil
+import time
 import uuid
-from datetime import datetime, timezone
 
 from fastapi import APIRouter, UploadFile, File, Depends, Header, HTTPException
 from pydantic import BaseModel
@@ -11,15 +12,19 @@ from database import get_db
 from models import KnowledgeDoc, QAPair, KnowledgeChunk, Avatar, User
 from responses import ok, fail
 import embeddings
+from services.knowledge_vectorizer import knowledge_vectorizer
 
 router = APIRouter()
-
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 UPLOAD_DIR = os.path.abspath(os.getenv("UPLOAD_DIR", os.path.join(BASE_DIR, "uploads")))
 os.makedirs(UPLOAD_DIR, exist_ok=True)
 
 ALLOWED_EXT = {".md", ".txt", ".pdf", ".doc", ".docx", ".xlsx"}
-MAX_UPLOAD_BYTES = 10 * 1024 * 1024
+MAX_UPLOAD_BYTES = 50 * 1024 * 1024
+UPLOAD_CHUNK_BYTES = 1024 * 1024
+MULTIPART_CHUNK_BYTES = 5 * 1024 * 1024
+MULTIPART_ROOT = ".multipart"
+MULTIPART_TTL_SECONDS = 24 * 60 * 60
 
 
 class QAIn(BaseModel):
@@ -32,6 +37,74 @@ class EnabledIn(BaseModel):
     enabled: bool = True
 
 
+class MultipartUploadIn(BaseModel):
+    filename: str
+    fileSize: int
+    totalChunks: int
+
+
+def _validate_document(filename: str, file_size: int):
+    ext = os.path.splitext(filename or "")[1].lower()
+    if ext not in ALLOWED_EXT:
+        return None, f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx"
+    if file_size <= 0:
+        return None, "文件内容不能为空"
+    if file_size > MAX_UPLOAD_BYTES:
+        return None, "文件不能超过 50MB"
+    return ext, ""
+
+
+def _multipart_dir(avatar_id: str, upload_id: str) -> str:
+    safe_avatar_id = os.path.basename(avatar_id)
+    safe_upload_id = os.path.basename(upload_id)
+    if (
+        safe_avatar_id != avatar_id
+        or safe_upload_id != upload_id
+        or len(upload_id) != 32
+        or any(character not in "0123456789abcdef" for character in upload_id)
+    ):
+        raise HTTPException(status_code=400, detail="上传标识无效")
+    return os.path.join(UPLOAD_DIR, MULTIPART_ROOT, safe_avatar_id, safe_upload_id)
+
+
+def _purge_stale_multipart_uploads(avatar_id: str):
+    avatar_upload_root = os.path.join(UPLOAD_DIR, MULTIPART_ROOT, os.path.basename(avatar_id))
+    if not os.path.isdir(avatar_upload_root):
+        return
+    cutoff = time.time() - MULTIPART_TTL_SECONDS
+    for entry in os.scandir(avatar_upload_root):
+        if entry.is_dir(follow_symlinks=False) and entry.stat(follow_symlinks=False).st_mtime < cutoff:
+            shutil.rmtree(entry.path, ignore_errors=True)
+
+
+def _read_multipart_metadata(avatar_id: str, upload_id: str) -> tuple[str, dict]:
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    metadata_path = os.path.join(upload_dir, "metadata.json")
+    if not os.path.isfile(metadata_path):
+        raise HTTPException(status_code=404, detail="上传任务不存在或已过期")
+    with open(metadata_path, "r", encoding="utf-8") as stream:
+        return upload_dir, json.load(stream)
+
+
+def _create_knowledge_doc(db: Session, avatar_id: str, filename: str, ext: str, file_size: int, stored: str):
+    doc = KnowledgeDoc(
+        id=uuid.uuid4().hex,
+        avatar_id=avatar_id,
+        filename=filename,
+        file_type=ext.lstrip("."),
+        file_size=file_size,
+        file_url=f"/api/files/{avatar_id}/{stored}",
+        status="parsing",
+        index_stage="queued",
+        index_progress=0,
+    )
+    db.add(doc)
+    db.commit()
+    db.refresh(doc)
+    knowledge_vectorizer.enqueue(doc.id)
+    return doc
+
+
 def _doc_payload(doc: KnowledgeDoc) -> dict:
     payload = doc.to_dict()
     stored_name = os.path.basename(doc.file_url or "")
@@ -75,60 +148,172 @@ def list_docs(avatar_id: str, authorization: str = Header(None), db: Session = D
 @router.post("/avatar/{avatar_id}/knowledge/docs")
 async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization: str = Header(None), db: Session = Depends(get_db)):
     _require_owned_avatar(db, avatar_id, authorization)
-    ext = os.path.splitext(file.filename or "")[1].lower()
-    if ext not in ALLOWED_EXT:
-        return fail(f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx", code=400)
+    ext, validation_error = _validate_document(file.filename or "", 1)
+    if validation_error:
+        return fail(validation_error, code=400)
     avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
     os.makedirs(avatar_dir, exist_ok=True)
     stored = f"{uuid.uuid4().hex}{ext}"
     path = os.path.join(avatar_dir, stored)
-    content = await file.read()
-    if len(content) > MAX_UPLOAD_BYTES:
-        return fail("文件不能超过 10MB", code=400)
-    with open(path, "wb") as f:
-        f.write(content)
-    doc = KnowledgeDoc(
-        avatar_id=avatar_id,
-        filename=file.filename,
-        file_type=ext.lstrip("."),
-        file_size=len(content),
-        file_url=f"/api/files/{avatar_id}/{stored}",
-        status="parsing",
-    )
-    db.add(doc)
-    db.commit()
-    db.refresh(doc)
-
-    # 向量化：抽取文本 -> 分块 -> 调第三方/本地嵌入 -> 存切片
+    file_size = 0
     try:
-        text = embeddings.extract_text(path, ext)
-        chunks = embeddings.chunk_text(text)
-        if chunks:
-            vectors = embeddings.embed(chunks)
-            for i, (c, v) in enumerate(zip(chunks, vectors)):
-                db.add(
-                    KnowledgeChunk(
-                        doc_id=doc.id,
-                        avatar_id=avatar_id,
-                        content=c,
-                        vector=json.dumps(v),
-                        chunk_index=i,
-                        embedding_model=embeddings.MODEL,
-                    )
-                )
-            doc.vectorized = True
-            doc.embedding_model = embeddings.MODEL
-            doc.chunk_count = len(chunks)
-            doc.vectorized_at = datetime.now(timezone.utc)
-        doc.status = "ready"
-        db.commit()
-        db.refresh(doc)
-    except Exception as e:
-        print("vectorize failed:", e)
-        doc.status = "ready"  # 上传成功但向量化失败，仍可展示
-        db.commit()
-        db.refresh(doc)
+        # Stream large files to disk so a 100MB upload does not occupy 100MB RAM.
+        with open(path, "wb") as f:
+            while chunk := await file.read(UPLOAD_CHUNK_BYTES):
+                file_size += len(chunk)
+                if file_size > MAX_UPLOAD_BYTES:
+                    raise ValueError("文件不能超过 50MB")
+                f.write(chunk)
+    except ValueError as exc:
+        if os.path.exists(path):
+            os.remove(path)
+        return fail(str(exc), code=400)
+    if file_size == 0:
+        if os.path.exists(path):
+            os.remove(path)
+        return fail("文件内容不能为空", code=400)
 
+    doc = _create_knowledge_doc(db, avatar_id, file.filename or stored, ext, file_size, stored)
+    return ok(_doc_payload(doc))
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads")
+def create_multipart_upload(
+    avatar_id: str,
+    body: MultipartUploadIn,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    ext, validation_error = _validate_document(body.filename, body.fileSize)
+    if validation_error:
+        return fail(validation_error, code=400)
+    expected_chunks = (body.fileSize + MULTIPART_CHUNK_BYTES - 1) // MULTIPART_CHUNK_BYTES
+    if body.totalChunks != expected_chunks:
+        return fail("文件分片数量不正确", code=400)
+
+    _purge_stale_multipart_uploads(avatar_id)
+    upload_id = uuid.uuid4().hex
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    os.makedirs(upload_dir, exist_ok=False)
+    metadata = {
+        "filename": body.filename,
+        "fileSize": body.fileSize,
+        "totalChunks": body.totalChunks,
+        "extension": ext,
+    }
+    with open(os.path.join(upload_dir, "metadata.json"), "w", encoding="utf-8") as stream:
+        json.dump(metadata, stream, ensure_ascii=False)
+    return ok({"uploadId": upload_id, "chunkSize": MULTIPART_CHUNK_BYTES})
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/chunks/{chunk_index}")
+async def upload_multipart_chunk(
+    avatar_id: str,
+    upload_id: str,
+    chunk_index: int,
+    file: UploadFile = File(...),
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    if chunk_index < 0 or chunk_index >= total_chunks:
+        return fail("文件分片序号不正确", code=400)
+
+    expected_size = min(
+        MULTIPART_CHUNK_BYTES,
+        int(metadata["fileSize"]) - chunk_index * MULTIPART_CHUNK_BYTES,
+    )
+    part_path = os.path.join(upload_dir, f"{chunk_index}.part")
+    temporary_path = f"{part_path}.uploading"
+    received = 0
+    try:
+        with open(temporary_path, "wb") as stream:
+            while chunk := await file.read(UPLOAD_CHUNK_BYTES):
+                received += len(chunk)
+                if received > expected_size:
+                    raise ValueError("文件分片大小不正确")
+                stream.write(chunk)
+        if received != expected_size:
+            raise ValueError("文件分片大小不正确")
+        os.replace(temporary_path, part_path)
+    except ValueError as exc:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        return fail(str(exc), code=400)
+    return ok({"chunkIndex": chunk_index, "uploadedBytes": received})
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/complete")
+def complete_multipart_upload(
+    avatar_id: str,
+    upload_id: str,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    part_paths = [os.path.join(upload_dir, f"{index}.part") for index in range(total_chunks)]
+    if not all(os.path.isfile(path) for path in part_paths):
+        return fail("文件分片尚未上传完整", code=400)
+    if sum(os.path.getsize(path) for path in part_paths) != int(metadata["fileSize"]):
+        return fail("文件分片总大小不正确", code=400)
+
+    avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
+    os.makedirs(avatar_dir, exist_ok=True)
+    stored = f"{uuid.uuid4().hex}{metadata['extension']}"
+    final_path = os.path.join(avatar_dir, stored)
+    temporary_path = f"{final_path}.assembling"
+    try:
+        with open(temporary_path, "wb") as output:
+            for part_path in part_paths:
+                with open(part_path, "rb") as source:
+                    shutil.copyfileobj(source, output, UPLOAD_CHUNK_BYTES)
+        os.replace(temporary_path, final_path)
+        doc = _create_knowledge_doc(
+            db,
+            avatar_id,
+            metadata["filename"],
+            metadata["extension"],
+            int(metadata["fileSize"]),
+            stored,
+        )
+    except Exception:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        raise
+    shutil.rmtree(upload_dir, ignore_errors=True)
+    return ok(_doc_payload(doc))
+
+
+@router.post("/avatar/{avatar_id}/knowledge/docs/{doc_id}/retry")
+def retry_doc(avatar_id: str, doc_id: str, authorization: str = Header(None), db: Session = Depends(get_db)):
+    _require_owned_avatar(db, avatar_id, authorization)
+    doc = db.query(KnowledgeDoc).filter(
+        KnowledgeDoc.id == doc_id, KnowledgeDoc.avatar_id == avatar_id
+    ).first()
+    if not doc:
+        return fail("文档不存在", code=404)
+    if doc.vectorized and doc.status == "ready":
+        return ok(_doc_payload(doc))
+    stored_name = os.path.basename(doc.file_url or "")
+    if not stored_name or not os.path.isfile(os.path.join(UPLOAD_DIR, avatar_id, stored_name)):
+        return fail("原文件不可用，请重新上传", code=400)
+    db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == doc.id).delete()
+    doc.status = "parsing"
+    doc.vectorized = False
+    doc.embedding_model = ""
+    doc.chunk_count = 0
+    doc.vectorized_at = None
+    doc.error_message = ""
+    doc.index_stage = "queued"
+    doc.index_progress = 0
+    db.commit()
+    db.refresh(doc)
+    knowledge_vectorizer.enqueue(doc.id)
     return ok(_doc_payload(doc))
 
 

digital-avatar-app/backend/routers/takeover.py

@@ -8,13 +8,25 @@ from sqlalchemy.orm import Session
 from database import get_db
 from models import TakeoverCursor, TakeoverReplyTask, User
 from responses import fail, ok
-from routers.authorizations import _require_authorization
+from routers.authorizations import (
+    DEFAULT_TAKEOVER_DELAY_SECONDS,
+    MAX_TAKEOVER_DELAY_SECONDS,
+    MIN_TAKEOVER_DELAY_SECONDS,
+    _require_authorization,
+    _stored_takeover_delay,
+)
 from routers.avatars import _require_owned_avatar
 
 router = APIRouter(tags=["分身接管"])
 BOXIM_STATUS_FRESH_SECONDS = 60
 
 
+def _delay_label(seconds: int) -> str:
+    if seconds % 60 == 0:
+        return f"{seconds // 60} 分钟"
+    return f"{seconds} 秒"
+
+
 @router.get("/avatar/{avatar_id}/takeover/status")
 def get_takeover_status(
     avatar_id: str,
@@ -24,6 +36,7 @@ def get_takeover_status(
     avatar = _require_owned_avatar(db, avatar_id, authorization)
     permissions = (avatar.config or {}).get("authorizationPermissions", [])
     enabled = isinstance(permissions, list) and "takeover" in permissions
+    reply_delay_seconds = _stored_takeover_delay(avatar)
     user = db.query(User).filter(User.huihui_user_id == avatar.owner_id).first()
     cursor = db.query(TakeoverCursor).filter(TakeoverCursor.avatar_id == avatar.id).first()
     pending_count = (
@@ -49,7 +62,10 @@ def get_takeover_status(
         and cursor.last_polled_at
         >= datetime.utcnow() - timedelta(seconds=BOXIM_STATUS_FRESH_SECONDS)
     ):
-        status, message = "ready", "BOXIM 已连接，收到私聊消息 3 秒后自动回复"
+        status, message = (
+            "ready",
+            f"BOXIM 已连接，收到私聊消息 {_delay_label(reply_delay_seconds)}后自动回复",
+        )
     else:
         status, message = "connecting", "正在连接 BOXIM"
 
@@ -59,6 +75,7 @@ def get_takeover_status(
             "status": status,
             "message": message,
             "pendingCount": pending_count,
+            "takeoverReplyDelaySeconds": reply_delay_seconds,
             "lastPolledAt": cursor.last_polled_at.isoformat() if cursor and cursor.last_polled_at else None,
         }
     )
@@ -91,7 +108,7 @@ def update_takeover_config(
     auth = _require_authorization(db, avatar_id, str(auth_id))
     enabled = bool(auth.takeover_enabled)
     mode = auth.takeover_mode or "immediate"
-    delay = auth.takeover_delay_seconds or 30
+    delay = auth.takeover_delay_seconds or DEFAULT_TAKEOVER_DELAY_SECONDS
 
     if _has(payload, "takeoverEnabled", "takeover_enabled"):
         raw_enabled = _read(payload, "takeoverEnabled", "takeover_enabled")
@@ -106,8 +123,12 @@ def update_takeover_config(
 
     if _has(payload, "takeoverDelaySeconds", "takeover_delay_seconds"):
         delay = _read(payload, "takeoverDelaySeconds", "takeover_delay_seconds")
-        if isinstance(delay, bool) or not isinstance(delay, int) or not 5 <= delay <= 3600:
-            return fail("延迟时间需在 5 到 3600 秒之间", 400)
+        if (
+            isinstance(delay, bool)
+            or not isinstance(delay, int)
+            or not MIN_TAKEOVER_DELAY_SECONDS <= delay <= MAX_TAKEOVER_DELAY_SECONDS
+        ):
+            return fail("延迟时间需在 3 秒到 24 小时之间", 400)
 
     if enabled and auth.target_type != "user":
         return fail("本期仅支持对会会用户开启单聊接管", 400)

digital-avatar-app/backend/routers/tokens.py

@@ -1,37 +1,346 @@
-from fastapi import APIRouter, Depends, Body
+import hashlib
+import hmac
+import json
+import os
+import uuid
+from datetime import datetime
+from decimal import Decimal, InvalidOperation, ROUND_HALF_UP
+from urllib.parse import parse_qs
+
+from fastapi import APIRouter, Body, Depends, Header, HTTPException, Request
+from sqlalchemy import func
 from sqlalchemy.orm import Session
 
 from database import get_db
-from models import TokenAccount, TokenPlan
-from responses import ok, fail
+from models import TokenAccount, TokenPaymentOrder, TokenPlan, TokenUsage, User
+from responses import fail, ok
+from services.huihui_payment import HuihuiPaymentClient, HuihuiPaymentError
+from services.token_billing import DEFAULT_TOKEN_GRANT, get_or_create_account
 
 router = APIRouter(tags=["Token"])
 
+PAYMENT_METHODS = {"wechat": "WECHAT", "alipay": "ALIPAY"}
+PAYMENT_SCENES = {"APP", "LITE", "JSAPI"}
+SUCCESS_STATUSES = {"SUCCESS", "SUCCEEDED", "PAID", "COMPLETED", "TRADE_SUCCESS"}
+FAILED_STATUSES = {"FAIL", "FAILED", "CLOSED", "CANCELLED", "CANCELED", "EXPIRED"}
+
+
+def _require_user(authorization: str | None, db: Session) -> User:
+    if not authorization:
+        raise HTTPException(status_code=401, detail="未登录")
+    token = authorization.replace("Bearer ", "", 1).replace("bearer ", "", 1).strip()
+    user = db.query(User).filter(User.app_token == token).first()
+    if not user:
+        raise HTTPException(status_code=401, detail="会话无效或已过期")
+    return user
+
+
+def _payment_client() -> HuihuiPaymentClient:
+    return HuihuiPaymentClient({
+        "HUIHUI_PAYMENT_BASE_URL": os.getenv(
+            "HUIHUI_PAYMENT_BASE_URL", "https://open.99hui.com/api/payment-v3"
+        ),
+        "HUIHUI_APP_ID": os.getenv("HUIHUI_APP_ID", ""),
+        "HUIHUI_ACCESS_ID": os.getenv("HUIHUI_ACCESS_ID", ""),
+        "HUIHUI_ACCESS_SECRET": os.getenv("HUIHUI_ACCESS_SECRET", ""),
+        "HUIHUI_PAYMENT_TIMEOUT_SECONDS": os.getenv("HUIHUI_PAYMENT_TIMEOUT_SECONDS", "30"),
+    })
+
+
+def _callback_url(order_no: str) -> str:
+    base = os.getenv(
+        "HUIHUI_PAYMENT_CALLBACK_BASE_URL", "https://digital.99hui.com"
+    ).rstrip("/")
+    secret = os.getenv("HUIHUI_PAYMENT_CALLBACK_SECRET", "").strip()
+    if len(secret) < 16:
+        raise HuihuiPaymentError("会会支付回调密钥未配置")
+    signature = hmac.new(secret.encode(), order_no.encode(), hashlib.sha256).hexdigest()
+    return f"{base}/api/token/payment/callback/{order_no}/{signature}"
+
+
+def _price_cents(price: float) -> int:
+    return int(
+        (Decimal(str(price)) * Decimal("100")).quantize(
+            Decimal("1"), rounding=ROUND_HALF_UP
+        )
+    )
+
+
+def _payment_payload(order: TokenPaymentOrder, account: TokenAccount) -> dict:
+    return {**order.to_dict(), "balance": account.balance}
+
+
+def _nested_payload(value):
+    if isinstance(value, str):
+        text = value.strip()
+        if text[:1] in ("{", "["):
+            try:
+                return _nested_payload(json.loads(text))
+            except (TypeError, ValueError):
+                return value
+        return value
+    if isinstance(value, list):
+        return [_nested_payload(item) for item in value]
+    if isinstance(value, dict):
+        return {key: _nested_payload(item) for key, item in value.items()}
+    return value
+
+
+def _find_value(payload, *names):
+    expected = {name.lower() for name in names}
+    if isinstance(payload, dict):
+        for key, value in payload.items():
+            if key.lower() in expected and value not in (None, ""):
+                return value
+        for value in payload.values():
+            found = _find_value(value, *names)
+            if found not in (None, ""):
+                return found
+    elif isinstance(payload, list):
+        for value in payload:
+            found = _find_value(value, *names)
+            if found not in (None, ""):
+                return found
+    return None
+
+
+def _callback_amount_cents(payload) -> int | None:
+    value = _find_value(
+        payload,
+        "actualAmt",
+        "payAmt",
+        "masterOrderAmt",
+        "orderAmt",
+        "amount",
+        "totalAmount",
+    )
+    if value in (None, ""):
+        return None
+    try:
+        return int(
+            (Decimal(str(value)) * Decimal("100")).quantize(
+                Decimal("1"), rounding=ROUND_HALF_UP
+            )
+        )
+    except (InvalidOperation, TypeError, ValueError):
+        return None
+
 
 @router.get("/token/balance")
-def balance(db: Session = Depends(get_db)):
-    acc = db.query(TokenAccount).first()
-    return ok({"balance": acc.balance if acc else 0})
+def balance(authorization: str = Header(None), db: Session = Depends(get_db)):
+    user = _require_user(authorization, db)
+    acc = get_or_create_account(db, user.id)
+    return ok({
+        "balance": acc.balance,
+        "totalGranted": acc.total_granted,
+        "totalConsumed": acc.total_consumed,
+    })
 
 
 @router.get("/token/plans")
-def plans(db: Session = Depends(get_db)):
+def plans(authorization: str = Header(None), db: Session = Depends(get_db)):
+    _require_user(authorization, db)
     items = db.query(TokenPlan).order_by(TokenPlan.price.asc()).all()
     return ok([p.to_dict() for p in items])
 
 
+# 积分只会在会会支付回调确认成功后到账。
 @router.post("/token/charge")
-def charge(payload: dict = Body(...), db: Session = Depends(get_db)):
-    plan_id = payload.get("planId")
-    plan = db.query(TokenPlan).filter(TokenPlan.id == plan_id).first()
+def charge(payload: dict = Body(...), authorization: str = Header(None), db: Session = Depends(get_db)):
+    user = _require_user(authorization, db)
+    plan = db.query(TokenPlan).filter(TokenPlan.id == payload.get("planId")).first()
     if not plan:
         return fail("套餐不存在", 404)
-    acc = db.query(TokenAccount).first()
-    if not acc:
-        acc = TokenAccount(balance=0)
-        db.add(acc)
+
+    payment_method = str(payload.get("paymentMethod") or "").lower()
+    pay_type = PAYMENT_METHODS.get(payment_method)
+    if not pay_type:
+        return fail("请选择正确的支付方式", 400)
+    pay_way = str(payload.get("payScene") or "APP").upper()
+    if pay_way not in PAYMENT_SCENES:
+        return fail("当前支付场景不受支持", 400)
+
+    cents = _price_cents(plan.price)
+    order = TokenPaymentOrder(
+        order_no=f"AV{datetime.utcnow().strftime('%Y%m%d%H%M%S')}{uuid.uuid4().hex[:12].upper()}",
+        user_id=user.id,
+        plan_id=plan.id,
+        payment_method=payment_method,
+        pay_type=pay_type,
+        pay_way=pay_way,
+        points_amount=plan.amount,
+        price_cents=cents,
+        status="pending",
+    )
+    db.add(order)
     db.commit()
-        db.refresh(acc)
-    acc.balance += plan.amount
+
+    try:
+        callback_url = _callback_url(order.order_no)
+    except HuihuiPaymentError as exc:
+        order.status = "failed"
+        order.failure_reason = str(exc)
         db.commit()
-    return ok({"balance": acc.balance, "charged": plan.amount})
+        return fail(str(exc), 503)
+
+    try:
+        result = _payment_client().create_payment(
+            huihui_token=user.huihui_token,
+            huihui_user_id=user.huihui_user_id,
+            real_name=user.nickname,
+            order_no=order.order_no,
+            amount=f"{cents / 100:.2f}",
+            points_amount=plan.amount,
+            pay_type=pay_type,
+            pay_way=pay_way,
+            callback_url=callback_url,
+        )
+    except HuihuiPaymentError as exc:
+        order.status = "failed"
+        order.failure_reason = str(exc)[:500]
+        db.commit()
+        return fail(str(exc), 502)
+
+    db.refresh(order)
+    if order.status != "paid":
+        order.provider_order_id = str(result.get("orderId") or "")
+        order.provider_order_no = str(result.get("orderNo") or "")
+        order.provider_status = str(result.get("status") or "pending")
+        message = result.get("payMessage") or ""
+        order.pay_message = (
+            json.dumps(message, ensure_ascii=False)
+            if isinstance(message, (dict, list))
+            else str(message)
+        )
+        if order.provider_status.upper() in FAILED_STATUSES:
+            order.status = "failed"
+            order.failure_reason = str(result.get("bankReturnMsg") or "支付下单失败")[:500]
+        db.commit()
+
+    return ok(_payment_payload(order, get_or_create_account(db, user.id)))
+
+
+@router.get("/token/payment/{order_id}")
+def payment_status(order_id: str, authorization: str = Header(None), db: Session = Depends(get_db)):
+    user = _require_user(authorization, db)
+    order = db.query(TokenPaymentOrder).filter(
+        TokenPaymentOrder.id == order_id,
+        TokenPaymentOrder.user_id == user.id,
+    ).first()
+    if not order:
+        return fail("支付订单不存在", 404)
+    return ok(_payment_payload(order, get_or_create_account(db, user.id)))
+
+
+@router.post("/token/payment/callback/{order_no}/{callback_signature}")
+async def payment_callback(
+    order_no: str,
+    callback_signature: str,
+    request: Request,
+    db: Session = Depends(get_db),
+):
+    secret = os.getenv("HUIHUI_PAYMENT_CALLBACK_SECRET", "").strip()
+    expected = hmac.new(secret.encode(), order_no.encode(), hashlib.sha256).hexdigest()
+    if len(secret) < 16 or not hmac.compare_digest(callback_signature, expected):
+        raise HTTPException(status_code=404, detail="Not found")
+
+    content_type = request.headers.get("content-type", "").lower()
+    if "application/json" in content_type:
+        try:
+            payload = await request.json()
+        except ValueError:
+            return fail("支付回调格式不正确", 400)
+    else:
+        raw = (await request.body()).decode("utf-8", errors="replace")
+        payload = {key: values[-1] for key, values in parse_qs(raw).items()}
+    payload = _nested_payload(payload)
+
+    payload_order_no = str(_find_value(
+        payload,
+        "masterOrderNo",
+        "master_order_no",
+        "orderNo",
+        "order_no",
+        "bizOrderNo",
+    ) or "").strip()
+    if payload_order_no and payload_order_no != order_no:
+        return fail("支付回调订单号不匹配", 422)
+
+    order = db.query(TokenPaymentOrder).filter(TokenPaymentOrder.order_no == order_no).first()
+    if not order:
+        return fail("支付订单不存在", 404)
+    if order.status == "paid":
+        return ok({"received": True, "duplicate": True})
+
+    provider_status = str(_find_value(
+        payload, "status", "payStatus", "tradeStatus", "paymentStatus"
+    ) or "").upper()
+    order.provider_status = provider_status
+    if provider_status not in SUCCESS_STATUSES:
+        if provider_status in FAILED_STATUSES:
+            order.status = "failed"
+            order.failure_reason = str(
+                _find_value(payload, "message", "errorMsg", "failReason") or "支付失败"
+            )[:500]
+        db.commit()
+        return ok({"received": True, "paid": False})
+
+    paid_cents = _callback_amount_cents(payload)
+    if paid_cents is None or paid_cents != order.price_cents:
+        order.failure_reason = "支付回调金额不匹配"
+        db.commit()
+        return fail("支付金额不匹配", 422)
+
+    updated = db.query(TokenPaymentOrder).filter(
+        TokenPaymentOrder.id == order.id,
+        TokenPaymentOrder.status != "paid",
+    ).update({
+        TokenPaymentOrder.status: "paid",
+        TokenPaymentOrder.provider_status: provider_status,
+        TokenPaymentOrder.paid_at: datetime.utcnow(),
+        TokenPaymentOrder.failure_reason: "",
+    }, synchronize_session=False)
+    if updated:
+        account = db.query(TokenAccount).filter(TokenAccount.user_id == order.user_id).first()
+        if account is None:
+            account = TokenAccount(
+                user_id=order.user_id,
+                balance=DEFAULT_TOKEN_GRANT,
+                total_granted=DEFAULT_TOKEN_GRANT,
+                total_consumed=0,
+            )
+            db.add(account)
+            db.flush()
+        account.balance = int(account.balance or 0) + order.points_amount
+        account.total_granted = int(account.total_granted or 0) + order.points_amount
+    db.commit()
+    return ok({"received": True, "paid": True})
+
+
+@router.get("/token/usage")
+def usage(authorization: str = Header(None), db: Session = Depends(get_db)):
+    user = _require_user(authorization, db)
+    rows = (
+        db.query(
+            TokenUsage.avatar_id,
+            TokenUsage.source,
+            func.sum(TokenUsage.prompt_tokens),
+            func.sum(TokenUsage.completion_tokens),
+            func.sum(TokenUsage.total_tokens),
+            func.count(TokenUsage.id),
+        )
+        .filter(TokenUsage.user_id == user.id, TokenUsage.status == "completed")
+        .group_by(TokenUsage.avatar_id, TokenUsage.source)
+        .all()
+    )
+    return ok([
+        {
+            "avatarId": avatar_id,
+            "source": source,
+            "promptTokens": int(prompt_tokens or 0),
+            "completionTokens": int(completion_tokens or 0),
+            "totalTokens": int(total_tokens or 0),
+            "requestCount": int(request_count or 0),
+        }
+        for avatar_id, source, prompt_tokens, completion_tokens, total_tokens, request_count in rows
+    ])

digital-avatar-app/backend/services/boxim_image_service.py

@@ -0,0 +1,151 @@
+"""Parse and safely download image payloads from BOXIM private messages."""
+
+import ipaddress
+import json
+import os
+import socket
+from dataclasses import dataclass
+from pathlib import PurePosixPath
+from urllib.parse import unquote, urljoin, urlsplit
+
+import httpx
+
+
+MAX_REDIRECTS = 3
+
+
+class BoxIMImageError(RuntimeError):
+    pass
+
+
+@dataclass(frozen=True)
+class DownloadedBoxIMImage:
+    content: bytes
+    filename: str
+    mime_type: str
+    source_url: str
+
+
+def parse_boxim_image_url(content: str, *, base_url: str = "") -> str:
+    try:
+        payload = json.loads(content or "")
+    except (TypeError, ValueError) as exc:
+        raise BoxIMImageError("BOXIM 图片消息格式无效") from exc
+    if not isinstance(payload, dict):
+        raise BoxIMImageError("BOXIM 图片消息格式无效")
+
+    value = payload.get("originUrl") or payload.get("thumbUrl") or payload.get("url")
+    if not isinstance(value, str) or not value.strip():
+        raise BoxIMImageError("BOXIM 图片消息缺少图片地址")
+    value = value.strip()
+    if value.startswith("/"):
+        if not base_url:
+            raise BoxIMImageError("BOXIM 图片地址不完整")
+        value = urljoin(f"{base_url.rstrip('/')}/", value)
+    return value
+
+
+def _configured_hosts(name: str) -> set[str]:
+    return {
+        value.strip().lower().rstrip(".")
+        for value in os.getenv(name, "").split(",")
+        if value.strip()
+    }
+
+
+def _host_matches(host: str, configured: set[str]) -> bool:
+    return any(host == value or host.endswith(f".{value}") for value in configured)
+
+
+def _resolved_addresses(host: str, port: int) -> set[ipaddress.IPv4Address | ipaddress.IPv6Address]:
+    try:
+        return {
+            ipaddress.ip_address(item[4][0])
+            for item in socket.getaddrinfo(host, port, type=socket.SOCK_STREAM)
+        }
+    except (OSError, ValueError) as exc:
+        raise BoxIMImageError("BOXIM 图片地址无法解析") from exc
+
+
+def _is_safe_remote_url(url: str) -> None:
+    parsed = urlsplit(url)
+    scheme = parsed.scheme.lower()
+    allow_http = os.getenv("BOXIM_IMAGE_ALLOW_HTTP", "").lower() in {"1", "true", "yes"}
+    if scheme not in ({"https", "http"} if allow_http else {"https"}):
+        raise BoxIMImageError("BOXIM 图片地址必须使用 HTTPS")
+    if parsed.username or parsed.password or not parsed.hostname:
+        raise BoxIMImageError("BOXIM 图片地址无效")
+
+    host = parsed.hostname.lower().rstrip(".")
+    allowed_hosts = _configured_hosts("BOXIM_IMAGE_ALLOWED_HOSTS")
+    if allowed_hosts and not _host_matches(host, allowed_hosts):
+        raise BoxIMImageError("BOXIM 图片地址不在允许的域名范围内")
+
+    private_hosts = _configured_hosts("BOXIM_IMAGE_PRIVATE_HOSTS")
+    try:
+        addresses = {ipaddress.ip_address(host)}
+    except ValueError:
+        addresses = _resolved_addresses(host, parsed.port or (443 if scheme == "https" else 80))
+    if not addresses:
+        raise BoxIMImageError("BOXIM 图片地址无法解析")
+    if _host_matches(host, private_hosts):
+        return
+    if any(not address.is_global for address in addresses):
+        raise BoxIMImageError("BOXIM 图片地址指向受限网络")
+
+
+def _filename_from_url(url: str) -> str:
+    value = unquote(PurePosixPath(urlsplit(url).path).name).strip()
+    value = value.replace("\x00", "")
+    return (value or "boxim-image")[:255]
+
+
+def download_boxim_image(
+    content: str,
+    *,
+    base_url: str = "",
+    transport: httpx.BaseTransport | None = None,
+) -> DownloadedBoxIMImage:
+    """Download one BOXIM image without redirects or oversized responses escaping checks."""
+    url = parse_boxim_image_url(content, base_url=base_url)
+    max_bytes = max(1024, int(os.getenv("CHAT_IMAGE_MAX_BYTES", str(8 * 1024 * 1024))))
+    timeout = max(1.0, min(float(os.getenv("BOXIM_IMAGE_TIMEOUT_SECONDS", "15")), 60.0))
+
+    with httpx.Client(
+        timeout=timeout,
+        follow_redirects=False,
+        trust_env=False,
+        transport=transport,
+    ) as client:
+        for _ in range(MAX_REDIRECTS + 1):
+            _is_safe_remote_url(url)
+            try:
+                with client.stream("GET", url, headers={"Accept": "image/*"}) as response:
+                    if response.status_code in {301, 302, 303, 307, 308}:
+                        location = response.headers.get("location", "").strip()
+                        if not location:
+                            raise BoxIMImageError("BOXIM 图片跳转地址无效")
+                        url = urljoin(url, location)
+                        continue
+                    response.raise_for_status()
+                    raw_length = response.headers.get("content-length", "")
+                    if raw_length.isdigit() and int(raw_length) > max_bytes:
+                        raise BoxIMImageError("BOXIM 图片超过大小限制")
+                    chunks = bytearray()
+                    for chunk in response.iter_bytes():
+                        chunks.extend(chunk)
+                        if len(chunks) > max_bytes:
+                            raise BoxIMImageError("BOXIM 图片超过大小限制")
+                    if not chunks:
+                        raise BoxIMImageError("BOXIM 图片内容为空")
+                    return DownloadedBoxIMImage(
+                        content=bytes(chunks),
+                        filename=_filename_from_url(url),
+                        mime_type=response.headers.get("content-type", "").split(";", 1)[0][:100],
+                        source_url=url,
+                    )
+            except BoxIMImageError:
+                raise
+            except (httpx.HTTPError, OSError) as exc:
+                raise BoxIMImageError("BOXIM 图片下载失败") from exc
+    raise BoxIMImageError("BOXIM 图片跳转次数过多")

digital-avatar-app/backend/services/chat_attachment_service.py

@@ -0,0 +1,20 @@
+from datetime import datetime
+
+from sqlalchemy.orm import Session
+
+from models import ChatAttachment
+
+
+def purge_expired_chat_attachments(
+    db: Session,
+    *,
+    now: datetime | None = None,
+) -> int:
+    """Remove expired derived image data; raw image bytes are never persisted."""
+    count = db.query(ChatAttachment).filter(
+        ChatAttachment.expires_at < (now or datetime.utcnow())
+    ).delete(synchronize_session=False)
+    if count:
+        db.commit()
+        db.expire_all()
+    return count

digital-avatar-app/backend/services/chat_model_config.py

@@ -0,0 +1,115 @@
+import logging
+import os
+import threading
+import time
+from dataclasses import dataclass
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass(frozen=True)
+class ChatModelConfig:
+    api_base_url: str
+    api_key: str
+    model: str
+    max_tokens: int
+    timeout_seconds: float
+    vision_model: str
+    ocr_model: str
+    vision_max_tokens: int
+    vision_timeout_seconds: float
+    source: str
+
+
+_cache_lock = threading.Lock()
+_cached_config: ChatModelConfig | None = None
+_cache_expires_at = 0.0
+
+
+def _environment_config() -> ChatModelConfig:
+    return ChatModelConfig(
+        api_base_url=os.getenv(
+            "CHAT_API_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1"
+        ).rstrip("/"),
+        api_key=os.getenv("CHAT_API_KEY", ""),
+        model=os.getenv("CHAT_MODEL", "qwen-plus"),
+        max_tokens=max(128, int(os.getenv("CHAT_MAX_OUTPUT_TOKENS", "1024"))),
+        timeout_seconds=max(5.0, float(os.getenv("CHAT_TIMEOUT_SECONDS", "30"))),
+        vision_model=os.getenv("VISION_MODEL", "qwen3.6-flash"),
+        ocr_model=os.getenv("VISION_OCR_MODEL", "qwen-vl-ocr"),
+        vision_max_tokens=max(256, int(os.getenv("VISION_MAX_OUTPUT_TOKENS", "2048"))),
+        vision_timeout_seconds=max(10.0, float(os.getenv("VISION_TIMEOUT_SECONDS", "90"))),
+        source="environment",
+    )
+
+
+def _fetch_runtime_config() -> ChatModelConfig | None:
+    url = os.getenv("CHAT_MODEL_CONFIG_URL", "").strip()
+    token = os.getenv("AVATAR_MODEL_CONFIG_TOKEN", "").strip()
+    if not url or not token:
+        return None
+    response = httpx.get(
+        url,
+        headers={"X-Avatar-Config-Token": token},
+        timeout=max(2.0, float(os.getenv("CHAT_MODEL_CONFIG_TIMEOUT_SECONDS", "5"))),
+    )
+    response.raise_for_status()
+    payload = response.json().get("data") or {}
+    api_base_url = str(payload.get("api_base_url") or "").rstrip("/")
+    api_key = str(payload.get("api_key") or "")
+    model = str(payload.get("model") or "")
+    if not api_base_url or not api_key or not model:
+        raise ValueError("数字分身专用模型配置不完整")
+    return ChatModelConfig(
+        api_base_url=api_base_url,
+        api_key=api_key,
+        model=model,
+        max_tokens=max(128, int(payload.get("max_tokens") or 1024)),
+        timeout_seconds=max(5.0, float(payload.get("timeout_seconds") or 30)),
+        vision_model=str(
+            payload.get("vision_model")
+            or os.getenv("VISION_MODEL", "qwen3.6-flash")
+        ),
+        ocr_model=str(
+            payload.get("ocr_model")
+            or os.getenv("VISION_OCR_MODEL", "qwen-vl-ocr")
+        ),
+        vision_max_tokens=max(
+            256, int(os.getenv("VISION_MAX_OUTPUT_TOKENS", "2048"))
+        ),
+        vision_timeout_seconds=max(
+            10.0, float(os.getenv("VISION_TIMEOUT_SECONDS", "90"))
+        ),
+        source="admin",
+    )
+
+
+def get_chat_model_config(*, force_refresh: bool = False) -> ChatModelConfig:
+    global _cached_config, _cache_expires_at
+
+    now = time.monotonic()
+    if not force_refresh and _cached_config is not None and now < _cache_expires_at:
+        return _cached_config
+
+    with _cache_lock:
+        now = time.monotonic()
+        if not force_refresh and _cached_config is not None and now < _cache_expires_at:
+            return _cached_config
+        try:
+            config = _fetch_runtime_config() or _environment_config()
+        except (httpx.HTTPError, ValueError, TypeError) as exc:
+            logger.warning("读取数字分身专用模型配置失败，暂时使用环境变量配置: %s", exc)
+            config = _environment_config()
+        _cached_config = config
+        ttl = max(5, int(os.getenv("CHAT_MODEL_CONFIG_CACHE_SECONDS", "60")))
+        _cache_expires_at = now + ttl
+        return config
+
+
+def clear_chat_model_config_cache() -> None:
+    global _cached_config, _cache_expires_at
+    with _cache_lock:
+        _cached_config = None
+        _cache_expires_at = 0.0

digital-avatar-app/backend/services/huihui_payment.py

@@ -0,0 +1,124 @@
+"""Signed client for Huihui's production payment-v3 service."""
+
+import hashlib
+import random
+import string
+from datetime import datetime, timedelta, timezone
+from typing import Any
+
+import httpx
+
+
+_CN_TZ = timezone(timedelta(hours=8))
+
+
+class HuihuiPaymentError(RuntimeError):
+    pass
+
+
+class HuihuiPaymentClient:
+    def __init__(self, config: dict):
+        self.base_url = config.get(
+            "HUIHUI_PAYMENT_BASE_URL", "https://open.99hui.com/api/payment-v3"
+        ).rstrip("/")
+        self.app_id = config.get("HUIHUI_APP_ID", "")
+        self.access_id = config.get("HUIHUI_ACCESS_ID", "")
+        self.access_secret = config.get("HUIHUI_ACCESS_SECRET", "")
+        self.timeout = float(config.get("HUIHUI_PAYMENT_TIMEOUT_SECONDS", 30))
+
+    @property
+    def configured(self) -> bool:
+        return bool(self.base_url and self.app_id and self.access_id and self.access_secret)
+
+    def _signed_params(self, user_id: str) -> dict:
+        params = {
+            "appId": self.app_id,
+            "accessId": self.access_id,
+            "nonce": "".join(random.choices(string.ascii_lowercase + string.digits, k=12)),
+            "timestamp": datetime.now(_CN_TZ).strftime("%Y%m%d%H%M%S"),
+            "signType": "MD5",
+            "signVersion": "1.0",
+            "userId": user_id,
+        }
+        source = "&".join(
+            f"{key}={params[key]}"
+            for key in sorted(params)
+            if params[key] not in (None, "", [])
+        )
+        source += f"&accessSecret={self.access_secret}"
+        params["signature"] = hashlib.md5(source.encode("utf-8")).hexdigest().upper()
+        return params
+
+    @staticmethod
+    def _json(response: httpx.Response) -> dict:
+        try:
+            payload = response.json()
+        except ValueError as exc:
+            raise HuihuiPaymentError("会会支付返回了无效响应") from exc
+        if not isinstance(payload, dict):
+            raise HuihuiPaymentError("会会支付返回格式不正确")
+        return payload
+
+    def create_payment(
+        self,
+        *,
+        huihui_token: str,
+        huihui_user_id: str,
+        real_name: str,
+        order_no: str,
+        amount: str,
+        points_amount: int,
+        pay_type: str,
+        pay_way: str,
+        callback_url: str,
+    ) -> dict[str, Any]:
+        if not self.configured:
+            raise HuihuiPaymentError("会会支付服务未配置")
+        if not huihui_token or not huihui_user_id:
+            raise HuihuiPaymentError("当前会会登录凭证无法发起支付")
+
+        now = datetime.now(_CN_TZ)
+        body = {
+            "appId": self.app_id,
+            "callbackUrl": callback_url,
+            "chargeType": 4,
+            "currency": "cny",
+            "description": f"充值 {points_amount} 积分",
+            "expend": {},
+            "masterOrderAmt": amount,
+            "masterOrderNo": order_no,
+            "memberId": huihui_user_id,
+            "orderDesc": "数字分身积分充值",
+            "orderTime": now.isoformat(),
+            "orderTitle": "数字分身积分充值",
+            "payAmt": float(amount),
+            "payType": pay_type,
+            "payWay": pay_way,
+            "realName": real_name or "会会用户",
+            "timeExpire": (now + timedelta(hours=2)).strftime("%Y%m%d%H%M%S"),
+        }
+        headers = {
+            "Authorization": f"Bearer {huihui_token}",
+            "appId": self.app_id,
+            "windowAppId": self.app_id,
+        }
+        try:
+            response = httpx.post(
+                f"{self.base_url}/payment/pay",
+                headers=headers,
+                params=self._signed_params(huihui_user_id),
+                json=body,
+                timeout=self.timeout,
+                follow_redirects=True,
+            )
+        except httpx.HTTPError as exc:
+            raise HuihuiPaymentError("会会支付连接失败，请稍后重试") from exc
+
+        payload = self._json(response)
+        code = payload.get("code")
+        if response.status_code >= 400 or code not in (0, 200, "0", "200"):
+            raise HuihuiPaymentError(payload.get("message") or "会会支付下单失败")
+        data = payload.get("data") or {}
+        if not isinstance(data, dict):
+            raise HuihuiPaymentError("会会支付未返回订单信息")
+        return data

digital-avatar-app/backend/services/knowledge_vectorizer.py

@@ -0,0 +1,143 @@
+"""Durable, serial knowledge-document indexing for the avatar knowledge base."""
+
+import json
+import logging
+import os
+import queue
+import threading
+from datetime import datetime, timezone
+
+from database import SessionLocal
+from models import KnowledgeChunk, KnowledgeDoc
+import embeddings
+
+logger = logging.getLogger(__name__)
+
+BACKEND_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+UPLOAD_DIR = os.path.abspath(
+    os.getenv("UPLOAD_DIR", os.path.join(BACKEND_DIR, "routers", "uploads"))
+)
+
+
+class KnowledgeVectorizer:
+    """Indexes one document at a time so slow providers cannot block uploads."""
+
+    def __init__(self):
+        self._queue: queue.Queue[str] = queue.Queue()
+        self._queued: set[str] = set()
+        self._lock = threading.Lock()
+        self._thread: threading.Thread | None = None
+
+    def start(self):
+        if self._thread and self._thread.is_alive():
+            return
+        self._thread = threading.Thread(
+            target=self._run, name="knowledge-vectorizer", daemon=True
+        )
+        self._thread.start()
+        db = SessionLocal()
+        try:
+            # A process restart must not abandon documents already accepted by upload.
+            for (doc_id,) in db.query(KnowledgeDoc.id).filter(KnowledgeDoc.status == "parsing"):
+                self.enqueue(doc_id)
+        finally:
+            db.close()
+
+    def enqueue(self, doc_id: str):
+        with self._lock:
+            if doc_id in self._queued:
+                return
+            self._queued.add(doc_id)
+        self._queue.put(doc_id)
+
+    def _run(self):
+        while True:
+            doc_id = self._queue.get()
+            try:
+                self.vectorize_document(doc_id)
+            except Exception:
+                logger.exception("Unexpected knowledge vectorizer failure for %s", doc_id)
+            finally:
+                with self._lock:
+                    self._queued.discard(doc_id)
+                self._queue.task_done()
+
+    def vectorize_document(self, doc_id: str):
+        db = SessionLocal()
+        try:
+            doc = db.get(KnowledgeDoc, doc_id)
+            if not doc or doc.status != "parsing":
+                return
+
+            stored_name = os.path.basename(doc.file_url or "")
+            path = os.path.join(UPLOAD_DIR, doc.avatar_id, stored_name)
+            if not stored_name or not os.path.isfile(path):
+                raise FileNotFoundError("原文件不可用，请重新上传")
+
+            self._set_progress(db, doc, "extracting", 8)
+            text = embeddings.extract_text(path, f".{doc.file_type}")
+            self._set_progress(db, doc, "chunking", 22)
+            chunks = embeddings.chunk_text(text)
+            if not chunks:
+                raise ValueError("文档没有可建立索引的文字内容")
+            self._set_progress(db, doc, "embedding", 30)
+
+            def embedding_progress(done: int, total: int):
+                percent = 30 + int((done / max(1, total)) * 65)
+                self._set_progress(db, doc, "embedding", min(percent, 95))
+
+            vectors = embeddings.embed(chunks, on_progress=embedding_progress)
+            if len(vectors) != len(chunks):
+                raise ValueError("向量服务返回数量与文档分段不一致")
+
+            # Commit the document and every chunk together. Chat only sees complete indexes.
+            db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == doc.id).delete()
+            db.add_all(
+                [
+                    KnowledgeChunk(
+                        doc_id=doc.id,
+                        avatar_id=doc.avatar_id,
+                        content=chunk,
+                        vector=json.dumps(vector),
+                        chunk_index=index,
+                        embedding_model=embeddings.MODEL,
+                    )
+                    for index, (chunk, vector) in enumerate(zip(chunks, vectors))
+                ]
+            )
+            doc.vectorized = True
+            doc.embedding_model = embeddings.MODEL
+            doc.chunk_count = len(chunks)
+            doc.vectorized_at = datetime.now(timezone.utc)
+            doc.status = "ready"
+            doc.error_message = ""
+            doc.index_stage = "ready"
+            doc.index_progress = 100
+            db.commit()
+            logger.info("Knowledge document %s indexed with %s chunks", doc.id, len(chunks))
+        except Exception as exc:
+            db.rollback()
+            failed_doc = db.get(KnowledgeDoc, doc_id)
+            if failed_doc:
+                db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == failed_doc.id).delete()
+                failed_doc.status = "failed"
+                failed_doc.vectorized = False
+                failed_doc.embedding_model = ""
+                failed_doc.chunk_count = 0
+                failed_doc.vectorized_at = None
+                failed_doc.error_message = str(exc)[:300] or "建立知识索引失败"
+                failed_doc.index_stage = "failed"
+                failed_doc.index_progress = 0
+                db.commit()
+            logger.exception("Knowledge vectorization failed for %s: %s", doc_id, exc)
+        finally:
+            db.close()
+
+    @staticmethod
+    def _set_progress(db, doc, stage: str, progress: int):
+        doc.index_stage = stage
+        doc.index_progress = progress
+        db.commit()
+
+
+knowledge_vectorizer = KnowledgeVectorizer()

digital-avatar-app/backend/services/takeover_service.py

@@ -3,6 +3,7 @@
 import asyncio
 import hashlib
 import logging
+import os
 import re
 import secrets
 import time
@@ -13,21 +14,50 @@ from sqlalchemy.orm import Session
 
 from models import (
     Avatar,
+    ChatAttachment,
     TakeoverCursor,
     TakeoverMessage,
     TakeoverReplyTask,
     User,
 )
 from services.boxim_client import BoxIMClient, BoxIMError
+from services.boxim_image_service import (
+    BoxIMImageError,
+    download_boxim_image,
+    parse_boxim_image_url,
+)
+from services.vision_service import ImageValidationError
 
 logger = logging.getLogger(__name__)
 
 ACTIVE_TASK_STATUSES = ("pending", "generating", "ready", "sending")
 GENERATABLE_TASK_STATUSES = ("pending",)
 MAX_PROMPT_LENGTH = 4000
-MAX_STALE_SECONDS = 120
+DEFAULT_MAX_MESSAGE_AGE_SECONDS = 600
+MAX_SEND_OVERDUE_SECONDS = 120
 STUCK_LOCK_SECONDS = 90
 TAKEOVER_PERMISSION = "takeover"
+TAKEOVER_DELAY_KEY = "takeoverReplyDelaySeconds"
+DEFAULT_REPLY_DELAY_SECONDS = 180
+MIN_REPLY_DELAY_SECONDS = 3
+MAX_REPLY_DELAY_SECONDS = 86_400
+HUMAN_PAUSE_SECONDS = 600
+RATE_LIMIT_WINDOW_SECONDS = 300
+RATE_LIMIT_MAX_REPLIES = 5
+AVATAR_LOCAL_ID_PREFIX = "880"
+BOXIM_TEXT_MESSAGE_TYPE = 0
+BOXIM_IMAGE_MESSAGE_TYPE = 1
+BOXIM_IMAGE_PROMPT = "请看看这张图片。"
+BOXIM_IMAGE_UNAVAILABLE_REPLY = "这张图片我暂时没看清，麻烦重新发送一张清晰的原图。"
+IMAGE_CONTEXT_LOOKBACK_SECONDS = 1800
+IMAGE_REFERENCE_LOOKBACK_SECONDS = 172_800
+MAX_RECENT_IMAGE_CONTEXTS = 3
+
+_IMAGE_REFERENCE_PATTERN = re.compile(
+    r"(?:图片|图像|照片|截图|这张图|刚才.{0,8}图|病例|病历|检查单|检验单|化验单|报告|影像|"
+    r"\b(?:image|photo|picture|screenshot|scan|report)\b)",
+    re.IGNORECASE,
+)
 
 
 def _utcnow() -> datetime:
@@ -70,24 +100,69 @@ def _plain_text_reply(value: str) -> str:
     return "\n".join(line for line in lines if line).strip()
 
 
+def _avatar_local_id(owner_id: str, trigger_message_id: str) -> str:
+    """Build a deterministic BOXIM idempotency key that also marks avatar traffic."""
+    digest = hashlib.sha256(f"{owner_id}:{trigger_message_id}".encode("utf-8")).digest()
+    suffix = int.from_bytes(digest[:8], "big") % (10**15)
+    return f"{AVATAR_LOCAL_ID_PREFIX}{suffix:015d}"
+
+
+def _is_avatar_local_id(value: str | None) -> bool:
+    local_id = str(value or "").strip()
+    return len(local_id) == 18 and local_id.isdigit() and local_id.startswith(AVATAR_LOCAL_ID_PREFIX)
+
+
+def _configured_reply_delay(avatar: Avatar, fallback: int | None = None) -> int:
+    raw = (avatar.config or {}).get(
+        TAKEOVER_DELAY_KEY,
+        fallback if fallback is not None else DEFAULT_REPLY_DELAY_SECONDS,
+    )
+    if isinstance(raw, bool):
+        return DEFAULT_REPLY_DELAY_SECONDS
+    try:
+        delay = int(raw)
+    except (TypeError, ValueError):
+        return DEFAULT_REPLY_DELAY_SECONDS
+    if not MIN_REPLY_DELAY_SECONDS <= delay <= MAX_REPLY_DELAY_SECONDS:
+        return DEFAULT_REPLY_DELAY_SECONDS
+    return delay
+
+
+def _event_prompt(event: TakeoverMessage) -> str:
+    if event.message_type == BOXIM_TEXT_MESSAGE_TYPE:
+        return event.content.strip()
+    if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
+        return BOXIM_IMAGE_PROMPT
+    return ""
+
+
+def _references_recent_image(value: str) -> bool:
+    return bool(_IMAGE_REFERENCE_PATTERN.search(value or ""))
+
+
 class TakeoverService:
-    """Poll BOXIM, prepare replies during the grace period, then send at +3s."""
+    """Poll BOXIM, honor the owner grace period, then generate and send one reply."""
 
     def __init__(
         self,
         session_factory: Callable[[], Session],
         boxim_client: BoxIMClient,
         *,
-        reply_delay_seconds: int = 3,
+        reply_delay_seconds: int | None = None,
+        poll_concurrency: int = 8,
+        max_message_age_seconds: int = DEFAULT_MAX_MESSAGE_AGE_SECONDS,
         now: Callable[[], datetime] = _utcnow,
     ):
         self.session_factory = session_factory
         self.boxim = boxim_client
         self.reply_delay_seconds = reply_delay_seconds
+        self.poll_concurrency = max(1, min(int(poll_concurrency), 64))
+        self.max_message_age_seconds = max(60, int(max_message_age_seconds))
         self.now = now
         self._sessions: dict[str, dict] = {}
         self._poll_lock = asyncio.Lock()
         self._process_lock = asyncio.Lock()
+        self._persist_lock = asyncio.Lock()
 
     async def poll_and_process_messages(self):
         """Run one complete cycle for callers that do not use the split scheduler."""
@@ -102,8 +177,48 @@ class TakeoverService:
             self._recover_stuck_tasks()
             avatar_ids = self._enabled_avatar_ids()
             self._cancel_disabled_tasks(set(avatar_ids))
-            for avatar_id in avatar_ids:
-                await self._sync_avatar(avatar_id)
+            self._ensure_takeover_cursors(avatar_ids)
+            semaphore = asyncio.Semaphore(self.poll_concurrency)
+
+            async def sync(avatar_id: str):
+                async with semaphore:
+                    return await self._sync_avatar(avatar_id)
+
+            results = await asyncio.gather(
+                *(sync(avatar_id) for avatar_id in avatar_ids),
+                return_exceptions=True,
+            )
+            for avatar_id, result in zip(avatar_ids, results):
+                if isinstance(result, Exception):
+                    logger.warning("BOXIM poll crashed for avatar %s: %s", avatar_id, result)
+
+    def _ensure_takeover_cursors(self, avatar_ids: list[str]):
+        """Create durable cursors before concurrent network polling starts."""
+        if not avatar_ids:
+            return
+        db = self.session_factory()
+        try:
+            existing = {
+                row[0]
+                for row in db.query(TakeoverCursor.avatar_id)
+                .filter(TakeoverCursor.avatar_id.in_(avatar_ids))
+                .all()
+            }
+            avatars = (
+                db.query(Avatar.id, Avatar.owner_id)
+                .filter(
+                    Avatar.id.in_(
+                        [avatar_id for avatar_id in avatar_ids if avatar_id not in existing]
+                    )
+                )
+                .all()
+            )
+            for avatar_id, owner_id in avatars:
+                db.add(TakeoverCursor(avatar_id=avatar_id, owner_id=owner_id))
+            if avatars:
+                db.commit()
+        finally:
+            db.close()
 
     async def process_reply_tasks(self):
         """Generate and send replies independently from BOXIM's long poll."""
@@ -119,11 +234,17 @@ class TakeoverService:
     def _enabled_avatar_ids(self) -> list[str]:
         db = self.session_factory()
         try:
-            return [
-                avatar.id
-                for avatar in db.query(Avatar).filter(Avatar.status == "active").all()
-                if _takeover_enabled(avatar)
-            ]
+            avatars = (
+                db.query(Avatar)
+                .filter(Avatar.status == "active")
+                .order_by(Avatar.updated_at.desc(), Avatar.created_at.desc())
+                .all()
+            )
+            selected = {}
+            for avatar in avatars:
+                if _takeover_enabled(avatar) and avatar.owner_id not in selected:
+                    selected[avatar.owner_id] = avatar.id
+            return list(selected.values())
         finally:
             db.close()
 
@@ -201,13 +322,20 @@ class TakeoverService:
     def _forget_boxim_session(self, user_id: str):
         self._sessions.pop(user_id, None)
 
-    def _disable_after_connection_failure(
+    def _record_connection_failure(
         self,
         db: Session,
         avatar: Avatar,
         cursor: TakeoverCursor,
         message: str,
+        *,
+        disable_takeover: bool,
     ):
+        cursor.last_error = message
+        cursor.last_polled_at = self.now()
+        if not disable_takeover:
+            return
+
         permissions = (avatar.config or {}).get("authorizationPermissions", [])
         avatar.config = {
             **(avatar.config or {}),
@@ -217,8 +345,6 @@ class TakeoverService:
                 if permission != TAKEOVER_PERMISSION
             ],
         }
-        cursor.last_error = message
-        cursor.last_polled_at = self.now()
         tasks = (
             db.query(TakeoverReplyTask)
             .filter(
@@ -243,13 +369,17 @@ class TakeoverService:
             if not cursor:
                 cursor = TakeoverCursor(avatar_id=avatar.id, owner_id=avatar.owner_id)
                 db.add(cursor)
-                db.flush()
+                db.commit()
+            else:
+                # Release SQLite's read transaction before the long network poll.
+                db.commit()
             if not user or not user.huihui_token:
-                self._disable_after_connection_failure(
+                self._record_connection_failure(
                     db,
                     avatar,
                     cursor,
                     "请重新登录会会生产账号后再开启主动接管",
+                    disable_takeover=True,
                 )
                 db.commit()
                 return False
@@ -268,11 +398,24 @@ class TakeoverService:
                 if isinstance(exc, BoxIMError) and exc.auth_error:
                     self._forget_boxim_session(user.id)
                     message = "BOXIM 授权已失效，请重新登录会会生产账号"
+                    disable_takeover = True
                 else:
                     message = f"BOXIM 暂时连接失败：{str(exc)[:160]}"
-                self._disable_after_connection_failure(db, avatar, cursor, message)
+                    disable_takeover = False
+                self._record_connection_failure(
+                    db,
+                    avatar,
+                    cursor,
+                    message,
+                    disable_takeover=disable_takeover,
+                )
                 db.commit()
-                logger.warning("BOXIM sync failed for avatar %s: %s", avatar.id, exc)
+                logger.warning(
+                    "BOXIM sync failed for avatar %s (will_retry=%s): %s",
+                    avatar.id,
+                    not disable_takeover,
+                    exc,
+                )
                 return False
 
             messages.sort(key=lambda item: (_numeric_id(item.get("id")), item.get("sendTime") or 0))
@@ -280,13 +423,6 @@ class TakeoverService:
             max_message_id = _numeric_id(cursor.last_message_id)
             read_receipts: dict[str, int] = {}
             for message in messages:
-                self._record_message(
-                    db,
-                    avatar,
-                    cursor.boxim_owner_id,
-                    message,
-                    schedule_reply=not priming,
-                )
                 message_id = _numeric_id(message.get("id"))
                 max_message_id = max(max_message_id, message_id)
                 send_id = str(message.get("sendId") or "")
@@ -301,6 +437,17 @@ class TakeoverService:
                     session["access_token"], peer_id, message_id
                 )
 
+            # Keep SQLite write transactions short. The read-receipt request above
+            # can block on the network and must not hold the database write lock.
+            async with self._persist_lock:
+                for message in messages:
+                    self._record_message(
+                        db,
+                        avatar,
+                        cursor.boxim_owner_id,
+                        message,
+                        schedule_reply=not priming,
+                    )
                 cursor.last_message_id = str(max_message_id)
                 cursor.initialized = True
                 cursor.last_polled_at = self.now()
@@ -350,13 +497,21 @@ class TakeoverService:
 
         now = self.now()
         send_time = _boxim_time(message.get("sendTime"), now)
-        is_avatar = False
-        if direction == "outgoing" and local_id:
+        is_avatar = _is_avatar_local_id(local_id)
+        if not is_avatar and local_id:
             is_avatar = bool(
                 db.query(TakeoverReplyTask)
                 .filter(
-                    TakeoverReplyTask.owner_id == avatar.owner_id,
                     TakeoverReplyTask.boxim_local_id == local_id,
+                    TakeoverReplyTask.status.in_(("ready", "sending", "sent")),
+                )
+                .first()
+            )
+        if not is_avatar:
+            is_avatar = bool(
+                db.query(TakeoverReplyTask)
+                .filter(
+                    TakeoverReplyTask.boxim_sent_message_id == message_id,
                     TakeoverReplyTask.status == "sent",
                 )
                 .first()
@@ -381,12 +536,96 @@ class TakeoverService:
             if not is_avatar:
                 self._cancel_conversation(db, avatar.owner_id, peer_id, "owner_replied")
             return
-        if not schedule_reply or event.message_type != 0 or not event.content.strip():
+        if not schedule_reply or event.message_type not in {
+            BOXIM_TEXT_MESSAGE_TYPE,
+            BOXIM_IMAGE_MESSAGE_TYPE,
+        }:
             return
-        if (now - send_time).total_seconds() > MAX_STALE_SECONDS:
+        if event.message_type == BOXIM_TEXT_MESSAGE_TYPE and not event.content.strip():
+            return
+        if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
+            try:
+                parse_boxim_image_url(
+                    event.content,
+                    base_url=getattr(self.boxim, "im_base_url", ""),
+                )
+            except BoxIMImageError as exc:
+                logger.warning(
+                    "Ignored invalid BOXIM image message %s for avatar %s: %s",
+                    message_id,
+                    avatar.id,
+                    exc,
+                )
+                return
+        if (now - send_time).total_seconds() > self.max_message_age_seconds:
+            logger.info(
+                "Ignored stale BOXIM message %s for avatar %s (age=%ss)",
+                message_id,
+                avatar.id,
+                int((now - send_time).total_seconds()),
+            )
+            return
+        if is_avatar:
+            self._cancel_conversation(db, avatar.owner_id, peer_id, "peer_avatar_message")
+            logger.info(
+                "Skipped BOXIM reply for avatar %s message %s: peer_avatar_message",
+                avatar.id,
+                message_id,
+            )
+            return
+        if self._human_pause_active(db, avatar.owner_id, peer_id, now):
+            self._cancel_conversation(db, avatar.owner_id, peer_id, "owner_active")
+            logger.info(
+                "Skipped BOXIM reply for avatar %s message %s: owner_active",
+                avatar.id,
+                message_id,
+            )
+            return
+        if self._conversation_rate_limited(db, avatar.owner_id, peer_id, now):
+            self._cancel_conversation(db, avatar.owner_id, peer_id, "rate_limited")
+            logger.info(
+                "Skipped BOXIM reply for avatar %s message %s: rate_limited",
+                avatar.id,
+                message_id,
+            )
             return
         self._schedule_reply(db, avatar, event)
 
+    @staticmethod
+    def _human_pause_active(db: Session, owner_id: str, peer_id: str, now: datetime) -> bool:
+        threshold = now - timedelta(seconds=HUMAN_PAUSE_SECONDS)
+        return bool(
+            db.query(TakeoverMessage.id)
+            .filter(
+                TakeoverMessage.owner_id == owner_id,
+                TakeoverMessage.peer_id == peer_id,
+                TakeoverMessage.direction == "outgoing",
+                TakeoverMessage.is_avatar.is_(False),
+                TakeoverMessage.send_time >= threshold,
+            )
+            .first()
+        )
+
+    @staticmethod
+    def _conversation_rate_limited(
+        db: Session,
+        owner_id: str,
+        peer_id: str,
+        now: datetime,
+    ) -> bool:
+        threshold = now - timedelta(seconds=RATE_LIMIT_WINDOW_SECONDS)
+        return (
+            db.query(TakeoverReplyTask.id)
+            .filter(
+                TakeoverReplyTask.owner_id == owner_id,
+                TakeoverReplyTask.peer_id == peer_id,
+                TakeoverReplyTask.status == "sent",
+                TakeoverReplyTask.sent_at >= threshold,
+            )
+            .count()
+            >= RATE_LIMIT_MAX_REPLIES
+        )
+
     @staticmethod
     def _cancel_conversation(db: Session, owner_id: str, peer_id: str, reason: str):
         tasks = (
@@ -424,12 +663,25 @@ class TakeoverService:
                 task.status = "cancelled"
                 task.cancel_reason = "newer_incoming_message"
                 task.locked_at = None
-        prompt_parts.append(event.content.strip())
+        if event.message_type == BOXIM_TEXT_MESSAGE_TYPE:
+            for image_event in self._recent_unhandled_images(
+                db,
+                avatar,
+                event,
+                source_ids,
+            ):
+                prompt_parts.append(_event_prompt(image_event))
+                source_ids.append(image_event.boxim_message_id)
+        prompt_parts.append(_event_prompt(event))
         source_ids.append(event.boxim_message_id)
         prompt = "\n".join(part for part in prompt_parts if part).strip()[-MAX_PROMPT_LENGTH:]
-        due_at = event.send_time + timedelta(seconds=self.reply_delay_seconds)
+        due_at = max(
+            event.send_time
+            + timedelta(seconds=_configured_reply_delay(avatar, self.reply_delay_seconds)),
+            self.now(),
+        )
         task_id = secrets.token_hex(16)
-        local_id = int(time.time() * 1000) * 1000 + secrets.randbelow(1000)
+        local_id = _avatar_local_id(avatar.owner_id, event.boxim_message_id)
         db.add(
             TakeoverReplyTask(
                 id=task_id,
@@ -445,6 +697,68 @@ class TakeoverService:
             )
         )
 
+    @staticmethod
+    def _recent_unhandled_images(
+        db: Session,
+        avatar: Avatar,
+        event: TakeoverMessage,
+        current_source_ids: list[str],
+    ) -> list[TakeoverMessage]:
+        """Recover missed images, or reuse a referenced image from the last two days."""
+        references_image = _references_recent_image(event.content)
+        lookback_seconds = (
+            IMAGE_REFERENCE_LOOKBACK_SECONDS
+            if references_image
+            else IMAGE_CONTEXT_LOOKBACK_SECONDS
+        )
+        threshold = event.send_time - timedelta(seconds=lookback_seconds)
+        candidates = (
+            db.query(TakeoverMessage)
+            .filter(
+                TakeoverMessage.avatar_id == avatar.id,
+                TakeoverMessage.owner_id == avatar.owner_id,
+                TakeoverMessage.peer_id == event.peer_id,
+                TakeoverMessage.direction == "incoming",
+                TakeoverMessage.message_type == BOXIM_IMAGE_MESSAGE_TYPE,
+                TakeoverMessage.is_avatar.is_(False),
+                TakeoverMessage.send_time >= threshold,
+                TakeoverMessage.send_time <= event.send_time,
+            )
+            .order_by(TakeoverMessage.send_time.desc())
+            .limit(MAX_RECENT_IMAGE_CONTEXTS)
+            .all()
+        )
+        if not candidates:
+            return []
+
+        current_ids = set(current_source_ids)
+        if references_image:
+            return [
+                image
+                for image in reversed(candidates)
+                if image.boxim_message_id not in current_ids
+            ]
+
+        handled_ids = set(current_ids)
+        task_sources = (
+            db.query(TakeoverReplyTask.source_message_ids)
+            .filter(
+                TakeoverReplyTask.avatar_id == avatar.id,
+                TakeoverReplyTask.owner_id == avatar.owner_id,
+                TakeoverReplyTask.peer_id == event.peer_id,
+                TakeoverReplyTask.created_at >= threshold,
+            )
+            .all()
+        )
+        for (source_message_ids,) in task_sources:
+            handled_ids.update(source_message_ids or [])
+
+        return [
+            image
+            for image in reversed(candidates)
+            if image.boxim_message_id not in handled_ids
+        ]
+
     async def _prepare_replies(self) -> int:
         db = self.session_factory()
         try:
@@ -455,6 +769,7 @@ class TakeoverService:
                     .filter(
                         TakeoverReplyTask.status.in_(GENERATABLE_TASK_STATUSES),
                         TakeoverReplyTask.response_text == "",
+                        TakeoverReplyTask.scheduled_at <= self.now(),
                     )
                     .order_by(TakeoverReplyTask.created_at.asc())
                     .limit(10)
@@ -478,6 +793,50 @@ class TakeoverService:
         results = await asyncio.gather(*(generate(task_id) for task_id in task_ids))
         return sum(bool(result) for result in results)
 
+    def _takeover_image_attachment(
+        self,
+        db: Session,
+        avatar: Avatar,
+        event: TakeoverMessage,
+    ) -> ChatAttachment:
+        now = self.now()
+        if event.attachment_id:
+            cached = db.get(ChatAttachment, event.attachment_id)
+            if cached and cached.status == "ready" and cached.expires_at > now:
+                cached.used_at = now
+                db.commit()
+                return cached
+
+        downloaded = download_boxim_image(
+            event.content,
+            base_url=getattr(
+                self.boxim,
+                "im_base_url",
+                os.getenv("BOXIM_API_BASE_URL", "https://im.99hui.com/api"),
+            ),
+        )
+        from routers.chat import _analyze_image_bytes
+
+        attachment = _analyze_image_bytes(
+            db,
+            avatar,
+            downloaded.content,
+            filename=downloaded.filename,
+            mime_type=downloaded.mime_type,
+            uploader_kind="boxim",
+        )
+        event.attachment_id = attachment.id
+        attachment.used_at = now
+        db.commit()
+        logger.info(
+            "BOXIM image analyzed message=%s attachment=%s avatar=%s category=%s",
+            event.boxim_message_id,
+            attachment.id,
+            avatar.id,
+            attachment.category,
+        )
+        return attachment
+
     def _generate_reply(self, task_id: str) -> bool:
         db = self.session_factory()
         try:
@@ -496,19 +855,59 @@ class TakeoverService:
             db.commit()
 
             excluded_ids = set(task.source_message_ids or [])
+            source_events = {
+                event.boxim_message_id: event
+                for event in (
+                    db.query(TakeoverMessage)
+                    .filter(
+                        TakeoverMessage.owner_id == task.owner_id,
+                        TakeoverMessage.peer_id == task.peer_id,
+                        TakeoverMessage.avatar_id == task.avatar_id,
+                        TakeoverMessage.boxim_message_id.in_(excluded_ids),
+                    )
+                    .all()
+                    if excluded_ids
+                    else []
+                )
+            }
             events = (
                 db.query(TakeoverMessage)
                 .filter(
                     TakeoverMessage.owner_id == task.owner_id,
                     TakeoverMessage.peer_id == task.peer_id,
+                    TakeoverMessage.avatar_id == task.avatar_id,
                 )
                 .order_by(TakeoverMessage.send_time.desc())
                 .limit(30)
                 .all()
             )
+            image_attachments = []
+            image_failed = False
+            for message_id in (task.source_message_ids or [])[-3:]:
+                event = source_events.get(message_id)
+                if not event or event.message_type != BOXIM_IMAGE_MESSAGE_TYPE:
+                    continue
+                try:
+                    image_attachments.append(
+                        self._takeover_image_attachment(db, avatar, event)
+                    )
+                except (BoxIMImageError, ImageValidationError) as exc:
+                    image_failed = True
+                    logger.warning(
+                        "BOXIM image unavailable message=%s avatar=%s: %s",
+                        event.boxim_message_id,
+                        avatar.id,
+                        exc,
+                    )
             history = []
             for event in reversed(events):
-                if event.boxim_message_id in excluded_ids or not event.content.strip():
+                if (
+                    event.boxim_message_id in excluded_ids
+                    or event.message_type != BOXIM_TEXT_MESSAGE_TYPE
+                    or not event.content.strip()
+                ):
+                    continue
+                if event.direction == "incoming" and event.is_avatar:
                     continue
                 history.append(
                     {
@@ -518,9 +917,20 @@ class TakeoverService:
                 )
             history = history[-10:]
 
-            from routers.chat import _resolve_reply
+            from routers.chat import _attachment_contexts, _resolve_reply
 
-            result = _resolve_reply(db, avatar, task.prompt, history)
+            image_contexts = _attachment_contexts(image_attachments)
+            if image_failed and not image_contexts:
+                answer = BOXIM_IMAGE_UNAVAILABLE_REPLY
+            else:
+                result = _resolve_reply(
+                    db,
+                    avatar,
+                    task.prompt,
+                    history,
+                    usage_source="takeover",
+                    image_contexts=image_contexts,
+                )
                 answer = _plain_text_reply(result.get("answer", ""))
             db.refresh(task)
             if task.status != "generating":
@@ -582,11 +992,20 @@ class TakeoverService:
                 task.cancel_reason = "takeover_disabled"
                 db.commit()
                 return False
-            if (self.now() - task.scheduled_at).total_seconds() > MAX_STALE_SECONDS:
+            if (self.now() - task.scheduled_at).total_seconds() > MAX_SEND_OVERDUE_SECONDS:
                 task.status = "cancelled"
                 task.cancel_reason = "stale_reply"
                 db.commit()
                 return False
+            cursor = (
+                db.query(TakeoverCursor)
+                .filter(TakeoverCursor.avatar_id == task.avatar_id)
+                .first()
+            )
+            if not cursor or not cursor.last_polled_at or cursor.last_polled_at < task.scheduled_at:
+                # Do not race the owner's final seconds of the grace period. A
+                # completed poll at/after the due time must confirm no human reply.
+                return False
             user = db.query(User).filter(User.huihui_user_id == task.owner_id).first()
             if not user or not user.huihui_token:
                 raise BoxIMError("缺少会会登录凭证", auth_error=True)

digital-avatar-app/backend/services/token_billing.py

@@ -0,0 +1,203 @@
+"""User-scoped token accounting for every avatar model request."""
+
+import math
+from dataclasses import dataclass
+from datetime import datetime, timedelta
+
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
+
+from models import Avatar, TokenAccount, TokenUsage, User
+
+DEFAULT_TOKEN_GRANT = 1_000_000
+
+
+class InsufficientTokensError(RuntimeError):
+    pass
+
+
+@dataclass(frozen=True)
+class TokenReservation:
+    usage_id: str
+    user_id: str
+    reserved_tokens: int
+
+
+def get_or_create_account(db: Session, user_id: str) -> TokenAccount:
+    account = db.query(TokenAccount).filter(TokenAccount.user_id == user_id).first()
+    if account:
+        return account
+    account = TokenAccount(
+        user_id=user_id,
+        balance=DEFAULT_TOKEN_GRANT,
+        total_granted=DEFAULT_TOKEN_GRANT,
+        total_consumed=0,
+    )
+    db.add(account)
+    try:
+        db.commit()
+    except IntegrityError:
+        # A concurrent first request may have created the same user account.
+        db.rollback()
+        account = db.query(TokenAccount).filter(TokenAccount.user_id == user_id).first()
+        if account is None:
+            raise
+    db.refresh(account)
+    return account
+
+
+def avatar_owner_user(db: Session, avatar: Avatar) -> User | None:
+    owner_id = (avatar.owner_id or "").strip()
+    if not owner_id:
+        return None
+    return db.query(User).filter(User.huihui_user_id == owner_id).first()
+
+
+def estimate_request_tokens(messages: list[dict], max_output_tokens: int) -> int:
+    # UTF-8 bytes / 2 deliberately overestimates mixed Chinese/English prompts;
+    # the unused reservation is returned after provider usage is received.
+    content_bytes = sum(
+        len(str(item.get("content", "")).encode("utf-8"))
+        for item in messages
+    )
+    prompt_reserve = max(1, math.ceil(content_bytes / 2) + len(messages) * 6)
+    return prompt_reserve + max(1, int(max_output_tokens))
+
+
+def estimate_fallback_usage(messages: list[dict], output: str) -> int:
+    content_bytes = sum(
+        len(str(item.get("content", "")).encode("utf-8"))
+        for item in messages
+    ) + len((output or "").encode("utf-8"))
+    return max(1, math.ceil(content_bytes / 3) + len(messages) * 4)
+
+
+def reserve_avatar_tokens(
+    db: Session,
+    avatar: Avatar,
+    source: str,
+    model: str,
+    messages: list[dict],
+    max_output_tokens: int,
+    *,
+    minimum_reserve_tokens: int = 0,
+) -> TokenReservation:
+    user = avatar_owner_user(db, avatar)
+    if not user:
+        raise InsufficientTokensError("分身尚未关联有效用户，暂时无法使用积分")
+    account = get_or_create_account(db, user.id)
+    reserved = max(
+        estimate_request_tokens(messages, max_output_tokens),
+        max(0, int(minimum_reserve_tokens or 0)),
+    )
+    updated = (
+        db.query(TokenAccount)
+        .filter(TokenAccount.id == account.id, TokenAccount.balance >= reserved)
+        .update(
+            {TokenAccount.balance: TokenAccount.balance - reserved},
+            synchronize_session=False,
+        )
+    )
+    if updated != 1:
+        db.rollback()
+        raise InsufficientTokensError("积分余额不足，请充值后继续")
+    db.refresh(account)
+    usage = TokenUsage(
+        user_id=user.id,
+        avatar_id=avatar.id,
+        source=source,
+        model=model,
+        status="reserved",
+        reserved_tokens=reserved,
+    )
+    db.add(usage)
+    db.flush()
+    usage.balance_after = account.balance
+    db.commit()
+    return TokenReservation(usage.id, user.id, reserved)
+
+
+def settle_reservation(
+    db: Session,
+    reservation: TokenReservation,
+    usage: dict | None,
+    *,
+    fallback_total: int,
+) -> dict:
+    record = db.query(TokenUsage).filter(TokenUsage.id == reservation.usage_id).first()
+    if not record or record.status != "reserved":
+        return {}
+    provider_usage = usage or {}
+    prompt_tokens = max(0, int(provider_usage.get("prompt_tokens") or 0))
+    completion_tokens = max(0, int(provider_usage.get("completion_tokens") or 0))
+    provider_total = max(
+        int(provider_usage.get("total_tokens") or 0),
+        prompt_tokens + completion_tokens,
+    )
+    total_tokens = max(1, provider_total or int(fallback_total or 0))
+    updated = (
+        db.query(TokenAccount)
+        .filter(TokenAccount.user_id == reservation.user_id)
+        .update(
+            {
+                TokenAccount.balance: TokenAccount.balance + reservation.reserved_tokens - total_tokens,
+                TokenAccount.total_consumed: TokenAccount.total_consumed + total_tokens,
+            },
+            synchronize_session=False,
+        )
+    )
+    if updated != 1:
+        raise RuntimeError("积分账户不存在")
+    db.expire_all()
+    account = db.query(TokenAccount).filter(TokenAccount.user_id == reservation.user_id).first()
+    record.prompt_tokens = prompt_tokens
+    record.completion_tokens = completion_tokens
+    record.total_tokens = total_tokens
+    record.balance_after = account.balance
+    record.status = "completed"
+    record.settled_at = datetime.utcnow()
+    db.commit()
+    return {
+        "promptTokens": prompt_tokens,
+        "completionTokens": completion_tokens,
+        "totalTokens": total_tokens,
+        "balance": account.balance,
+    }
+
+
+def release_reservation(db: Session, reservation: TokenReservation, reason: str = "") -> None:
+    record = db.query(TokenUsage).filter(TokenUsage.id == reservation.usage_id).first()
+    if not record or record.status != "reserved":
+        return
+    updated = (
+        db.query(TokenAccount)
+        .filter(TokenAccount.user_id == reservation.user_id)
+        .update(
+            {TokenAccount.balance: TokenAccount.balance + reservation.reserved_tokens},
+            synchronize_session=False,
+        )
+    )
+    if updated:
+        db.expire_all()
+        account = db.query(TokenAccount).filter(TokenAccount.user_id == reservation.user_id).first()
+        record = db.query(TokenUsage).filter(TokenUsage.id == reservation.usage_id).first()
+        record.balance_after = account.balance
+    record.status = "failed"
+    record.failure_reason = (reason or "model_request_failed")[:255]
+    record.settled_at = datetime.utcnow()
+    db.commit()
+
+
+def release_stale_reservations(db: Session, older_than_minutes: int = 10) -> int:
+    cutoff = datetime.utcnow() - timedelta(minutes=older_than_minutes)
+    stale = db.query(TokenUsage).filter(
+        TokenUsage.status == "reserved",
+        TokenUsage.created_at < cutoff,
+    ).all()
+    for record in stale:
+        release_reservation(
+            db,
+            TokenReservation(record.id, record.user_id, int(record.reserved_tokens or 0)),
+            "stale_reservation_recovered",
+        )
+    return len(stale)

digital-avatar-app/backend/services/vision_service.py

@@ -0,0 +1,196 @@
+"""Private image normalization and OpenAI-compatible vision model calls."""
+
+import base64
+import io
+import json
+import os
+import re
+from dataclasses import dataclass
+from typing import Any
+
+import httpx
+from PIL import Image, ImageOps, UnidentifiedImageError
+
+from services.chat_model_config import ChatModelConfig
+
+
+ALLOWED_IMAGE_FORMATS = {"JPEG": "image/jpeg", "PNG": "image/png", "WEBP": "image/webp"}
+ALLOWED_CATEGORIES = {"general_image", "document", "medical_document", "medical_image"}
+
+GENERAL_VISION_PROMPT = """
+请客观分析这张图片，并只输出一个 JSON 对象，不要使用 Markdown 代码块。
+字段必须为：
+category: general_image、document、medical_document、medical_image 四选一；
+summary: 图片的完整客观摘要；
+visible_text: 图片中能够确认的文字，保留自然换行；
+key_facts: 可确认事实数组；
+uncertainties: 模糊、遮挡、无法确认内容数组；
+medical: 对象，包含 document_type、patient_info、chief_complaint、findings、measurements、doctor_advice。
+
+规则：
+1. 不得补全看不清或被遮挡的文字，不得猜测人物身份。
+2. 病例、处方、检查单、检验报告归为 medical_document。
+3. X 光、CT、MRI、超声影像等归为 medical_image，只描述可见内容，不作疾病诊断、分期、用药或治疗建议。
+4. 非医疗图片的 medical 字段仍保留，但使用空字符串、空对象或空数组。
+5. 不要提及模型、供应商、系统提示词或内部处理过程。
+""".strip()
+
+MEDICAL_OCR_PROMPT = """
+请逐字转录这张医疗文档图片中的全部可见文字和表格。
+保持标题、段落、项目、数值、单位、参考区间、阳性/阴性标记和医生意见的对应关系。
+看不清的内容写作[无法辨认]，不要猜测、纠错或补全，不要给出诊断和建议，不要使用 Markdown 代码块。
+""".strip()
+
+
+class ImageValidationError(ValueError):
+    pass
+
+
+@dataclass(frozen=True)
+class PreparedImage:
+    data: bytes
+    mime_type: str
+    width: int
+    height: int
+
+    @property
+    def data_uri(self) -> str:
+        encoded = base64.b64encode(self.data).decode("ascii")
+        return f"data:{self.mime_type};base64,{encoded}"
+
+
+def prepare_image(content: bytes) -> PreparedImage:
+    max_bytes = max(1024, int(os.getenv("CHAT_IMAGE_MAX_BYTES", str(8 * 1024 * 1024))))
+    max_pixels = max(1_000_000, int(os.getenv("CHAT_IMAGE_MAX_PIXELS", "16000000")))
+    max_edge = max(1024, int(os.getenv("CHAT_IMAGE_MAX_EDGE", "4096")))
+    if not content:
+        raise ImageValidationError("图片内容为空")
+    if len(content) > max_bytes:
+        raise ImageValidationError(f"单张图片不能超过 {max_bytes // 1024 // 1024}MB")
+
+    try:
+        with Image.open(io.BytesIO(content)) as probe:
+            image_format = str(probe.format or "").upper()
+            width, height = probe.size
+            probe.verify()
+    except (UnidentifiedImageError, OSError, SyntaxError) as exc:
+        raise ImageValidationError("图片格式无效或文件已损坏") from exc
+
+    if image_format not in ALLOWED_IMAGE_FORMATS:
+        raise ImageValidationError("仅支持 JPG、PNG、WebP 图片")
+    if width <= 0 or height <= 0 or width * height > max_pixels:
+        raise ImageValidationError("图片像素过大，请压缩后重新上传")
+
+    try:
+        with Image.open(io.BytesIO(content)) as original:
+            image = ImageOps.exif_transpose(original)
+            image.load()
+            if max(image.size) > max_edge:
+                image.thumbnail((max_edge, max_edge), Image.Resampling.LANCZOS)
+            if image.mode in {"RGBA", "LA"}:
+                canvas = Image.new("RGB", image.size, "white")
+                alpha = image.getchannel("A")
+                canvas.paste(image.convert("RGB"), mask=alpha)
+                image = canvas
+            elif image.mode != "RGB":
+                image = image.convert("RGB")
+            output = io.BytesIO()
+            image.save(output, format="JPEG", quality=92, optimize=True)
+            normalized = output.getvalue()
+            normalized_width, normalized_height = image.size
+    except (OSError, ValueError) as exc:
+        raise ImageValidationError("图片解码失败，请重新选择图片") from exc
+
+    return PreparedImage(
+        data=normalized,
+        mime_type="image/jpeg",
+        width=normalized_width,
+        height=normalized_height,
+    )
+
+
+def call_vision_model(
+    prepared: PreparedImage,
+    model_config: ChatModelConfig,
+    *,
+    model: str,
+    prompt: str,
+    json_output: bool,
+) -> dict:
+    if not model_config.api_key:
+        raise RuntimeError("视觉模型服务未配置")
+    payload: dict[str, Any] = {
+        "model": model,
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "image_url", "image_url": {"url": prepared.data_uri}},
+                    {"type": "text", "text": prompt},
+                ],
+            }
+        ],
+        "temperature": 0,
+        "max_tokens": model_config.vision_max_tokens,
+    }
+    if json_output:
+        payload["response_format"] = {"type": "json_object"}
+    try:
+        response = httpx.post(
+            f"{model_config.api_base_url}/chat/completions",
+            headers={"Authorization": f"Bearer {model_config.api_key}"},
+            json=payload,
+            timeout=model_config.vision_timeout_seconds,
+        )
+        response.raise_for_status()
+        data = response.json()
+        content = data.get("choices", [{}])[0].get("message", {}).get("content", "")
+    except (httpx.HTTPError, ValueError, KeyError, IndexError) as exc:
+        raise RuntimeError("图片识别服务暂时不可用") from exc
+    if not isinstance(content, str) or not content.strip():
+        raise RuntimeError("图片识别服务没有返回有效结果")
+    return {"content": content.strip(), "usage": data.get("usage") or {}}
+
+
+def parse_vision_analysis(content: str) -> dict:
+    value = (content or "").strip()
+    fenced = re.match(r"^```(?:json)?\s*(.*?)\s*```$", value, re.DOTALL | re.IGNORECASE)
+    if fenced:
+        value = fenced.group(1).strip()
+    try:
+        payload = json.loads(value)
+    except (TypeError, ValueError) as exc:
+        raise RuntimeError("图片识别结果格式无效") from exc
+    if not isinstance(payload, dict):
+        raise RuntimeError("图片识别结果格式无效")
+
+    category = str(payload.get("category") or "general_image").strip().lower()
+    if category not in ALLOWED_CATEGORIES:
+        category = "general_image"
+    medical = payload.get("medical") if isinstance(payload.get("medical"), dict) else {}
+    return {
+        "category": category,
+        "summary": str(payload.get("summary") or "").strip(),
+        "visible_text": str(payload.get("visible_text") or "").strip(),
+        "key_facts": _string_list(payload.get("key_facts")),
+        "uncertainties": _string_list(payload.get("uncertainties")),
+        "medical": medical,
+    }
+
+
+def build_attachment_warning(analysis: dict, *, ocr_failed: bool = False) -> str:
+    warnings = list(analysis.get("uncertainties") or [])
+    category = analysis.get("category")
+    if ocr_failed:
+        warnings.append("精确文字识别暂时不可用，请人工核对图片原文")
+    if category == "medical_document":
+        warnings.append("病例识别结果仅供辅助，不能替代医生诊断，请核对原始文档")
+    elif category == "medical_image":
+        warnings.append("医学影像仅作客观描述，不能替代影像报告和医生诊断")
+    return "；".join(dict.fromkeys(item for item in warnings if item))
+
+
+def _string_list(value: Any) -> list[str]:
+    if not isinstance(value, list):
+        return []
+    return [str(item).strip() for item in value if str(item).strip()]

digital-avatar-app/backend/tests/conftest.py

@@ -5,9 +5,13 @@ from database import init_db, SessionLocal
 from models import (
     Authorization,
     Avatar,
+    ChatAttachment,
     TakeoverCursor,
     TakeoverMessage,
     TakeoverReplyTask,
+    TokenAccount,
+    TokenPaymentOrder,
+    TokenUsage,
     User,
 )
 
@@ -92,6 +96,9 @@ def authorization_context():
     finally:
         db.rollback()
         avatar_ids = [avatar.id, other_avatar.id]
+        db.query(ChatAttachment).filter(
+            ChatAttachment.avatar_id.in_(avatar_ids)
+        ).delete(synchronize_session=False)
         db.query(TakeoverReplyTask).filter(
             TakeoverReplyTask.avatar_id.in_(avatar_ids)
         ).delete(synchronize_session=False)
@@ -107,6 +114,16 @@ def authorization_context():
         db.query(Avatar).filter(Avatar.id.in_(avatar_ids)).delete(
             synchronize_session=False
         )
+        user_ids = [owner.id, other.id]
+        db.query(TokenPaymentOrder).filter(TokenPaymentOrder.user_id.in_(user_ids)).delete(
+            synchronize_session=False
+        )
+        db.query(TokenUsage).filter(TokenUsage.user_id.in_(user_ids)).delete(
+            synchronize_session=False
+        )
+        db.query(TokenAccount).filter(TokenAccount.user_id.in_(user_ids)).delete(
+            synchronize_session=False
+        )
         db.query(User).filter(User.id.in_([owner.id, other.id])).delete(
             synchronize_session=False
         )

digital-avatar-app/backend/tests/test_authorizations_api.py

@@ -103,6 +103,7 @@ def test_avatar_permission_settings_default_and_persist(authorization_context):
     assert initial["data"] == {
         "avatarId": context["avatar"].id,
         "permissions": ["friend", "chat"],
+        "takeoverReplyDelaySeconds": 180,
     }
 
     updated = client.put(
@@ -115,6 +116,7 @@ def test_avatar_permission_settings_default_and_persist(authorization_context):
 
     reloaded = client.get(endpoint, headers=context["owner_headers"]).json()
     assert reloaded["data"]["permissions"] == ["friend", "publish", "interact", "takeover"]
+    assert reloaded["data"]["takeoverReplyDelaySeconds"] == 180
 
 
 def test_avatar_permission_settings_allow_all_disabled(authorization_context):
@@ -156,3 +158,57 @@ def test_avatar_permission_settings_validate_owner_and_permissions(authorization
 
     unauthenticated = client.get(endpoint)
     assert unauthenticated.status_code == 401
+
+
+def test_takeover_delay_minimum_and_single_active_avatar_per_owner(authorization_context):
+    from database import SessionLocal
+    from models import Avatar
+
+    context = authorization_context
+    endpoint = f"/api/avatar/{context['avatar'].id}/permission-settings"
+    invalid = client.put(
+        endpoint,
+        headers=context["owner_headers"],
+        json={"permissions": ["chat"], "takeoverReplyDelaySeconds": 2},
+    ).json()
+    assert invalid["code"] == 400
+
+    second_avatar_id = f"second-{context['suffix']}"
+    db = SessionLocal()
+    try:
+        db.add(
+            Avatar(
+                id=second_avatar_id,
+                owner_id=context["owner"].huihui_user_id,
+                name="第二个分身",
+                status="active",
+                config={"authorizationPermissions": ["chat", "takeover"]},
+            )
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    try:
+        updated = client.put(
+            endpoint,
+            headers=context["owner_headers"],
+            json={"permissions": ["chat", "takeover"], "takeoverReplyDelaySeconds": 3},
+        ).json()
+        assert updated["code"] == 200
+        assert updated["data"]["takeoverReplyDelaySeconds"] == 3
+        assert updated["data"]["disabledAvatarIds"] == [second_avatar_id]
+
+        db = SessionLocal()
+        try:
+            second = db.query(Avatar).filter(Avatar.id == second_avatar_id).one()
+            assert "takeover" not in second.config["authorizationPermissions"]
+        finally:
+            db.close()
+    finally:
+        db = SessionLocal()
+        try:
+            db.query(Avatar).filter(Avatar.id == second_avatar_id).delete()
+            db.commit()
+        finally:
+            db.close()

digital-avatar-app/backend/tests/test_avatar_api.py

@@ -0,0 +1,93 @@
+"""Ownership and configuration-isolation tests for digital avatars."""
+
+from fastapi.testclient import TestClient
+
+from database import SessionLocal
+from main import app
+from models import Avatar
+
+
+client = TestClient(app)
+
+
+def test_avatar_detail_and_update_require_the_owner(authorization_context):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+
+    assert client.get(f"/api/avatar/{avatar_id}").status_code == 401
+    assert client.get(
+        f"/api/avatar/{avatar_id}", headers=context["other_headers"]
+    ).status_code == 403
+
+    updated = client.put(
+        f"/api/avatar/{avatar_id}",
+        headers=context["owner_headers"],
+        json={
+            "description": "独立描述",
+            "config": {"replyStyle": "concise"},
+        },
+    )
+    assert updated.status_code == 200
+    assert updated.json()["data"]["description"] == "独立描述"
+
+    forbidden = client.put(
+        f"/api/avatar/{avatar_id}",
+        headers=context["other_headers"],
+        json={"description": "越权修改"},
+    )
+    assert forbidden.status_code == 403
+
+
+def test_avatar_config_updates_do_not_erase_takeover_or_knowledge_scope(authorization_context):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    db = SessionLocal()
+    try:
+        avatar = db.query(Avatar).filter(Avatar.id == avatar_id).one()
+        avatar.config = {
+            "authorizationPermissions": ["chat", "takeover"],
+            "takeoverReplyDelaySeconds": 180,
+        }
+        db.commit()
+    finally:
+        db.close()
+
+    response = client.put(
+        f"/api/avatar/{avatar_id}",
+        headers=context["owner_headers"],
+        json={"config": {"replyStyle": "warm", "creativity": 25}},
+    ).json()
+    config = response["data"]["config"]
+    assert config["replyStyle"] == "warm"
+    assert config["creativity"] == 25
+    assert config["authorizationPermissions"] == ["chat", "takeover"]
+    assert config["takeoverReplyDelaySeconds"] == 180
+
+
+def test_avatar_create_and_delete_require_login_and_ownership(authorization_context):
+    context = authorization_context
+    assert client.post("/api/avatar", json={"name": "匿名分身"}).status_code == 401
+
+    created = client.post(
+        "/api/avatar",
+        headers=context["owner_headers"],
+        json={"name": "待删除分身"},
+    )
+    assert created.status_code == 200
+    avatar_id = created.json()["data"]["id"]
+
+    try:
+        assert client.delete(
+            f"/api/avatar/{avatar_id}", headers=context["other_headers"]
+        ).status_code == 403
+        deleted = client.delete(
+            f"/api/avatar/{avatar_id}", headers=context["owner_headers"]
+        ).json()
+        assert deleted["code"] == 200
+    finally:
+        db = SessionLocal()
+        try:
+            db.query(Avatar).filter(Avatar.id == avatar_id).delete()
+            db.commit()
+        finally:
+            db.close()

digital-avatar-app/backend/tests/test_boxim_image_service.py

@@ -0,0 +1,71 @@
+import ipaddress
+import json
+
+import httpx
+import pytest
+
+from services.boxim_image_service import (
+    BoxIMImageError,
+    download_boxim_image,
+    parse_boxim_image_url,
+)
+
+
+def test_parse_boxim_image_prefers_origin_and_supports_relative_url():
+    content = json.dumps({"originUrl": "/files/original.png", "thumbUrl": "/thumb.png"})
+    assert parse_boxim_image_url(content, base_url="https://im.example/api") == (
+        "https://im.example/files/original.png"
+    )
+
+
+def test_download_boxim_image_streams_public_https(monkeypatch):
+    monkeypatch.setattr(
+        "services.boxim_image_service._resolved_addresses",
+        lambda _host, _port: {ipaddress.ip_address("8.8.8.8")},
+    )
+    transport = httpx.MockTransport(
+        lambda request: httpx.Response(
+            200,
+            headers={"content-type": "image/png"},
+            content=b"png-bytes",
+            request=request,
+        )
+    )
+
+    image = download_boxim_image(
+        json.dumps({"originUrl": "https://cdn.example/case%20photo.png"}),
+        transport=transport,
+    )
+
+    assert image.content == b"png-bytes"
+    assert image.filename == "case photo.png"
+    assert image.mime_type == "image/png"
+
+
+def test_download_boxim_image_rejects_private_network_url():
+    with pytest.raises(BoxIMImageError, match="受限网络"):
+        download_boxim_image(
+            json.dumps({"originUrl": "https://127.0.0.1/private.png"}),
+            transport=httpx.MockTransport(lambda request: httpx.Response(200, request=request)),
+        )
+
+
+def test_download_boxim_image_stops_oversized_stream(monkeypatch):
+    monkeypatch.setenv("CHAT_IMAGE_MAX_BYTES", "1024")
+    monkeypatch.setattr(
+        "services.boxim_image_service._resolved_addresses",
+        lambda _host, _port: {ipaddress.ip_address("8.8.8.8")},
+    )
+    transport = httpx.MockTransport(
+        lambda request: httpx.Response(
+            200,
+            headers={"content-length": "2048"},
+            request=request,
+        )
+    )
+
+    with pytest.raises(BoxIMImageError, match="超过大小限制"):
+        download_boxim_image(
+            json.dumps({"originUrl": "https://cdn.example/large.png"}),
+            transport=transport,
+        )

digital-avatar-app/backend/tests/test_chat_images.py

@@ -0,0 +1,332 @@
+import json
+from datetime import datetime, timedelta
+from types import SimpleNamespace
+from unittest.mock import Mock, patch
+
+import pytest
+from fastapi import HTTPException
+from fastapi.testclient import TestClient
+
+from database import SessionLocal
+from main import app
+from models import ChatAttachment
+from routers.chat import (
+    ChatIn,
+    _answer_denies_available_image,
+    _attachment_contexts,
+    _load_chat_attachments,
+    _resolve_reply,
+)
+from services.chat_attachment_service import purge_expired_chat_attachments
+from services.token_billing import InsufficientTokensError
+from services.vision_service import PreparedImage
+
+
+client = TestClient(app)
+
+
+GENERAL_RESULT = {
+    "content": json.dumps({
+        "category": "general_image",
+        "summary": "一张包含产品路线图的截图",
+        "visible_text": "产品路线图",
+        "key_facts": ["包含三个阶段"],
+        "uncertainties": [],
+        "medical": {},
+    }, ensure_ascii=False),
+    "usage": {"total_tokens": 120},
+}
+
+
+def test_owner_can_upload_and_cache_image_analysis(authorization_context):
+    context = authorization_context
+    prepared = PreparedImage(b"jpeg", "image/jpeg", 100, 80)
+    with (
+        patch("routers.chat.prepare_image", return_value=prepared),
+        patch("routers.chat._run_billed_vision_call", return_value=GENERAL_RESULT),
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/chat/images",
+            headers=context["owner_headers"],
+            files={"file": ("roadmap.png", b"image-bytes", "image/png")},
+        )
+
+    assert response.status_code == 200
+    payload = response.json()["data"]
+    assert payload["status"] == "ready"
+    assert payload["category"] == "general_image"
+    assert payload["summary"] == "一张包含产品路线图的截图"
+    db = SessionLocal()
+    try:
+        stored = db.query(ChatAttachment).filter(ChatAttachment.id == payload["id"]).one()
+        assert stored.avatar_id == context["avatar"].id
+        assert stored.extracted_text == "产品路线图"
+        assert stored.structured_data["key_facts"] == ["包含三个阶段"]
+    finally:
+        db.close()
+
+
+def test_non_owner_cannot_upload_chat_image(authorization_context):
+    context = authorization_context
+    response = client.post(
+        f"/api/avatar/{context['avatar'].id}/chat/images",
+        headers=context["other_headers"],
+        files={"file": ("private.png", b"image-bytes", "image/png")},
+    )
+
+    assert response.status_code == 403
+
+
+def test_image_upload_preserves_insufficient_points_response(authorization_context):
+    context = authorization_context
+    with patch(
+        "routers.chat._analyze_image_bytes",
+        side_effect=InsufficientTokensError("积分余额不足"),
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/chat/images",
+            headers=context["owner_headers"],
+            files={"file": ("private.png", b"image-bytes", "image/png")},
+        )
+
+    assert response.status_code == 402
+    assert response.json()["detail"] == "积分余额不足"
+
+
+def test_public_share_can_upload_without_exposing_analysis_details(authorization_context):
+    context = authorization_context
+    db = SessionLocal()
+    try:
+        avatar = db.get(type(context["avatar"]), context["avatar"].id)
+        avatar.share_token = f"share-{context['suffix']}"
+        db.commit()
+        share_token = avatar.share_token
+    finally:
+        db.close()
+
+    with (
+        patch(
+            "routers.chat.prepare_image",
+            return_value=PreparedImage(b"jpeg", "image/jpeg", 100, 80),
+        ),
+        patch("routers.chat._run_billed_vision_call", return_value=GENERAL_RESULT),
+    ):
+        response = client.post(
+            f"/api/public/avatar/{share_token}/chat/images",
+            files={"file": ("visitor.png", b"image-bytes", "image/png")},
+        )
+
+    assert response.status_code == 200
+    payload = response.json()["data"]
+    assert payload["status"] == "ready"
+    assert "structuredData" not in payload
+    assert "extractedText" not in payload
+    assert "visionModel" not in payload
+    assert "ocrModel" not in payload
+
+    db = SessionLocal()
+    try:
+        stored = db.get(ChatAttachment, payload["id"])
+        assert stored.uploader_kind == "public"
+        assert stored.avatar_id == context["avatar"].id
+    finally:
+        db.close()
+
+
+def test_medical_document_uses_ocr_result(authorization_context):
+    context = authorization_context
+    general = {
+        "content": json.dumps({
+            "category": "medical_document",
+            "summary": "血常规报告",
+            "visible_text": "初步文字",
+            "key_facts": [],
+            "uncertainties": [],
+            "medical": {"document_type": "检验报告"},
+        }, ensure_ascii=False),
+        "usage": {},
+    }
+    ocr = {"content": "白细胞 11.2 x10^9/L", "usage": {}}
+    with (
+        patch("routers.chat.prepare_image", return_value=PreparedImage(b"jpeg", "image/jpeg", 100, 80)),
+        patch("routers.chat._run_billed_vision_call", side_effect=[general, ocr]) as model,
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/chat/images",
+            headers=context["owner_headers"],
+            files={"file": ("report.jpg", b"image-bytes", "image/jpeg")},
+        )
+
+    assert response.status_code == 200
+    attachment_id = response.json()["data"]["id"]
+    assert model.call_count == 2
+    assert model.call_args_list[1].kwargs["source"] == "vision_medical_ocr"
+    db = SessionLocal()
+    try:
+        stored = db.query(ChatAttachment).filter(ChatAttachment.id == attachment_id).one()
+        assert stored.extracted_text == "白细胞 11.2 x10^9/L"
+        assert stored.ocr_model == "qwen-vl-ocr"
+        assert "不能替代医生诊断" in stored.warning
+    finally:
+        db.close()
+
+
+def test_attachment_cannot_cross_avatar_boundary(authorization_context):
+    context = authorization_context
+    db = SessionLocal()
+    try:
+        attachment = ChatAttachment(
+            avatar_id=context["avatar"].id,
+            filename="private.jpg",
+            status="ready",
+            expires_at=datetime.utcnow() + timedelta(hours=1),
+        )
+        db.add(attachment)
+        db.commit()
+        body = ChatIn(message="看看图片", attachmentIds=[attachment.id])
+        with pytest.raises(HTTPException, match="不属于当前分身") as caught:
+            _load_chat_attachments(db, context["other_avatar"].id, body)
+        assert caught.value.status_code == 400
+    finally:
+        db.close()
+
+
+def test_expired_attachment_is_removed(authorization_context):
+    context = authorization_context
+    db = SessionLocal()
+    try:
+        attachment = ChatAttachment(
+            avatar_id=context["avatar"].id,
+            filename="expired.jpg",
+            status="ready",
+            expires_at=datetime.utcnow() - timedelta(seconds=1),
+        )
+        db.add(attachment)
+        db.commit()
+        attachment_id = attachment.id
+        body = ChatIn(message="看看图片", attachmentIds=[attachment_id])
+        with pytest.raises(HTTPException):
+            _load_chat_attachments(db, context["avatar"].id, body)
+        assert db.query(ChatAttachment).filter(ChatAttachment.id == attachment_id).first() is None
+    finally:
+        db.close()
+
+
+def test_cleanup_keeps_unexpired_attachment(authorization_context):
+    context = authorization_context
+    now = datetime.utcnow()
+    db = SessionLocal()
+    try:
+        expired = ChatAttachment(
+            avatar_id=context["avatar"].id,
+            filename="expired.jpg",
+            status="ready",
+            expires_at=now - timedelta(seconds=1),
+        )
+        active = ChatAttachment(
+            avatar_id=context["avatar"].id,
+            filename="active.jpg",
+            status="ready",
+            expires_at=now + timedelta(hours=1),
+        )
+        db.add_all([expired, active])
+        db.commit()
+        expired_id, active_id = expired.id, active.id
+
+        assert purge_expired_chat_attachments(db, now=now) == 1
+        assert db.get(ChatAttachment, expired_id) is None
+        assert db.get(ChatAttachment, active_id) is not None
+    finally:
+        db.close()
+
+
+def test_image_context_keeps_standard_answer_authoritative():
+    avatar = SimpleNamespace(
+        id="avatar-vision",
+        name="测试分身",
+        description="产品顾问",
+        config={},
+    )
+    model = Mock(return_value="标准退款期限是七天；图片显示的是商品包装。")
+    result = _resolve_reply(
+        None,
+        avatar,
+        "退款期限是多少？",
+        [],
+        qa_pairs=[SimpleNamespace(question="退款期限是多少？", answer="七天", enabled=True)],
+        search_fn=Mock(return_value=[]),
+        model_client=model,
+        image_contexts=[{
+            "id": "attachment",
+            "filename": "product.jpg",
+            "category": "general_image",
+            "summary": "商品包装",
+            "extractedText": "",
+            "structuredData": {},
+            "warning": "",
+        }],
+    )
+
+    assert result["source"] == "qa"
+    system = model.call_args.kwargs["messages"][0]["content"]
+    assert "已确认标准答案" in system
+    assert "七天" in system
+    assert "商品包装" in system
+    assert "标准答题对中的事实优先级高于图片资料" in system
+
+
+def test_ready_image_context_never_returns_whole_image_access_denial():
+    avatar = SimpleNamespace(
+        id="avatar-vision",
+        name="测试分身",
+        description="产品顾问",
+        config={},
+    )
+    model = Mock(return_value="抱歉，我无法查看或识别图片，请重新上传。")
+    result = _resolve_reply(
+        None,
+        avatar,
+        "请看看这张图片",
+        [],
+        qa_pairs=[],
+        search_fn=Mock(return_value=[]),
+        model_client=model,
+        image_contexts=[{
+            "id": "attachment",
+            "filename": "report.jpg",
+            "category": "medical_document",
+            "summary": "一份耳鼻喉科门诊记录",
+            "extractedText": "主诉：咽痛三天",
+            "structuredData": {"key_facts": ["主诉为咽痛三天"]},
+            "warning": "请核对原始资料",
+        }],
+    )
+
+    assert result["source"] == "vision"
+    assert "一份耳鼻喉科门诊记录" in result["answer"]
+    assert "主诉为咽痛三天" in result["answer"]
+    assert "无法查看" not in result["answer"]
+    system = model.call_args.kwargs["messages"][0]["content"]
+    assert "当前会话图片已经成功读取" in system
+    assert "禁止声称无法查看" in system
+
+
+def test_image_denial_detector_allows_uncertain_field_in_ready_image():
+    assert _answer_denies_available_image("我无法查看这张图片") is True
+    assert _answer_denies_available_image("图片中患者姓名无法辨认，主诉为咽痛三天。") is False
+
+
+def test_attachment_context_does_not_expose_internal_fields():
+    row = SimpleNamespace(
+        id="attachment",
+        filename="case.jpg",
+        category="medical_document",
+        summary="门诊病例",
+        extracted_text="主诉：咳嗽",
+        structured_data={"medical": {"chief_complaint": "咳嗽"}},
+        warning="请核对原文",
+    )
+    context = _attachment_contexts([row])[0]
+    assert context["filename"] == "case.jpg"
+    assert "avatar_id" not in context
+    assert "vision_model" not in context

digital-avatar-app/backend/tests/test_chat_model_config.py

@@ -0,0 +1,87 @@
+from unittest.mock import Mock, patch
+
+import httpx
+
+from services.chat_model_config import (
+    clear_chat_model_config_cache,
+    get_chat_model_config,
+)
+
+
+def setup_function():
+    clear_chat_model_config_cache()
+
+
+def teardown_function():
+    clear_chat_model_config_cache()
+
+
+def test_admin_runtime_config_takes_priority(monkeypatch):
+    monkeypatch.setenv("CHAT_MODEL_CONFIG_URL", "http://config.test/runtime")
+    monkeypatch.setenv("AVATAR_MODEL_CONFIG_TOKEN", "shared-secret")
+    response = Mock()
+    response.raise_for_status.return_value = None
+    response.json.return_value = {
+        "data": {
+            "api_base_url": "https://model.test/v1/",
+            "api_key": "runtime-key",
+            "model": "avatar-model",
+            "vision_model": "avatar-vision-model",
+            "ocr_model": "avatar-ocr-model",
+            "max_tokens": 2048,
+            "timeout_seconds": 42,
+        }
+    }
+
+    with patch("services.chat_model_config.httpx.get", return_value=response) as request:
+        config = get_chat_model_config()
+
+    assert config.source == "admin"
+    assert config.api_base_url == "https://model.test/v1"
+    assert config.model == "avatar-model"
+    assert config.vision_model == "avatar-vision-model"
+    assert config.ocr_model == "avatar-ocr-model"
+    assert config.max_tokens == 2048
+    request.assert_called_once_with(
+        "http://config.test/runtime",
+        headers={"X-Avatar-Config-Token": "shared-secret"},
+        timeout=5.0,
+    )
+
+
+def test_runtime_failure_falls_back_to_environment(monkeypatch):
+    monkeypatch.setenv("CHAT_MODEL_CONFIG_URL", "http://config.test/runtime")
+    monkeypatch.setenv("AVATAR_MODEL_CONFIG_TOKEN", "shared-secret")
+    monkeypatch.setenv("CHAT_API_URL", "https://fallback.test/v1/")
+    monkeypatch.setenv("CHAT_API_KEY", "fallback-key")
+    monkeypatch.setenv("CHAT_MODEL", "fallback-model")
+    monkeypatch.setenv("VISION_MODEL", "fallback-vision")
+    monkeypatch.setenv("VISION_OCR_MODEL", "fallback-ocr")
+    monkeypatch.setenv("CHAT_MAX_OUTPUT_TOKENS", "1536")
+
+    request = httpx.Request("GET", "http://config.test/runtime")
+    with patch(
+        "services.chat_model_config.httpx.get",
+        side_effect=httpx.ConnectError("offline", request=request),
+    ):
+        config = get_chat_model_config()
+
+    assert config.source == "environment"
+    assert config.api_base_url == "https://fallback.test/v1"
+    assert config.api_key == "fallback-key"
+    assert config.model == "fallback-model"
+    assert config.vision_model == "fallback-vision"
+    assert config.ocr_model == "fallback-ocr"
+    assert config.max_tokens == 1536
+
+
+def test_runtime_config_is_cached(monkeypatch):
+    monkeypatch.setenv("CHAT_MODEL_CONFIG_URL", "")
+    monkeypatch.setenv("CHAT_MODEL", "first-model")
+    first = get_chat_model_config()
+    monkeypatch.setenv("CHAT_MODEL", "second-model")
+
+    second = get_chat_model_config()
+
+    assert first is second
+    assert second.model == "first-model"

digital-avatar-app/backend/tests/test_chat_orchestration.py

@@ -5,7 +5,16 @@ from unittest.mock import Mock
 from fastapi import HTTPException
 
 from models import Avatar, User
-from routers.chat import _build_prompt, _iter_text_chunks, _match_standard_qa, _public_avatar_payload, _require_owned_avatar, _resolve_reply
+from routers.chat import (
+    _build_prompt,
+    _iter_text_chunks,
+    _match_standard_qa,
+    _public_avatar_payload,
+    _qa_requires_language_adaptation,
+    _qa_requires_per_turn_rendering,
+    _require_owned_avatar,
+    _resolve_reply,
+)
 
 
 class ChatOrchestrationTests(unittest.TestCase):
@@ -50,6 +59,69 @@ class ChatOrchestrationTests(unittest.TestCase):
         self.assertEqual(result["answer"], "标准地址")
         fake_model.assert_not_called()
 
+    def test_cross_language_qa_is_faithfully_adapted_by_model(self):
+        fake_model = Mock(return_value="Our address is Test Road 1.")
+        fake_search = Mock(return_value=[])
+        result = _resolve_reply(
+            None,
+            self.avatar,
+            "Where is your office?",
+            [],
+            qa_pairs=[SimpleNamespace(question="Where is your office?", answer="地址是测试路1号。", enabled=True)],
+            search_fn=fake_search,
+            model_client=fake_model,
+        )
+
+        self.assertEqual(result["source"], "qa")
+        self.assertEqual(result["answer"], "Our address is Test Road 1.")
+        self.assertEqual(fake_model.call_args.kwargs["temperature"], 0.0)
+        system = fake_model.call_args.kwargs["messages"][0]["content"]
+        self.assertIn("已确认标准答案", system)
+        self.assertIn("地址是测试路1号", system)
+        self.assertIn("只使用该语言回答", system)
+        fake_search.assert_not_called()
+
+    def test_qa_language_adaptation_detects_common_writing_system_changes(self):
+        self.assertTrue(_qa_requires_language_adaptation("Hello", "你好"))
+        self.assertTrue(_qa_requires_language_adaptation("こんにちは", "你好"))
+        self.assertTrue(_qa_requires_language_adaptation("안녕하세요", "你好"))
+        self.assertFalse(_qa_requires_language_adaptation("你好", "您好"))
+
+    def test_conversation_qa_is_rendered_for_the_current_turn_language(self):
+        history = [SimpleNamespace(role="user", content="Please answer in English.")]
+        self.assertTrue(_qa_requires_per_turn_rendering("Quelle est votre adresse ?", "Our address is Test Road 1.", history))
+
+        fake_model = Mock(return_value="Notre adresse est Test Road 1.")
+        result = _resolve_reply(
+            None,
+            self.avatar,
+            "Quelle est votre adresse ?",
+            history,
+            qa_pairs=[SimpleNamespace(question="Quelle est votre adresse ?", answer="Our address is Test Road 1.", enabled=True)],
+            search_fn=Mock(),
+            model_client=fake_model,
+        )
+
+        self.assertEqual(result["source"], "qa")
+        self.assertEqual(result["answer"], "Notre adresse est Test Road 1.")
+        messages = fake_model.call_args.kwargs["messages"]
+        self.assertEqual(messages[-1], {"role": "user", "content": "Quelle est votre adresse ?"})
+        self.assertEqual(messages[-2]["role"], "system")
+        self.assertIn("本轮语言覆盖指令", messages[-2]["content"])
+        self.assertIn("不要沿用上一轮语言", messages[-2]["content"])
+
+    def test_latest_user_message_has_an_adjacent_language_override(self):
+        history = [
+            SimpleNamespace(role="user", content="请用中文回答"),
+            SimpleNamespace(role="assistant", content="好的，请问有什么可以帮你？"),
+        ]
+        messages = _build_prompt(self.avatar, history, "What can you help me with?", [])
+
+        self.assertEqual(messages[-1], {"role": "user", "content": "What can you help me with?"})
+        self.assertEqual(messages[-2]["role"], "system")
+        self.assertIn("最新用户消息", messages[-2]["content"])
+        self.assertIn("立即切换到相同语言", messages[-2]["content"])
+
     def test_conversational_paraphrase_matches_standard_qa(self):
         for question in ("请问一下，你们公司在哪里呀？", "请问去你们那边怎么走"):
             with self.subTest(question=question):
@@ -106,6 +178,9 @@ class ChatOrchestrationTests(unittest.TestCase):
         self.assertIn("像熟人之间微信聊天一样", messages[0]["content"])
         self.assertIn("不隶属于任何机构", messages[0]["content"])
         self.assertIn("不要连续输出空行", messages[0]["content"])
+        self.assertIn("回答语言规则", messages[0]["content"])
+        self.assertIn("当前最后一条用户消息", messages[0]["content"])
+        self.assertIn("历史消息", messages[0]["content"])
 
     def test_prompt_blocks_ungrounded_factual_answers(self):
         messages = _build_prompt(self.avatar, [], "聊聊国际新闻", [])
@@ -113,6 +188,8 @@ class ChatOrchestrationTests(unittest.TestCase):
         self.assertIn("没有检索到可靠资料", system)
         self.assertIn("不要凭通用知识", system)
         self.assertIn("不要提及知识库", system)
+        self.assertIn("不得推断服务对象", system)
+        self.assertIn("工作场所", system)
 
     def test_public_avatar_payload_excludes_internal_configuration(self):
         payload = _public_avatar_payload(self.avatar)

digital-avatar-app/backend/tests/test_embeddings.py

@@ -48,9 +48,12 @@ class RemoteEmbeddingTests(unittest.TestCase):
     def test_large_input_is_split_into_provider_safe_batches(self):
         texts = [f"chunk-{index}" for index in range(14)]
         batch_sizes = []
+        requested_urls = []
+        progress_updates = []
 
         def fake_urlopen(request, timeout):
             self.assertEqual(timeout, 30)
+            requested_urls.append(request.full_url)
             payload = json.loads(request.data.decode("utf-8"))
             batch_sizes.append(len(payload["input"]))
             return FakeResponse({
@@ -61,15 +64,29 @@ class RemoteEmbeddingTests(unittest.TestCase):
             })
 
         with patch.dict(os.environ, {
-            "EMBEDDING_API_URL": "https://embedding.example/v1/embeddings",
+            "EMBEDDING_API_URL": "https://embedding.example/v1",
             "EMBEDDING_API_KEY": "test-key",
             "EMBEDDING_MODEL": "text-embedding-v4",
             "EMBEDDING_BATCH_SIZE": "10",
         }), patch("embeddings.urllib.request.urlopen", side_effect=fake_urlopen):
-            result = embeddings.embed(texts)
+            result = embeddings.embed(
+                texts,
+                on_progress=lambda completed, total: progress_updates.append((completed, total)),
+            )
 
         self.assertEqual(batch_sizes, [10, 4])
+        self.assertEqual(requested_urls, [
+            "https://embedding.example/v1/embeddings",
+            "https://embedding.example/v1/embeddings",
+        ])
         self.assertEqual(result, [[float(index)] for index in range(14)])
+        self.assertEqual(progress_updates, [(10, 14), (14, 14)])
+
+    def test_full_embedding_endpoint_is_not_modified(self):
+        self.assertEqual(
+            embeddings._embedding_endpoint("https://embedding.example/v1/embeddings/"),
+            "https://embedding.example/v1/embeddings",
+        )
 
 
 if __name__ == "__main__":

digital-avatar-app/backend/tests/test_huihui_auth.py

@@ -1,6 +1,7 @@
 """Tests for preserving local avatar ownership when Huihui IDs change."""
 
 from datetime import datetime
+from unittest.mock import AsyncMock, patch
 
 import pytest
 from sqlalchemy import create_engine
@@ -9,7 +10,8 @@ from sqlalchemy.pool import StaticPool
 
 from database import Base
 from models import Avatar, TakeoverCursor, TakeoverMessage, TakeoverReplyTask, User
-from routers.huihui_auth import _issue_session
+from routers.huihui_auth import _issue_session, token_login
+from services.boxim_client import BoxIMError
 
 
 @pytest.fixture
@@ -135,3 +137,55 @@ def test_ambiguous_phone_matches_do_not_move_existing_avatars(db):
     assert db.query(User).count() == 3
     _assert_avatar_data_owner(db, first_avatar.id, "fat-1")
     _assert_avatar_data_owner(db, second_avatar.id, "fat-2")
+
+
+@pytest.mark.asyncio
+async def test_token_login_uses_huihui_user_id_and_keeps_upstream_token_server_side(db):
+    existing = User(
+        id="existing-local",
+        huihui_user_id="huihui-user-88",
+        app_token="existing-app-session",
+    )
+    db.add(existing)
+    db.commit()
+
+    client = AsyncMock()
+    client.exchange_access_token.return_value = {"accessToken": "boxim-token"}
+    client.get_self.return_value = {
+        "id": 998877,
+        "huihuiUserId": "huihui-user-88",
+        "nickName": "会会用户",
+        "headImage": "https://cdn.example/avatar.jpg",
+    }
+    with patch("routers.huihui_auth._cfg_ready", return_value=True), patch(
+        "routers.huihui_auth._create_boxim_client", return_value=client
+    ):
+        response = await token_login({"token": "production-huihui-token"}, db)
+
+    assert response["code"] == 200
+    assert response["data"]["token"] == "existing-app-session"
+    assert "token" not in response["data"]["huihui"]
+    user = db.query(User).one()
+    assert user.huihui_user_id == "huihui-user-88"
+    assert user.huihui_user_id != "998877"
+    assert user.huihui_token == "production-huihui-token"
+    assert user.nickname == "会会用户"
+    assert user.avatar_url == "https://cdn.example/avatar.jpg"
+    client.exchange_access_token.assert_awaited_once_with("production-huihui-token")
+    client.get_self.assert_awaited_once_with("boxim-token")
+
+
+@pytest.mark.asyncio
+async def test_token_login_rejects_expired_huihui_token_without_creating_user(db):
+    client = AsyncMock()
+    client.exchange_access_token.side_effect = BoxIMError(
+        "expired", auth_error=True
+    )
+    with patch("routers.huihui_auth._cfg_ready", return_value=True), patch(
+        "routers.huihui_auth._create_boxim_client", return_value=client
+    ):
+        response = await token_login({"token": "expired-token"}, db)
+
+    assert response["code"] == 401
+    assert response["message"] == "会会登录凭证无效或已过期"
+    assert db.query(User).count() == 0

digital-avatar-app/backend/tests/test_huihui_payment.py

@@ -0,0 +1,50 @@
+from unittest.mock import Mock, patch
+
+from services.huihui_payment import HuihuiPaymentClient
+
+
+def test_create_payment_uses_huihui_payment_v3_contract():
+    client = HuihuiPaymentClient({
+        "HUIHUI_PAYMENT_BASE_URL": "https://open.example/api/payment-v3",
+        "HUIHUI_APP_ID": "app-id",
+        "HUIHUI_ACCESS_ID": "access-id",
+        "HUIHUI_ACCESS_SECRET": "access-secret",
+    })
+    response = Mock()
+    response.status_code = 200
+    response.json.return_value = {
+        "code": 0,
+        "data": {"orderId": "provider-id", "status": "pending", "payMessage": "mock"},
+    }
+
+    with patch("services.huihui_payment.httpx.post", return_value=response) as post:
+        result = client.create_payment(
+            huihui_token="user-token",
+            huihui_user_id="user-id",
+            real_name="测试用户",
+            order_no="AV202608260001",
+            amount="10.00",
+            points_amount=2_000_000,
+            pay_type="WECHAT",
+            pay_way="APP",
+            callback_url="https://digital.example/api/token/payment/callback/secret",
+        )
+
+    assert result["orderId"] == "provider-id"
+    assert post.call_args.args[0] == "https://open.example/api/payment-v3/payment/pay"
+    assert post.call_args.kwargs["headers"] == {
+        "Authorization": "Bearer user-token",
+        "appId": "app-id",
+        "windowAppId": "app-id",
+    }
+    params = post.call_args.kwargs["params"]
+    assert params["appId"] == "app-id"
+    assert params["accessId"] == "access-id"
+    assert params["userId"] == "user-id"
+    assert params["signature"]
+    assert "accessSecret" not in params
+    body = post.call_args.kwargs["json"]
+    assert body["payType"] == "WECHAT"
+    assert body["payWay"] == "APP"
+    assert body["masterOrderAmt"] == "10.00"
+    assert body["payAmt"] == 10.0

digital-avatar-app/backend/tests/test_knowledge_storage.py

@@ -2,7 +2,16 @@ from pathlib import Path
 from types import SimpleNamespace
 from unittest.mock import patch
 
+from fastapi.testclient import TestClient
+
+from database import SessionLocal
+from main import app
+from models import Avatar, KnowledgeChunk, KnowledgeDoc, QAPair
 from routers.knowledge import _doc_payload
+from services.knowledge_vectorizer import knowledge_vectorizer
+
+
+client = TestClient(app)
 
 
 def test_doc_payload_reports_whether_the_persisted_file_exists(tmp_path: Path):
@@ -21,3 +30,347 @@ def test_doc_payload_reports_whether_the_persisted_file_exists(tmp_path: Path):
         assert _doc_payload(doc)["filePresent"] is False
         stored_file.write_text("knowledge", encoding="utf-8")
         assert _doc_payload(doc)["filePresent"] is True
+
+
+def test_upload_returns_before_background_vectorization(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
+            headers=context["owner_headers"],
+            files={"file": ("knowledge.md", b"# Knowledge\n\nTest content", "text/markdown")},
+        )
+
+    payload = response.json()["data"]
+    assert payload["status"] == "parsing"
+    assert payload["vectorized"] is False
+    assert payload["chunkCount"] == 0
+    enqueue.assert_called_once_with(payload["id"])
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == payload["id"]).one()
+        assert stored.status == "parsing"
+        assert db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).count() == 0
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_upload_rejects_oversize_file_before_queuing_indexing(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MAX_UPLOAD_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
+            headers=context["owner_headers"],
+            files={"file": ("oversize.md", b"12345", "text/markdown")},
+        )
+
+    payload = response.json()
+    assert payload["code"] == 400
+    assert payload["message"] == "文件不能超过 50MB"
+    enqueue.assert_not_called()
+    assert not list((tmp_path / context["avatar"].id).glob("*"))
+
+
+def test_multipart_upload_reassembles_file_before_queuing_indexing(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    content = b"0123456789"
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": len(content), "totalChunks": 3},
+        ).json()["data"]
+
+        for index, chunk in enumerate((content[:4], content[4:8], content[8:])):
+            response = client.post(
+                f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/{index}",
+                headers=context["owner_headers"],
+                files={"file": (f"chunk-{index}", chunk, "application/octet-stream")},
+            )
+            assert response.json()["code"] == 200
+
+        completed = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        ).json()["data"]
+
+    assert completed["status"] == "parsing"
+    assert completed["fileSize"] == len(content)
+    enqueue.assert_called_once_with(completed["id"])
+    stored_path = tmp_path / avatar_id / Path(completed["fileUrl"]).name
+    assert stored_path.read_bytes() == content
+    assert not (tmp_path / ".multipart" / avatar_id / created["uploadId"]).exists()
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == completed["id"]).one()
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_multipart_upload_rejects_incomplete_parts(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": 6, "totalChunks": 2},
+        ).json()["data"]
+        client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/0",
+            headers=context["owner_headers"],
+            files={"file": ("chunk-0", b"0123", "application/octet-stream")},
+        )
+        response = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        )
+
+    assert response.json()["code"] == 400
+    assert response.json()["message"] == "文件分片尚未上传完整"
+    enqueue.assert_not_called()
+
+
+def test_background_vectorizer_commits_ready_document_and_chunks_together(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue"),
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
+            headers=context["owner_headers"],
+            files={"file": ("knowledge.md", b"# Knowledge\n\nTest content", "text/markdown")},
+        )
+
+    payload = response.json()["data"]
+    assert payload["status"] == "parsing"
+    with (
+        patch("services.knowledge_vectorizer.UPLOAD_DIR", str(tmp_path)),
+        patch("services.knowledge_vectorizer.embeddings.embed", return_value=[[1.0, 0.0]]),
+    ):
+        knowledge_vectorizer.vectorize_document(payload["id"])
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == payload["id"]).one()
+        assert stored.status == "ready"
+        assert stored.vectorized is True
+        assert stored.chunk_count == 1
+        assert stored.index_stage == "ready"
+        assert stored.index_progress == 100
+        assert db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).count() == 1
+        db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).delete()
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_background_vectorizer_keeps_failure_reason_for_retry(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue"),
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
+            headers=context["owner_headers"],
+            files={"file": ("knowledge.md", b"# Knowledge\n\nTest content", "text/markdown")},
+        )
+
+    payload = response.json()["data"]
+    with (
+        patch("services.knowledge_vectorizer.UPLOAD_DIR", str(tmp_path)),
+        patch("services.knowledge_vectorizer.embeddings.embed", side_effect=RuntimeError("provider unavailable")),
+    ):
+        knowledge_vectorizer.vectorize_document(payload["id"])
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == payload["id"]).one()
+        assert stored.status == "failed"
+        assert stored.error_message == "provider unavailable"
+        assert db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).count() == 0
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_retry_queues_a_failed_document_again(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    document_id = f"retry-doc-{context['suffix']}"
+    avatar_dir = tmp_path / context["avatar"].id
+    avatar_dir.mkdir()
+    (avatar_dir / "retry.md").write_text("retry content", encoding="utf-8")
+    db = SessionLocal()
+    try:
+        db.add(
+            KnowledgeDoc(
+                id=document_id,
+                avatar_id=context["avatar"].id,
+                filename="retry.md",
+                file_type="md",
+                file_url=f"/api/files/{context['avatar'].id}/retry.md",
+                status="failed",
+                error_message="provider unavailable",
+            )
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs/{document_id}/retry",
+            headers=context["owner_headers"],
+        )
+
+    payload = response.json()["data"]
+    assert payload["status"] == "parsing"
+    assert payload["errorMessage"] == ""
+    enqueue.assert_called_once_with(document_id)
+    db = SessionLocal()
+    try:
+        db.query(KnowledgeDoc).filter(KnowledgeDoc.id == document_id).delete()
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_each_avatar_has_an_independent_document_and_qa_scope(authorization_context):
+    context = authorization_context
+    first_avatar_id = context["avatar"].id
+    second_avatar_id = f"knowledge-second-{context['suffix']}"
+    first_doc_id = f"knowledge-first-doc-{context['suffix']}"
+    second_doc_id = f"knowledge-second-doc-{context['suffix']}"
+    first_qa_id = f"knowledge-first-qa-{context['suffix']}"
+    second_qa_id = f"knowledge-second-qa-{context['suffix']}"
+
+    db = SessionLocal()
+    try:
+        db.add_all(
+            [
+                Avatar(
+                    id=second_avatar_id,
+                    owner_id=context["owner"].huihui_user_id,
+                    name="独立知识库分身",
+                    status="active",
+                    config={},
+                ),
+                KnowledgeDoc(
+                    id=first_doc_id,
+                    avatar_id=first_avatar_id,
+                    filename="first.md",
+                    status="ready",
+                    vectorized=True,
+                ),
+                KnowledgeDoc(
+                    id=second_doc_id,
+                    avatar_id=second_avatar_id,
+                    filename="second.md",
+                    status="ready",
+                    vectorized=True,
+                ),
+                QAPair(
+                    id=first_qa_id,
+                    avatar_id=first_avatar_id,
+                    question="第一个分身问题",
+                    answer="第一个分身答案",
+                ),
+                QAPair(
+                    id=second_qa_id,
+                    avatar_id=second_avatar_id,
+                    question="第二个分身问题",
+                    answer="第二个分身答案",
+                ),
+            ]
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    try:
+        first_docs = client.get(
+            f"/api/avatar/{first_avatar_id}/knowledge/docs",
+            headers=context["owner_headers"],
+        ).json()["data"]
+        second_docs = client.get(
+            f"/api/avatar/{second_avatar_id}/knowledge/docs",
+            headers=context["owner_headers"],
+        ).json()["data"]
+        first_qa = client.get(
+            f"/api/avatar/{first_avatar_id}/knowledge/qa",
+            headers=context["owner_headers"],
+        ).json()["data"]
+        second_qa = client.get(
+            f"/api/avatar/{second_avatar_id}/knowledge/qa",
+            headers=context["owner_headers"],
+        ).json()["data"]
+
+        assert [item["id"] for item in first_docs if item["id"] == first_doc_id] == [first_doc_id]
+        assert second_doc_id not in {item["id"] for item in first_docs}
+        assert [item["id"] for item in second_docs] == [second_doc_id]
+        assert first_qa_id in {item["id"] for item in first_qa}
+        assert second_qa_id not in {item["id"] for item in first_qa}
+        assert [item["id"] for item in second_qa] == [second_qa_id]
+    finally:
+        db = SessionLocal()
+        try:
+            db.query(QAPair).filter(QAPair.id.in_([first_qa_id, second_qa_id])).delete(
+                synchronize_session=False
+            )
+            db.query(KnowledgeDoc).filter(
+                KnowledgeDoc.id.in_([first_doc_id, second_doc_id])
+            ).delete(synchronize_session=False)
+            db.query(Avatar).filter(Avatar.id == second_avatar_id).delete()
+            db.commit()
+        finally:
+            db.close()

digital-avatar-app/backend/tests/test_takeover_model.py

@@ -13,7 +13,7 @@ def test_authorization_takeover_fields():
         assert hasattr(auth, 'takeover_delay_seconds')
         assert auth.takeover_enabled == False
         assert auth.takeover_mode == 'immediate'
-        assert auth.takeover_delay_seconds == 30
+        assert auth.takeover_delay_seconds == 180
     finally:
         db.close()
 

digital-avatar-app/backend/tests/test_takeover_scheduler.py

@@ -20,8 +20,9 @@ def test_scheduler_uses_boxim_and_restart_safe_service(
 ):
     import main
 
+    maintenance_scheduler = MagicMock()
     scheduler = MagicMock()
-    mock_scheduler_class.return_value = scheduler
+    mock_scheduler_class.side_effect = [maintenance_scheduler, scheduler]
     boxim = MagicMock()
     mock_boxim_class.return_value = boxim
     takeover = MagicMock()
@@ -45,7 +46,16 @@ def test_scheduler_uses_boxim_and_restart_safe_service(
     config = mock_boxim_class.call_args.args[0]
     assert config["HUIHUI_PLATFORM_BASE_URL"] == "https://open.example/api"
     assert config["BOXIM_API_BASE_URL"] == "https://im.example/api"
-    mock_takeover_class.assert_called_once_with(main.SessionLocal, boxim)
+    mock_takeover_class.assert_called_once_with(
+        main.SessionLocal,
+        boxim,
+        poll_concurrency=8,
+        max_message_age_seconds=600,
+    )
+
+    maintenance_scheduler.add_job.assert_called_once()
+    assert maintenance_scheduler.add_job.call_args.kwargs["id"] == "chat_attachment_cleanup"
+    maintenance_scheduler.start.assert_called_once_with()
 
     assert scheduler.add_job.call_count == 2
     poll_call, process_call = scheduler.add_job.call_args_list
@@ -62,6 +72,7 @@ def test_scheduler_uses_boxim_and_restart_safe_service(
     scheduler.start.assert_called_once_with()
 
     main.takeover_scheduler = None
+    main.maintenance_scheduler = None
 
 
 @patch("main.AsyncIOScheduler")
@@ -73,6 +84,7 @@ def test_scheduler_failure_does_not_stop_the_api(mock_scheduler_class):
         main.on_startup()
 
     assert main.takeover_scheduler is None
+    assert main.maintenance_scheduler is None
 
 
 def test_shutdown_stops_only_the_scheduler():
@@ -80,9 +92,14 @@ def test_shutdown_stops_only_the_scheduler():
 
     scheduler = MagicMock()
     scheduler.running = True
+    maintenance_scheduler = MagicMock()
+    maintenance_scheduler.running = True
     main.takeover_scheduler = scheduler
+    main.maintenance_scheduler = maintenance_scheduler
 
     main.on_shutdown()
 
     scheduler.shutdown.assert_called_once_with(wait=False)
+    maintenance_scheduler.shutdown.assert_called_once_with(wait=False)
     assert main.takeover_scheduler is None
+    assert main.maintenance_scheduler is None

digital-avatar-app/backend/tests/test_takeover_service.py

@@ -1,5 +1,7 @@
 """End-to-end service tests for BOXIM takeover timing and human priority."""
 
+import asyncio
+import json
 from datetime import datetime, timedelta, timezone
 from threading import Barrier
 from unittest.mock import AsyncMock, patch
@@ -7,12 +9,17 @@ from unittest.mock import AsyncMock, patch
 import pytest
 from sqlalchemy import create_engine
 from sqlalchemy.orm import sessionmaker
-from sqlalchemy.pool import StaticPool
 
 from database import Base
-from models import Avatar, TakeoverCursor, TakeoverMessage, TakeoverReplyTask, User
+from models import Avatar, ChatAttachment, TakeoverCursor, TakeoverMessage, TakeoverReplyTask, User
 from services.boxim_client import BoxIMError
-from services.takeover_service import TakeoverService, _plain_text_reply
+from services.boxim_image_service import DownloadedBoxIMImage
+from services.takeover_service import (
+    AVATAR_LOCAL_ID_PREFIX,
+    TakeoverService,
+    _avatar_local_id,
+    _plain_text_reply,
+)
 
 
 class Clock:
@@ -58,12 +65,54 @@ class FakeBoxIM:
         return {"id": 900 + len(self.sent), "localId": int(local_id)}
 
 
+class ConcurrentPollingBoxIM(FakeBoxIM):
+    def __init__(self):
+        super().__init__()
+        self.active_polls = 0
+        self.peak_active_polls = 0
+
+    async def exchange_access_token(self, huihui_token):
+        return {"accessToken": huihui_token, "accessTokenExpiresIn": 3600}
+
+    async def get_self(self, access_token):
+        return {"id": 100 if access_token == "prod-huihui-token" else 101}
+
+    async def fetch_private_messages(self, access_token, min_id="0"):
+        self.active_polls += 1
+        self.peak_active_polls = max(self.peak_active_polls, self.active_polls)
+        await asyncio.sleep(0.05)
+        self.active_polls -= 1
+        return []
+
+
+class ConcurrentMessagePollingBoxIM(ConcurrentPollingBoxIM):
+    async def fetch_private_messages(self, access_token, min_id="0"):
+        await super().fetch_private_messages(access_token, min_id)
+        owner_id = 100 if access_token == "prod-huihui-token" else 101
+        return [
+            {
+                "id": owner_id,
+                "localId": owner_id,
+                "sendId": owner_id + 100,
+                "recvId": owner_id,
+                "sendTime": 1_700_000_000_000,
+                "type": 0,
+                "content": "并发写入测试",
+            }
+        ]
+
+    async def mark_private_messages_read(self, access_token, friend_id, message_id):
+        await asyncio.sleep(0.05)
+        self.read_receipts.append(
+            {"friendId": str(friend_id), "messageId": str(message_id)}
+        )
+
+
 @pytest.fixture
-def service_context():
+def service_context(tmp_path):
     engine = create_engine(
-        "sqlite://",
+        f"sqlite:///{tmp_path / 'takeover.db'}",
         connect_args={"check_same_thread": False},
-        poolclass=StaticPool,
     )
     session_factory = sessionmaker(bind=engine, autoflush=False, expire_on_commit=False)
     Base.metadata.create_all(engine)
@@ -79,7 +128,10 @@ def service_context():
         owner_id=user.huihui_user_id,
         name="分身",
         status="active",
-        config={"authorizationPermissions": ["chat", "takeover"]},
+        config={
+            "authorizationPermissions": ["chat", "takeover"],
+            "takeoverReplyDelaySeconds": 3,
+        },
     )
     db.add_all([user, avatar])
     db.commit()
@@ -122,7 +174,6 @@ async def test_incoming_message_is_prepared_then_sent_at_three_seconds(service_c
         {"id": 11, "localId": 2, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "你好"}
     )
 
-    with patch("routers.chat._resolve_reply", return_value={"answer": "**你好**\n\n很高兴见到你"}):
     await service.poll_and_process_messages()
     assert boxim.sent == []
     assert boxim.read_receipts == [{"friendId": "200", "messageId": "11"}]
@@ -132,6 +183,7 @@ async def test_incoming_message_is_prepared_then_sent_at_three_seconds(service_c
     assert boxim.sent == []
 
     clock.advance(1)
+    with patch("routers.chat._resolve_reply", return_value={"answer": "**你好**\n\n很高兴见到你"}):
         await service.poll_and_process_messages()
     assert boxim.sent == [{"peerId": "200", "content": "你好\n很高兴见到你", "localId": boxim.sent[0]["localId"]}]
 
@@ -144,6 +196,453 @@ async def test_incoming_message_is_prepared_then_sent_at_three_seconds(service_c
         db.close()
 
 
+@pytest.mark.asyncio
+async def test_incoming_image_is_analyzed_and_used_in_takeover_reply(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {
+            "id": 111,
+            "localId": 111,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 1,
+            "content": json.dumps(
+                {
+                    "originUrl": "https://cdn.example/case.png",
+                    "thumbUrl": "https://cdn.example/case-thumb.png",
+                }
+            ),
+        }
+    )
+
+    await service.poll_and_process_messages()
+    db = session_factory()
+    try:
+        scheduled = db.query(TakeoverReplyTask).filter_by(trigger_message_id="111").one()
+        assert scheduled.status == "pending"
+        assert scheduled.prompt == "请看看这张图片。"
+    finally:
+        db.close()
+
+    clock.advance(3)
+
+    def analyze(db, avatar, content, **kwargs):
+        assert content == b"image-content"
+        attachment = ChatAttachment(
+            avatar_id=avatar.id,
+            uploader_kind=kwargs["uploader_kind"],
+            filename=kwargs["filename"],
+            mime_type="image/jpeg",
+            file_size=len(content),
+            status="ready",
+            category="medical_document",
+            summary="一张门诊病例",
+            extracted_text="主诉：咳嗽三天",
+            structured_data={"medical": {"chief_complaint": "咳嗽三天"}},
+            warning="请核对原始资料",
+            expires_at=clock.now() + timedelta(hours=24),
+        )
+        db.add(attachment)
+        db.commit()
+        db.refresh(attachment)
+        return attachment
+
+    downloaded = DownloadedBoxIMImage(
+        content=b"image-content",
+        filename="case.png",
+        mime_type="image/png",
+        source_url="https://cdn.example/case.png",
+    )
+    with (
+        patch("services.takeover_service.download_boxim_image", return_value=downloaded),
+        patch("routers.chat._analyze_image_bytes", side_effect=analyze) as analyzer,
+        patch("routers.chat._resolve_reply", return_value={"answer": "这份资料里写的是咳嗽三天。"}) as resolver,
+    ):
+        await service.poll_and_process_messages()
+
+    analyzer.assert_called_once()
+    assert resolver.call_args.args[2] == "请看看这张图片。"
+    image_contexts = resolver.call_args.kwargs["image_contexts"]
+    assert image_contexts[0]["summary"] == "一张门诊病例"
+    assert image_contexts[0]["extractedText"] == "主诉：咳嗽三天"
+    assert [item["content"] for item in boxim.sent] == ["这份资料里写的是咳嗽三天。"]
+
+    db = session_factory()
+    try:
+        event = db.query(TakeoverMessage).filter_by(boxim_message_id="111").one()
+        task = db.query(TakeoverReplyTask).filter_by(trigger_message_id="111").one()
+        assert event.attachment_id
+        assert db.get(ChatAttachment, event.attachment_id).uploader_kind == "boxim"
+        assert task.status == "sent"
+        with patch(
+            "services.takeover_service.download_boxim_image",
+            side_effect=AssertionError("cached image must not be downloaded again"),
+        ):
+            cached = service._takeover_image_attachment(db, db.get(Avatar, "avatar-1"), event)
+        assert cached.id == event.attachment_id
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_followup_text_recovers_recent_image_recorded_without_task(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    image_message = {
+        "id": 113,
+        "localId": 113,
+        "sendId": 200,
+        "recvId": 100,
+        "sendTime": clock.millis(),
+        "type": 1,
+        "content": json.dumps(
+            {
+                "originUrl": "https://cdn.example/case.png",
+                "thumbUrl": "https://cdn.example/case-thumb.png",
+            }
+        ),
+    }
+
+    db = session_factory()
+    try:
+        avatar = db.get(Avatar, "avatar-1")
+        service._record_message(db, avatar, "100", image_message, schedule_reply=False)
+        cursor = db.query(TakeoverCursor).one()
+        cursor.last_message_id = "113"
+        db.commit()
+    finally:
+        db.close()
+
+    clock.advance(60)
+    boxim.messages.extend(
+        [
+            image_message,
+            {
+                "id": 114,
+                "localId": 114,
+                "sendId": 200,
+                "recvId": 100,
+                "sendTime": clock.millis(),
+                "type": 0,
+                "content": "请帮我看看这张图",
+            },
+        ]
+    )
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        task = db.query(TakeoverReplyTask).filter_by(trigger_message_id="114").one()
+        assert task.source_message_ids == ["113", "114"]
+        assert task.prompt == "请看看这张图片。\n请帮我看看这张图"
+    finally:
+        db.close()
+
+    clock.advance(1)
+    boxim.messages.append(
+        {
+            "id": 115,
+            "localId": 115,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 0,
+            "content": "图里写了什么",
+        }
+    )
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        latest = db.query(TakeoverReplyTask).filter_by(trigger_message_id="115").one()
+        assert latest.source_message_ids == ["113", "114", "115"]
+        assert latest.source_message_ids.count("113") == 1
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_explicit_followup_reuses_handled_image_within_two_days(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    image_message = {
+        "id": 116,
+        "localId": 116,
+        "sendId": 200,
+        "recvId": 100,
+        "sendTime": clock.millis(),
+        "type": 1,
+        "content": json.dumps({"originUrl": "https://cdn.example/handled-case.png"}),
+    }
+    boxim.messages.append(image_message)
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        image_task = db.query(TakeoverReplyTask).filter_by(trigger_message_id="116").one()
+        image_task.status = "sent"
+        image_task.sent_at = clock.now()
+        db.commit()
+    finally:
+        db.close()
+
+    clock.advance(47 * 60 * 60)
+    boxim.messages.append(
+        {
+            "id": 117,
+            "localId": 117,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 0,
+            "content": "重新看一下刚才那张病例图片",
+        }
+    )
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        task = db.query(TakeoverReplyTask).filter_by(trigger_message_id="117").one()
+        assert task.source_message_ids == ["116", "117"]
+        assert task.prompt == "请看看这张图片。\n重新看一下刚才那张病例图片"
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_invalid_image_message_is_recorded_but_not_scheduled(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {
+            "id": 112,
+            "localId": 112,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 1,
+            "content": json.dumps({"width": 100, "height": 100}),
+        }
+    )
+
+    await service.poll_and_process_messages()
+
+    db = session_factory()
+    try:
+        assert db.query(TakeoverMessage).filter_by(boxim_message_id="112").one()
+        assert db.query(TakeoverReplyTask).count() == 0
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_default_reply_delay_is_three_minutes(service_context):
+    session_factory, service, boxim, clock = service_context
+    db = session_factory()
+    try:
+        avatar = db.query(Avatar).one()
+        avatar.config = {"authorizationPermissions": ["chat", "takeover"]}
+        db.commit()
+    finally:
+        db.close()
+
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {"id": 12, "localId": 12, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "三分钟后回复"}
+    )
+    await service.poll_and_process_messages()
+
+    db = session_factory()
+    try:
+        task = db.query(TakeoverReplyTask).one()
+        assert task.scheduled_at == clock.now() + timedelta(seconds=180)
+    finally:
+        db.close()
+
+    clock.advance(179)
+    await service.process_reply_tasks()
+    assert boxim.sent == []
+    clock.advance(1)
+    with patch("routers.chat._resolve_reply", return_value={"answer": "好的"}):
+        await service.poll_and_process_messages()
+    assert [item["content"] for item in boxim.sent] == ["好的"]
+
+
+@pytest.mark.asyncio
+async def test_multiple_avatar_owners_are_polled_concurrently(service_context):
+    session_factory, _service, _boxim, clock = service_context
+    db = session_factory()
+    try:
+        db.add_all(
+            [
+                User(
+                    id="owner-local-2",
+                    huihui_user_id="owner-huihui-2",
+                    huihui_token="prod-huihui-token-2",
+                    app_token="app-token-2",
+                ),
+                Avatar(
+                    id="avatar-2",
+                    owner_id="owner-huihui-2",
+                    name="分身二",
+                    status="active",
+                    config={"authorizationPermissions": ["chat", "takeover"]},
+                ),
+            ]
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    boxim = ConcurrentMessagePollingBoxIM()
+    service = TakeoverService(
+        session_factory,
+        boxim,
+        poll_concurrency=2,
+        now=clock.now,
+    )
+
+    await service.poll_messages()
+
+    assert boxim.peak_active_polls == 2
+    db = session_factory()
+    try:
+        assert db.query(TakeoverCursor).filter(TakeoverCursor.initialized.is_(True)).count() == 2
+        assert db.query(TakeoverMessage).count() == 2
+        assert len(boxim.read_receipts) == 2
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_delayed_poll_still_schedules_recent_message(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_messages()
+    delayed_send_time = int(
+        (clock.value - timedelta(seconds=150)).replace(tzinfo=timezone.utc).timestamp()
+        * 1000
+    )
+    boxim.messages.append(
+        {
+            "id": 13,
+            "localId": 13,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": delayed_send_time,
+            "type": 0,
+            "content": "排队后仍需回复",
+        }
+    )
+
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        task = db.query(TakeoverReplyTask).one()
+        assert task.status == "pending"
+        assert task.scheduled_at == clock.now()
+    finally:
+        db.close()
+
+    with patch("routers.chat._resolve_reply", return_value={"answer": "已经收到"}):
+        await service.process_reply_tasks()
+    assert [item["content"] for item in boxim.sent] == ["已经收到"]
+
+
+@pytest.mark.asyncio
+async def test_avatar_origin_message_never_schedules_a_reply(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    local_id = _avatar_local_id("peer-owner", "peer-trigger")
+    boxim.messages.append(
+        {"id": 15, "localId": local_id, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "另一端分身回复"}
+    )
+
+    with patch("routers.chat._resolve_reply") as resolver:
+        await service.poll_and_process_messages()
+
+    resolver.assert_not_called()
+    db = session_factory()
+    try:
+        event = db.query(TakeoverMessage).filter(TakeoverMessage.boxim_message_id == "15").one()
+        assert event.is_avatar is True
+        assert db.query(TakeoverReplyTask).count() == 0
+    finally:
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_peer_avatar_messages_are_excluded_from_later_human_context(service_context):
+    _session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {
+            "id": 16,
+            "localId": _avatar_local_id("peer-owner", "peer-trigger"),
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 0,
+            "content": "分身生成的夸张长文",
+        }
+    )
+    await service.poll_and_process_messages()
+
+    clock.advance(1)
+    boxim.messages.append(
+        {
+            "id": 17,
+            "localId": 17,
+            "sendId": 200,
+            "recvId": 100,
+            "sendTime": clock.millis(),
+            "type": 0,
+            "content": "真人的新问题",
+        }
+    )
+    await service.poll_and_process_messages()
+    clock.advance(3)
+    with patch("routers.chat._resolve_reply", return_value={"answer": "正常回复"}) as resolver:
+        await service.poll_and_process_messages()
+
+    assert resolver.call_args.args[3] == []
+
+
+@pytest.mark.asyncio
+async def test_owner_message_pauses_future_takeover_for_ten_minutes(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {"id": 18, "localId": 18, "sendId": 100, "recvId": 200, "sendTime": clock.millis(), "type": 0, "content": "我先来回复"}
+    )
+    await service.poll_and_process_messages()
+
+    clock.advance(30)
+    boxim.messages.append(
+        {"id": 19, "localId": 19, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "收到"}
+    )
+    with patch("routers.chat._resolve_reply") as resolver:
+        await service.poll_and_process_messages()
+
+    resolver.assert_not_called()
+    db = session_factory()
+    try:
+        assert db.query(TakeoverReplyTask).count() == 0
+    finally:
+        db.close()
+
+
+def test_avatar_local_id_is_deterministic_and_self_describing():
+    first = _avatar_local_id("owner", "message-1")
+    assert first == _avatar_local_id("owner", "message-1")
+    assert first != _avatar_local_id("owner", "message-2")
+    assert first.startswith(AVATAR_LOCAL_ID_PREFIX)
+    assert len(first) == 18
+    assert first.isdigit()
+
+
 @pytest.mark.asyncio
 async def test_different_contacts_generate_without_blocking_each_other(service_context):
     session_factory, service, boxim, clock = service_context
@@ -156,15 +655,15 @@ async def test_different_contacts_generate_without_blocking_each_other(service_c
     )
     both_generating = Barrier(2, timeout=2)
 
-    def resolve(_db, _avatar, prompt, _history):
+    def resolve(_db, _avatar, prompt, _history, **_kwargs):
         both_generating.wait()
         return {"answer": f"回复{prompt[-1]}"}
 
-    with patch("routers.chat._resolve_reply", side_effect=resolve):
     await service.poll_and_process_messages()
 
     clock.advance(3)
-    await service.process_reply_tasks()
+    with patch("routers.chat._resolve_reply", side_effect=resolve):
+        await service.poll_and_process_messages()
     assert {(item["peerId"], item["content"]) for item in boxim.sent} == {
         ("200", "回复甲"),
         ("300", "回复乙"),
@@ -239,6 +738,34 @@ async def test_owner_message_cancels_pending_reply(service_context):
         db.close()
 
 
+@pytest.mark.asyncio
+async def test_owner_message_in_final_second_wins_before_generation(service_context):
+    session_factory, service, boxim, clock = service_context
+    await service.poll_and_process_messages()
+    boxim.messages.append(
+        {"id": 23, "localId": 23, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "还在吗"}
+    )
+    await service.poll_and_process_messages()
+
+    clock.advance(2)
+    boxim.messages.append(
+        {"id": 24, "localId": 24, "sendId": 100, "recvId": 200, "sendTime": clock.millis(), "type": 0, "content": "我来处理"}
+    )
+    clock.advance(1)
+    with patch("routers.chat._resolve_reply") as resolver:
+        await service.poll_and_process_messages()
+
+    resolver.assert_not_called()
+    assert boxim.sent == []
+    db = session_factory()
+    try:
+        task = db.query(TakeoverReplyTask).filter(TakeoverReplyTask.trigger_message_id == "23").one()
+        assert task.status == "cancelled"
+        assert task.cancel_reason == "owner_replied"
+    finally:
+        db.close()
+
+
 @pytest.mark.asyncio
 async def test_quick_successive_messages_are_coalesced_into_one_reply(service_context):
     session_factory, service, boxim, clock = service_context
@@ -246,19 +773,18 @@ async def test_quick_successive_messages_are_coalesced_into_one_reply(service_co
     boxim.messages.append(
         {"id": 31, "localId": 5, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "第一句"}
     )
-    with patch("routers.chat._resolve_reply", return_value={"answer": "第一版"}):
     await service.poll_and_process_messages()
 
     clock.advance(1)
     boxim.messages.append(
         {"id": 32, "localId": 6, "sendId": 200, "recvId": 100, "sendTime": clock.millis(), "type": 0, "content": "第二句"}
     )
+    await service.poll_and_process_messages()
+
+    clock.advance(3)
     with patch("routers.chat._resolve_reply", return_value={"answer": "合并回复"}) as resolver:
         await service.poll_and_process_messages()
     assert resolver.call_args.args[2] == "第一句\n第二句"
-
-    clock.advance(3)
-    await service.poll_and_process_messages()
     assert [item["content"] for item in boxim.sent] == ["合并回复"]
 
     db = session_factory()
@@ -293,5 +819,41 @@ async def test_connection_failure_disables_takeover_and_stops_retrying(service_c
     boxim.exchange_access_token.assert_awaited_once_with("prod-huihui-token")
 
 
+@pytest.mark.asyncio
+async def test_transient_connection_failure_keeps_takeover_and_recovers(service_context):
+    session_factory, service, boxim, _ = service_context
+    boxim.exchange_access_token = AsyncMock(
+        side_effect=[
+            BoxIMError("连接超时"),
+            {"accessToken": "box-token", "accessTokenExpiresIn": 3600},
+        ]
+    )
+
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        avatar = db.query(Avatar).one()
+        cursor = db.query(TakeoverCursor).one()
+        assert "takeover" in avatar.config["authorizationPermissions"]
+        assert cursor.initialized is False
+        assert "暂时连接失败" in cursor.last_error
+    finally:
+        db.close()
+
+    await service.poll_messages()
+
+    db = session_factory()
+    try:
+        avatar = db.query(Avatar).one()
+        cursor = db.query(TakeoverCursor).one()
+        assert "takeover" in avatar.config["authorizationPermissions"]
+        assert cursor.initialized is True
+        assert cursor.last_error == ""
+    finally:
+        db.close()
+    assert boxim.exchange_access_token.await_count == 2
+
+
 def test_plain_text_reply_removes_markdown_and_empty_lines():
     assert _plain_text_reply("## 建议\n\n**不能自行用药**\n`必要时就医`") == "建议\n不能自行用药\n必要时就医"

digital-avatar-app/backend/tests/test_token_billing.py

@@ -0,0 +1,430 @@
+import uuid
+import os
+import hashlib
+import hmac
+from concurrent.futures import ThreadPoolExecutor
+from threading import Barrier
+from unittest.mock import Mock, patch
+
+import pytest
+from fastapi.testclient import TestClient
+
+from database import SessionLocal
+from main import app, seed
+from models import Avatar, TokenAccount, TokenPaymentOrder, TokenPlan, TokenUsage, User
+from routers.chat import _resolve_reply, _stream_reply
+from services.token_billing import (
+    DEFAULT_TOKEN_GRANT,
+    InsufficientTokensError,
+    get_or_create_account,
+    release_reservation,
+    reserve_avatar_tokens,
+    settle_reservation,
+)
+
+client = TestClient(app)
+
+
+def _enable_huihui_payment_login(context):
+    db = SessionLocal()
+    try:
+        user = db.query(User).filter(User.id == context["owner"].id).one()
+        user.huihui_token = f"huihui-payment-{context['suffix']}"
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_balance_is_user_scoped_and_defaults_to_one_million(authorization_context):
+    context = authorization_context
+    owner = client.get("/api/token/balance", headers=context["owner_headers"])
+    other = client.get("/api/token/balance", headers=context["other_headers"])
+
+    assert owner.status_code == 200
+    assert owner.json()["data"] == {
+        "balance": DEFAULT_TOKEN_GRANT,
+        "totalGranted": DEFAULT_TOKEN_GRANT,
+        "totalConsumed": 0,
+    }
+    assert other.json()["data"]["balance"] == DEFAULT_TOKEN_GRANT
+    assert client.get("/api/token/balance").status_code == 401
+
+
+def test_seed_synchronizes_requested_recharge_plans():
+    seed()
+    db = SessionLocal()
+    try:
+        plans = db.query(TokenPlan).order_by(TokenPlan.price.asc()).all()
+        assert [(plan.price, plan.amount) for plan in plans] == [
+            (10, 2_000_000),
+            (100, 20_000_000),
+            (1000, 250_000_000),
+            (10000, 2_500_000_000),
+        ]
+    finally:
+        db.close()
+
+
+def test_charge_creates_huihui_order_without_early_points(authorization_context):
+    context = authorization_context
+    _enable_huihui_payment_login(context)
+    payment_client = Mock()
+    payment_client.create_payment.return_value = {
+        "orderId": "huihui-payment-id",
+        "orderNo": "huihui-payment-no",
+        "payMessage": {"mock": "payment-params"},
+        "payType": "WECHAT",
+        "paySubType": "APP",
+        "status": "pending",
+    }
+    env = {
+        "HUIHUI_PAYMENT_CALLBACK_BASE_URL": "https://digital.example",
+        "HUIHUI_PAYMENT_CALLBACK_SECRET": "test-callback-secret-123456",
+    }
+    with patch.dict(os.environ, env), patch("routers.tokens._payment_client", return_value=payment_client):
+        response = client.post(
+            "/api/token/charge",
+            headers=context["owner_headers"],
+            json={"planId": "1", "paymentMethod": "wechat", "payScene": "APP"},
+        )
+
+    assert response.status_code == 200
+    result = response.json()["data"]
+    assert result["status"] == "pending"
+    assert result["payType"] == "WECHAT"
+    assert result["payWay"] == "APP"
+    assert result["balance"] == DEFAULT_TOKEN_GRANT
+    assert payment_client.create_payment.call_args.kwargs["amount"] == "10.00"
+    callback_url = payment_client.create_payment.call_args.kwargs["callback_url"]
+    assert callback_url.startswith("https://digital.example/api/token/payment/callback/AV")
+    assert "test-callback-secret-123456" not in callback_url
+
+
+def test_success_callback_credits_once_and_status_is_user_scoped(authorization_context):
+    context = authorization_context
+    _enable_huihui_payment_login(context)
+    payment_client = Mock()
+    payment_client.create_payment.return_value = {
+        "orderId": "huihui-payment-id",
+        "orderNo": "huihui-payment-no",
+        "payMessage": "payment-message",
+        "status": "pending",
+    }
+    secret = "test-callback-secret-123456"
+    env = {
+        "HUIHUI_PAYMENT_CALLBACK_BASE_URL": "https://digital.example",
+        "HUIHUI_PAYMENT_CALLBACK_SECRET": secret,
+    }
+    with patch.dict(os.environ, env), patch("routers.tokens._payment_client", return_value=payment_client):
+        created = client.post(
+            "/api/token/charge",
+            headers=context["owner_headers"],
+            json={"planId": "1", "paymentMethod": "alipay", "payScene": "APP"},
+        ).json()["data"]
+        callback_body = {
+            "data": {
+                "masterOrderNo": created["orderNo"],
+                "status": "succeeded",
+                "payAmt": "10.00",
+            }
+        }
+        signature = hmac.new(
+            secret.encode(), created["orderNo"].encode(), hashlib.sha256
+        ).hexdigest()
+        callback_path = f"/api/token/payment/callback/{created['orderNo']}/{signature}"
+        first = client.post(callback_path, json=callback_body)
+        second = client.post(callback_path, json=callback_body)
+
+    assert first.json()["data"] == {"received": True, "paid": True}
+    assert second.json()["data"] == {"received": True, "duplicate": True}
+    status = client.get(
+        f"/api/token/payment/{created['id']}", headers=context["owner_headers"]
+    ).json()["data"]
+    assert status["status"] == "paid"
+    assert status["balance"] == DEFAULT_TOKEN_GRANT + 2_000_000
+    assert client.get(
+        f"/api/token/payment/{created['id']}", headers=context["other_headers"]
+    ).json()["code"] == 404
+
+
+def test_callback_amount_mismatch_never_credits_points(authorization_context):
+    context = authorization_context
+    _enable_huihui_payment_login(context)
+    payment_client = Mock()
+    payment_client.create_payment.return_value = {"status": "pending", "payMessage": "mock"}
+    secret = "test-callback-secret-123456"
+    env = {
+        "HUIHUI_PAYMENT_CALLBACK_BASE_URL": "https://digital.example",
+        "HUIHUI_PAYMENT_CALLBACK_SECRET": secret,
+    }
+    with patch.dict(os.environ, env), patch("routers.tokens._payment_client", return_value=payment_client):
+        created = client.post(
+            "/api/token/charge",
+            headers=context["owner_headers"],
+            json={"planId": "1", "paymentMethod": "wechat", "payScene": "APP"},
+        ).json()["data"]
+        signature = hmac.new(
+            secret.encode(), created["orderNo"].encode(), hashlib.sha256
+        ).hexdigest()
+        callback = client.post(
+            f"/api/token/payment/callback/{created['orderNo']}/{signature}",
+            json={
+                "masterOrderNo": created["orderNo"],
+                "status": "success",
+                "actualAmt": "9.99",
+            },
+        )
+
+    assert callback.json()["code"] == 422
+    db = SessionLocal()
+    try:
+        order = db.query(TokenPaymentOrder).filter(TokenPaymentOrder.id == created["id"]).one()
+        account = get_or_create_account(db, context["owner"].id)
+        assert order.status == "pending"
+        assert account.balance == DEFAULT_TOKEN_GRANT
+    finally:
+        db.close()
+
+
+def test_payment_callback_creates_missing_account_in_same_settlement(authorization_context):
+    context = authorization_context
+    _enable_huihui_payment_login(context)
+    payment_client = Mock()
+    payment_client.create_payment.return_value = {"status": "pending", "payMessage": "mock"}
+    secret = "test-callback-secret-123456"
+    env = {
+        "HUIHUI_PAYMENT_CALLBACK_BASE_URL": "https://digital.example",
+        "HUIHUI_PAYMENT_CALLBACK_SECRET": secret,
+    }
+    with patch.dict(os.environ, env), patch("routers.tokens._payment_client", return_value=payment_client):
+        created = client.post(
+            "/api/token/charge",
+            headers=context["owner_headers"],
+            json={"planId": "1", "paymentMethod": "alipay", "payScene": "APP"},
+        ).json()["data"]
+        db = SessionLocal()
+        try:
+            db.query(TokenAccount).filter(TokenAccount.user_id == context["owner"].id).delete()
+            db.commit()
+        finally:
+            db.close()
+        signature = hmac.new(
+            secret.encode(), created["orderNo"].encode(), hashlib.sha256
+        ).hexdigest()
+        callback = client.post(
+            f"/api/token/payment/callback/{created['orderNo']}/{signature}",
+            json={
+                "masterOrderNo": created["orderNo"],
+                "status": "success",
+                "payAmt": "10.00",
+            },
+        )
+
+    assert callback.json()["data"] == {"received": True, "paid": True}
+    db = SessionLocal()
+    try:
+        account = db.query(TokenAccount).filter(TokenAccount.user_id == context["owner"].id).one()
+        assert account.balance == DEFAULT_TOKEN_GRANT + 2_000_000
+        assert account.total_granted == DEFAULT_TOKEN_GRANT + 2_000_000
+    finally:
+        db.close()
+
+
+def test_multiple_avatars_share_owner_balance_and_usage_is_itemized():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"token-user-{suffix}", huihui_user_id=f"token-huihui-{suffix}")
+    first = Avatar(id=f"token-avatar-a-{suffix}", owner_id=user.huihui_user_id, name="甲")
+    second = Avatar(id=f"token-avatar-b-{suffix}", owner_id=user.huihui_user_id, name="乙")
+    db.add_all([user, first, second])
+    db.commit()
+    try:
+        first_reservation = reserve_avatar_tokens(db, first, "chat", "qwen-test", [{"content": "问题一"}], 128)
+        settle_reservation(
+            db,
+            first_reservation,
+            {"prompt_tokens": 60, "completion_tokens": 40, "total_tokens": 100},
+            fallback_total=999,
+        )
+        second_reservation = reserve_avatar_tokens(db, second, "takeover", "qwen-test", [{"content": "问题二"}], 128)
+        settle_reservation(
+            db,
+            second_reservation,
+            {"prompt_tokens": 120, "completion_tokens": 80, "total_tokens": 200},
+            fallback_total=999,
+        )
+
+        account = get_or_create_account(db, user.id)
+        assert account.balance == DEFAULT_TOKEN_GRANT - 300
+        assert account.total_consumed == 300
+        usages = db.query(TokenUsage).filter(TokenUsage.user_id == user.id).order_by(TokenUsage.total_tokens).all()
+        assert [(row.avatar_id, row.source, row.total_tokens) for row in usages] == [
+            (first.id, "chat", 100),
+            (second.id, "takeover", 200),
+        ]
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id.in_([first.id, second.id])).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()
+
+
+def test_concurrent_settlements_do_not_overwrite_each_other():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"concurrent-user-{suffix}", huihui_user_id=f"concurrent-huihui-{suffix}")
+    avatar = Avatar(id=f"concurrent-avatar-{suffix}", owner_id=user.huihui_user_id, name="并发测试")
+    db.add_all([user, avatar])
+    db.commit()
+    first = reserve_avatar_tokens(db, avatar, "takeover", "qwen-test", [{"content": "甲"}], 128)
+    second = reserve_avatar_tokens(db, avatar, "takeover", "qwen-test", [{"content": "乙"}], 128)
+    db.close()
+    barrier = Barrier(2, timeout=3)
+
+    def settle(reservation, total):
+        thread_db = SessionLocal()
+        try:
+            barrier.wait()
+            settle_reservation(
+                thread_db,
+                reservation,
+                {"prompt_tokens": total - 20, "completion_tokens": 20, "total_tokens": total},
+                fallback_total=999,
+            )
+        finally:
+            thread_db.close()
+
+    with ThreadPoolExecutor(max_workers=2) as pool:
+        list(pool.map(lambda args: settle(*args), [(first, 100), (second, 200)]))
+
+    db = SessionLocal()
+    try:
+        account = get_or_create_account(db, user.id)
+        assert account.balance == DEFAULT_TOKEN_GRANT - 300
+        assert account.total_consumed == 300
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id == avatar.id).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()
+
+
+def test_failed_model_request_returns_the_full_reservation():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"refund-user-{suffix}", huihui_user_id=f"refund-huihui-{suffix}")
+    avatar = Avatar(id=f"refund-avatar-{suffix}", owner_id=user.huihui_user_id, name="退款测试")
+    db.add_all([user, avatar])
+    db.commit()
+    try:
+        reservation = reserve_avatar_tokens(db, avatar, "chat", "qwen-test", [{"content": "问题"}], 128)
+        release_reservation(db, reservation, "provider error")
+        account = get_or_create_account(db, user.id)
+        usage = db.query(TokenUsage).filter(TokenUsage.id == reservation.usage_id).one()
+        assert account.balance == DEFAULT_TOKEN_GRANT
+        assert account.total_consumed == 0
+        assert usage.status == "failed"
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id == avatar.id).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()
+
+
+def test_insufficient_balance_rejects_before_model_usage_is_created():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"empty-user-{suffix}", huihui_user_id=f"empty-huihui-{suffix}")
+    avatar = Avatar(id=f"empty-avatar-{suffix}", owner_id=user.huihui_user_id, name="余额不足")
+    db.add_all([user, avatar])
+    db.commit()
+    try:
+        account = get_or_create_account(db, user.id)
+        account.balance = 1
+        db.commit()
+        with pytest.raises(InsufficientTokensError):
+            reserve_avatar_tokens(db, avatar, "chat", "qwen-test", [{"content": "问题"}], 128)
+        db.refresh(account)
+        assert account.balance == 1
+        assert db.query(TokenUsage).filter(TokenUsage.user_id == user.id).count() == 0
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id == avatar.id).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()
+
+
+def test_chat_settles_from_provider_usage_not_fallback_estimate():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"chat-user-{suffix}", huihui_user_id=f"chat-huihui-{suffix}")
+    avatar = Avatar(id=f"chat-avatar-{suffix}", owner_id=user.huihui_user_id, name="聊天测试", config={})
+    db.add_all([user, avatar])
+    db.commit()
+    try:
+        with patch(
+            "routers.chat._call_qwen",
+            return_value={
+                "answer": "测试回答",
+                "usage": {"prompt_tokens": 80, "completion_tokens": 20, "total_tokens": 100},
+            },
+        ):
+            result = _resolve_reply(
+                db,
+                avatar,
+                "测试问题",
+                [],
+                qa_pairs=[],
+                search_fn=lambda *_args: [],
+            )
+        assert result["tokenUsage"]["totalTokens"] == 100
+        assert result["tokenUsage"]["balance"] == DEFAULT_TOKEN_GRANT - 100
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id == avatar.id).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()
+
+
+@pytest.mark.asyncio
+async def test_streaming_chat_settles_final_provider_usage():
+    suffix = uuid.uuid4().hex
+    db = SessionLocal()
+    user = User(id=f"stream-user-{suffix}", huihui_user_id=f"stream-huihui-{suffix}")
+    avatar = Avatar(id=f"stream-avatar-{suffix}", owner_id=user.huihui_user_id, name="流式测试", config={})
+    db.add_all([user, avatar])
+    db.commit()
+    try:
+        chunks = iter([
+            {"content": "流式"},
+            {"content": "回答"},
+            {"usage": {"prompt_tokens": 90, "completion_tokens": 10, "total_tokens": 100}},
+        ])
+        with patch("routers.chat._iter_qwen_stream", return_value=chunks):
+            response = _stream_reply(db, avatar, "测试问题", [])
+            body = []
+            async for chunk in response.body_iterator:
+                body.append(chunk.decode() if isinstance(chunk, bytes) else chunk)
+        assert "流式" in "".join(body)
+        account = get_or_create_account(db, user.id)
+        usage = db.query(TokenUsage).filter(TokenUsage.user_id == user.id).one()
+        assert account.balance == DEFAULT_TOKEN_GRANT - 100
+        assert usage.source == "chat_stream"
+        assert usage.total_tokens == 100
+    finally:
+        db.query(TokenUsage).filter(TokenUsage.user_id == user.id).delete(synchronize_session=False)
+        db.query(TokenAccount).filter(TokenAccount.user_id == user.id).delete(synchronize_session=False)
+        db.query(Avatar).filter(Avatar.id == avatar.id).delete(synchronize_session=False)
+        db.query(User).filter(User.id == user.id).delete(synchronize_session=False)
+        db.commit()
+        db.close()

digital-avatar-app/backend/tests/test_vision_service.py

@@ -0,0 +1,98 @@
+import io
+import json
+from unittest.mock import Mock, patch
+
+import pytest
+from PIL import Image
+
+from services.chat_model_config import ChatModelConfig
+from services.vision_service import (
+    ImageValidationError,
+    build_attachment_warning,
+    call_vision_model,
+    parse_vision_analysis,
+    prepare_image,
+)
+
+
+def _image_bytes(fmt="PNG", size=(120, 80)):
+    output = io.BytesIO()
+    Image.new("RGB", size, "#f97316").save(output, format=fmt)
+    return output.getvalue()
+
+
+def _config():
+    return ChatModelConfig(
+        api_base_url="https://model.test/v1",
+        api_key="secret-key",
+        model="chat-model",
+        max_tokens=1024,
+        timeout_seconds=30,
+        vision_model="vision-model",
+        ocr_model="ocr-model",
+        vision_max_tokens=2048,
+        vision_timeout_seconds=90,
+        source="test",
+    )
+
+
+def test_prepare_image_validates_and_reencodes_without_metadata():
+    prepared = prepare_image(_image_bytes())
+
+    assert prepared.mime_type == "image/jpeg"
+    assert prepared.width == 120
+    assert prepared.height == 80
+    with Image.open(io.BytesIO(prepared.data)) as image:
+        assert image.format == "JPEG"
+        assert not image.getexif()
+
+
+def test_prepare_image_rejects_non_image_content():
+    with pytest.raises(ImageValidationError, match="格式无效"):
+        prepare_image(b"not-an-image")
+
+
+def test_vision_request_uses_openai_compatible_image_content():
+    response = Mock()
+    response.raise_for_status.return_value = None
+    response.json.return_value = {
+        "choices": [{"message": {"content": '{"category":"general_image"}'}}],
+        "usage": {"total_tokens": 88},
+    }
+    prepared = prepare_image(_image_bytes())
+
+    with patch("services.vision_service.httpx.post", return_value=response) as request:
+        result = call_vision_model(
+            prepared,
+            _config(),
+            model="vision-model",
+            prompt="describe",
+            json_output=True,
+        )
+
+    payload = request.call_args.kwargs["json"]
+    content = payload["messages"][0]["content"]
+    assert payload["model"] == "vision-model"
+    assert payload["response_format"] == {"type": "json_object"}
+    assert content[0]["type"] == "image_url"
+    assert content[0]["image_url"]["url"].startswith("data:image/jpeg;base64,")
+    assert content[1] == {"type": "text", "text": "describe"}
+    assert result["usage"]["total_tokens"] == 88
+
+
+def test_parse_medical_analysis_and_build_warning():
+    analysis = parse_vision_analysis(json.dumps({
+        "category": "medical_document",
+        "summary": "血常规报告",
+        "visible_text": "白细胞 11.2",
+        "key_facts": ["白细胞偏高"],
+        "uncertainties": ["日期模糊"],
+        "medical": {"document_type": "检验报告"},
+    }, ensure_ascii=False))
+
+    assert analysis["category"] == "medical_document"
+    assert analysis["medical"]["document_type"] == "检验报告"
+    warning = build_attachment_warning(analysis, ocr_failed=True)
+    assert "日期模糊" in warning
+    assert "人工核对" in warning
+    assert "不能替代医生诊断" in warning

digital-avatar-app/docker-compose.yml

@@ -10,6 +10,9 @@ services:
     environment:
       DATABASE_URL: sqlite:////data/avatar.db
       UPLOAD_DIR: /data/uploads
+      CHAT_MODEL_CONFIG_URL: http://host.docker.internal:8000/api/ai-models/runtime/digital-avatar
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     volumes:
       - avatar-data:/data
     expose:

digital-avatar-app/docs/H5_PRODUCTION_DEPLOYMENT.md

@@ -0,0 +1,146 @@
+# 数字分身 H5 生产接入与部署
+
+## 1. 接入方式
+
+生产会会在用户已登录后打开以下地址：
+
+```text
+https://digital.99hui.com/#/avatar/manage?token=<encodeURIComponent(会会 access token)>
+```
+
+测试环境示例：
+
+```text
+http://192.168.1.188:8099/#/avatar/manage?token=<encodeURIComponent(token)>
+```
+
+兼容参数位于域名查询串的形式，但生产统一使用上面的 hash 路由形式。必须对 token 调用 `encodeURIComponent`，不能拼接用户 ID 代替 token。
+
+免登录流程如下：
+
+1. H5 在页面渲染前读取 `token`，立即通过 `history.replaceState` 从地址栏和浏览器历史中移除。
+2. H5 调用 `POST /api/huihui/token/login`，不会把会会 token 当作数字分身接口 token 直接使用。
+3. 后端通过会会生产接口 `/im/box/netease` 换取 BOXIM 凭证，再调用 BOXIM `/user/self` 校验用户身份。
+4. 后端以返回的 `huihuiUserId` 绑定本地用户，保存会会凭证供 BOXIM 接管功能使用，并签发本系统 `app_token`。
+5. 浏览器只保存 `app_token` 和非敏感用户资料。会会原始 token 不返回浏览器存储。
+6. token 无效、过期或上游校验失败时清除旧会话并进入登录页，不会沿用上一位用户的缓存身份。
+
+## 2. 生产配置
+
+后端 `.env` 至少配置以下内容，密钥由部署平台注入，禁止提交 Git：
+
+```dotenv
+HUIHUI_DEV_MOCK=false
+HUIHUI_AUTH_BASE_URL=https://99hui.com/api/usercenter
+HUIHUI_PLATFORM_BASE_URL=https://open.99hui.com/api
+BOXIM_API_BASE_URL=https://im.99hui.com/api
+HUIHUI_APP_ID=<production-app-id>
+HUIHUI_ACCESS_ID=<production-access-id>
+HUIHUI_ACCESS_SECRET=<production-access-secret>
+HUIHUI_CLIENT_CODE=<production-client-code>
+BOXIM_TIMEOUT_SECONDS=20
+BOXIM_POLL_CONCURRENCY=8
+BOXIM_MAX_MESSAGE_AGE_SECONDS=600
+HUIHUI_PAYMENT_BASE_URL=https://open.99hui.com/api/payment-v3
+HUIHUI_PAYMENT_CALLBACK_BASE_URL=https://digital.99hui.com
+HUIHUI_PAYMENT_CALLBACK_SECRET=<至少32位随机密钥>
+HUIHUI_PAYMENT_TIMEOUT_SECONDS=30
+
+DATABASE_URL=sqlite:////data/avatar.db
+UPLOAD_DIR=/data/uploads
+CHAT_MODEL_CONFIG_URL=http://<huihuisquare-api>/api/ai-models/runtime/digital-avatar
+EMBEDDING_API_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
+EMBEDDING_API_KEY=<production-embedding-api-key>
+EMBEDDING_MODEL=text-embedding-v3
+EMBEDDING_BATCH_SIZE=10
+
+VISION_MODEL=qwen3.6-flash
+VISION_OCR_MODEL=qwen-vl-ocr
+VISION_MAX_OUTPUT_TOKENS=2048
+VISION_TIMEOUT_SECONDS=90
+VISION_TOKEN_RESERVE=12000
+CHAT_IMAGE_MAX_BYTES=8388608
+CHAT_IMAGE_MAX_PIXELS=16000000
+CHAT_ATTACHMENT_RETENTION_HOURS=24
+CHAT_ATTACHMENT_CLEANUP_MINUTES=60
+```
+
+如生产 AI 配置中心不可用，还应提供当前项目支持的 `OPENAI_API_KEY`、`OPENAI_BASE_URL`、`CHAT_MODEL` 等兜底配置。`/data` 必须挂载持久卷，数据库与知识库文件不可存放在容器临时层。
+
+`EMBEDDING_API_URL` 同时支持 OpenAI 兼容基础地址（如上面的 `/v1`）和完整的 `/v1/embeddings` 地址，后端会统一请求 `/embeddings`。发布后必须在后端容器内执行一次最小向量探针，确认返回向量数量和维度，而不能只检查 `/api/health`。
+
+积分充值使用会会支付体系的 `payment-v3/payment/pay`，渠道值为 `WECHAT` / `ALIPAY`，端内支付场景为 `APP`，微信内 H5 使用 `JSAPI`。`HUIHUI_PAYMENT_CALLBACK_SECRET` 只用于为每笔订单生成 HMAC 回调签名，不会发送到前端或直接出现在回调地址中。支付回调确认状态成功且金额与套餐价格完全一致后才增加积分，重复回调不会重复到账。
+
+## 3. 构建与发布
+
+首次发布前备份数据：
+
+```bash
+BACKUP_DIR="backups/$(date +%Y%m%d-%H%M%S)"
+mkdir -p "$BACKUP_DIR"
+cp /srv/digital-avatar/data/avatar.db "$BACKUP_DIR/"
+tar -C /srv/digital-avatar/data -czf "$BACKUP_DIR/uploads.tgz" uploads
+```
+
+在发布目录执行：
+
+```bash
+git fetch origin
+git checkout <已验收的提交SHA>
+cd digital-avatar-app
+docker compose build --pull avatar-backend avatar-frontend
+docker compose up -d avatar-backend avatar-frontend
+docker compose ps
+curl -fsS http://127.0.0.1:8099/api/health
+docker compose exec avatar-backend python -c 'import embeddings; v=embeddings.embed(["部署向量探针"]); print(len(v), len(v[0]))'
+```
+
+生产编排应把示例中的测试端口改为内网暴露，由统一 HTTPS 网关接入。后端暂时使用 SQLite，必须保持单实例写入；若扩展为多后端实例，应先迁移到 PostgreSQL，并把延迟接管任务改为共享队列。
+
+## 4. 网关要求
+
+必须使用 HTTPS。同域部署时，H5 静态资源与 `/api/` 由同一域名提供，可避免跨域和 Cookie/来源策略问题。Nginx 关键配置示例：
+
+```nginx
+server_name digital.99hui.com;
+
+location / {
+    try_files $uri $uri/ /index.html;
+}
+
+location /api/ {
+    proxy_pass http://avatar-backend:8000;
+    proxy_http_version 1.1;
+    proxy_set_header Host $host;
+    proxy_set_header X-Real-IP $remote_addr;
+    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+    proxy_set_header X-Forwarded-Proto $scheme;
+    proxy_buffering off;
+    proxy_read_timeout 300s;
+    client_max_body_size 100m;
+}
+```
+
+`proxy_buffering off` 用于数字分身 SSE 流式吐字，`client_max_body_size` 同时用于知识库文件和聊天图片上传。应用只保存图片识别结果，不保存原图；识别结果 24 小时失效，后台默认每小时清理一次。公开分享图片识别会消耗分身所有者积分，生产网关应针对 `/api/public/avatar/*/chat/images` 设置每 IP 和每分享令牌的上传频率限制，防止恶意消耗。
+
+网关和应用日志必须关闭完整 URL 查询参数记录，任何异常日志都不得输出 token、Authorization、图片 Base64、病例正文或平台密钥。建议同时设置严格的 `Referrer-Policy: no-referrer`。
+
+## 5. 发布验收
+
+1. 已登录会会用户通过带 token 链接打开后直接进入 `/avatar/manage`，不出现登录页或创建新账号页。
+2. 页面加载后地址栏中不再包含 `token`，刷新页面仍使用本地 `app_token` 正常访问。
+3. 后端用户绑定的是 BOXIM 返回的 `huihuiUserId`，不是 BOXIM 内部 `id`；原有数字分身、独立知识库和积分余额均存在。
+4. A、B 两个会会用户分别进入时只能看到各自的数字分身与知识库，不会继承上一用户缓存。
+5. 使用过期或伪造 token 时进入登录页并显示凭证失效，不得继续访问旧用户数据。
+6. 分身聊天 SSE 逐段输出正常，Markdown 正常渲染，知识库优先级和积分扣费正常。
+7. 开启 BOXIM 主动接管后保持在线，默认三分钟回复、自定义等待时间、已读回执、分身防回环和主人发言暂停均正常。
+8. 重建容器后数据库、头像、知识库文档仍存在，`/api/health` 返回成功。
+9. `https://digital.99hui.com/api/health` 可访问，证书域名和有效期正确，HTTP 自动跳转 HTTPS。
+10. 微信和支付宝各创建一笔最小套餐订单，未付款时积分不变；支付成功后回调到账一次，重复回调积分不重复增加。
+11. 私聊和公开分享各上传 JPG、PNG、WebP 图片并完成追问；上传非图片、超过 8MB 或跨分身附件时必须拒绝。
+12. 病例图片可以提取可见文字并标记待核对内容，医学影像不作确定诊断；视觉与 OCR 调用分别扣减积分。
+13. 检查服务器上传目录不残留聊天原图，数据库过期图片识别记录在清理周期后删除，日志不出现 Base64 或病例正文。
+
+## 6. 回滚
+
+保留上一版前后端镜像标签和发布前数据库/上传文件备份。代码回滚优先切回上一镜像；只有新版本执行了不可逆数据变更时才恢复数据库。恢复前先停止后端写入，恢复后对比用户数、分身数、知识库文档数并完成一次免登录和聊天验收。

digital-avatar-app/docs/IMAGE_MEDICAL_UNDERSTANDING_DESIGN.md

@@ -0,0 +1,184 @@
+# 数字分身图片与病例理解详细设计
+
+## 1. 目标与边界
+
+本功能让数字分身在私聊和公开分享聊天中接收图片，并围绕图片内容继续使用现有的“标准答题对 -> 分身独立知识库 -> Qwen 兼容模型”链路回答。
+
+第一期支持 JPEG、PNG、WebP，覆盖以下场景：
+
+1. 普通照片、截图、图表和界面图片的内容理解。
+2. 病例、处方、检查单、检验报告等图片文档的文字和表格提取。
+3. X 光、CT、MRI 等医学影像的客观可见内容描述。
+
+第一期不把通用视觉模型的输出当作医学诊断，不自动把图片或病例写入知识库，不保存原图供长期访问，也不支持 DICOM 原始影像。
+
+## 2. 核心原则
+
+- **资料优先级不变**：标准答题对最高，分身独立知识库其次，图片识别结果属于待核对的会话资料，最后才由模型组织表达。
+- **病例最小留存**：应用不把原图写入业务存储，上传内容在内存中归一化并调用视觉服务；数据库只保存结构化结果和必要元数据。
+- **严格隔离**：每条图片记录必须绑定 `avatar_id`，私聊校验分身所有者，公开聊天校验分享令牌对应的分身。
+- **不确定性显式化**：OCR 看不清、表格列错位、医学影像无法确认时必须指出待核对项，不允许补齐缺失内容。
+- **可计量**：视觉理解和病例 OCR 分别计入分身所有者的积分消耗，失败时释放预留积分。
+- **可降级**：OCR 失败但通用视觉结果有效时仍可回答；视觉主调用失败则不进入聊天发送。
+
+## 3. 总体流程
+
+```text
+用户选择图片
+  -> 前端本地预览
+  -> 私聊/公开图片上传接口
+  -> 文件大小、MIME、真实格式、像素数校验
+  -> 自动旋转、缩放、去 EXIF、统一 JPEG
+  -> 通用视觉模型分类并输出结构化 JSON
+  -> 若为病例/检查单，再调用 OCR 模型精确转录
+  -> 保存结构化结果，不持久化原图
+  -> 返回 attachmentId
+  -> 用户发送文字 + attachmentIds
+  -> 标准答题对匹配
+  -> 用文字 + 图片提取结果检索独立知识库
+  -> 把标准答案、知识片段、图片资料注入系统上下文
+  -> Qwen SSE 流式回答
+```
+
+## 4. 模型编排
+
+### 4.1 通用视觉模型
+
+默认 `qwen3.6-flash`，可在后台数字分身专用模型配置中修改。输入为归一化后的 Base64 Data URL，要求返回 JSON：
+
+```json
+{
+  "category": "general_image|document|medical_document|medical_image",
+  "summary": "客观、完整的图片描述",
+  "visible_text": "图片中可确认的文字",
+  "key_facts": ["事实1", "事实2"],
+  "uncertainties": ["无法确认的内容"],
+  "medical": {
+    "document_type": "",
+    "patient_info": {},
+    "chief_complaint": "",
+    "findings": [],
+    "measurements": [],
+    "doctor_advice": ""
+  }
+}
+```
+
+模型提示词禁止诊断、补全被遮挡文字、猜测患者身份和输出模型信息。
+
+### 4.2 病例 OCR
+
+当 `category=medical_document` 时追加调用 `qwen-vl-ocr`，按原布局转录文字和表格。OCR 文本优先替换通用视觉输出中的 `visible_text`，但保留通用视觉模型提供的分类、摘要和不确定项。
+
+### 4.3 医学影像
+
+当 `category=medical_image` 时只保存客观描述，不输出疾病结论、分期、用药或治疗方案。聊天提示词必须要求结合正规影像报告和医生意见，并显示“图片识别结果仅供辅助，不能替代医生诊断”。
+
+## 5. 数据模型
+
+新增 `chat_attachments`：
+
+| 字段 | 说明 |
+|---|---|
+| `id` | 不可猜测的附件 ID |
+| `avatar_id` | 所属数字分身，强制隔离 |
+| `filename` | 原文件名，去除路径 |
+| `mime_type` / `file_size` | 上传元数据 |
+| `status` | `processing / ready / failed` |
+| `category` | 图片分类 |
+| `summary` | 通用视觉摘要 |
+| `extracted_text` | 可确认文字/OCR 结果 |
+| `structured_data` | 结构化 JSON |
+| `warning` | 不确定项和医学提示 |
+| `vision_model` / `ocr_model` | 实际调用模型 |
+| `created_at` / `used_at` | 创建和最近使用时间 |
+
+不保存公开原图 URL。应用层不落盘原图；框架上传缓冲在请求结束时关闭，处理结果在 24 小时后自动清理。
+
+## 6. API 设计
+
+### 6.1 上传并解析
+
+- `POST /api/avatar/{avatar_id}/chat/images`
+- `POST /api/public/avatar/{share_token}/chat/images`
+- `multipart/form-data: file`
+
+成功返回：
+
+```json
+{
+  "id": "attachment-id",
+  "filename": "病例.jpg",
+  "status": "ready",
+  "category": "medical_document",
+  "summary": "门诊检查单",
+  "warning": "部分手写内容需要人工核对"
+}
+```
+
+### 6.2 聊天
+
+原聊天接口增加：
+
+```json
+{
+  "message": "请帮我看看异常指标",
+  "attachmentIds": ["attachment-id"],
+  "history": [
+    {"role": "user", "content": "上一条问题", "attachmentIds": ["attachment-id"]}
+  ]
+}
+```
+
+当前消息最多 3 张图，历史最多引用最近 3 个不同附件。后端只读取与当前 `avatar_id` 相同且状态为 `ready` 的记录。
+
+## 7. 安全与隐私
+
+- 单图最大 8MB，解码后最大 1600 万像素，最长边归一化到 4096 像素以内。
+- 使用 Pillow 验证真实图片格式并防止解压炸弹；重新编码时清除 EXIF、GPS 和其他元数据。
+- 图片不会写入 FastAPI `StaticFiles` 或知识库目录，模型请求和日志不得输出 Base64 内容。
+- 日志只记录附件 ID、分身 ID、状态、耗时和模型，不记录图片 Base64、OCR 全文、病例内容或 API Key。
+- 公开分享上传仍消耗分身所有者积分；余额不足时拒绝视觉调用。
+- 生产环境需要补充用户授权、数据处理协议、存储地域和模型供应商留存策略确认。
+
+## 8. 前端交互
+
+- 输入框左侧增加图片按钮，支持相册选择和移动端拍照。
+- 选择后显示本地缩略图和“正在识别图片”，识别完成前禁止发送。
+- 用户可删除待发送图片；发送后图片保留在当前会话气泡中，但刷新页面后不恢复原图。
+- 病例和医学影像在输入区及回答下方显示辅助提示，不使用恐吓式红色告警。
+- 上传或识别失败时保留文字输入，明确提示重新选择图片，不产生空白消息。
+
+## 9. 配置
+
+数字分身专用模型配置新增：
+
+- `vision_model_version`，默认 `qwen3.6-flash`
+- `ocr_model_version`，默认 `qwen-vl-ocr`
+
+环境变量兜底：
+
+```dotenv
+VISION_MODEL=qwen3.6-flash
+VISION_OCR_MODEL=qwen-vl-ocr
+VISION_MAX_OUTPUT_TOKENS=2048
+VISION_TIMEOUT_SECONDS=90
+VISION_TOKEN_RESERVE=12000
+CHAT_IMAGE_MAX_BYTES=8388608
+CHAT_IMAGE_MAX_PIXELS=16000000
+CHAT_ATTACHMENT_RETENTION_HOURS=24
+CHAT_ATTACHMENT_CLEANUP_MINUTES=60
+```
+
+视觉调用复用数字分身专用配置的 `api_base_url` 和 `api_key`，不额外复制密钥。
+
+## 10. 验收标准
+
+1. 普通照片、截图和图表能够返回与图片一致的描述并支持追问。
+2. 病例图片可以提取标题、患者字段、检查结果、异常指标和医生意见，模糊内容明确标记待核对。
+3. 上传后服务器业务目录不残留原图，响应和日志不包含 Base64 或完整病例正文。
+4. A 分身无法引用 B 分身附件；公开分享令牌无法访问其他分身附件。
+5. 有图片时标准答题对仍作为最高优先级事实，知识库命中次之。
+6. 视觉与 OCR 积分分别结算，失败调用释放预留积分。
+7. SSE 打字效果、Markdown、用户头像、公开分享和纯文本聊天均无回归。
+8. CT、MRI、X 光回答不作确定诊断，并显示人工复核提示。

digital-avatar-app/index.html

@@ -7,10 +7,9 @@
       content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no, viewport-fit=cover"
     />
     <title>会会数字分身</title>
-    <!-- uniapp web-view 桥接：加载后全局出现 window.uni.webView，H5 才能与原生壳通信 -->
-    <script type="text/javascript" src="https://unpkg.com/@dcloudio/uni-webview-js@0.0.10/index.js"></script>
-    <!-- 混合架构部署配置：web-view 内请把 apiBase 设为后端公网地址（如 'https://geo.99hui.com/api'）。
-         留空则回退为 '/api'（开发态由 Vite 代理到 :8000）。 -->
+    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
+    <!-- digital.99hui.com 同域部署保持 apiBase 为空，接口使用 /api。
+         仅跨域部署时才显式配置后端公网地址。 -->
     <script type="text/javascript">
       window.__APP_CONFIG__ = { apiBase: '' }
     </script>

digital-avatar-app/nginx.conf

@@ -23,6 +23,10 @@ http {
 
         root /usr/share/nginx/html;
         index index.html;
+        # Keep the application gateway aligned with the production edge gateway.
+        # Without this Nginx rejects ordinary PDF uploads with HTTP 413 before
+        # FastAPI can return its user-facing file-size validation message.
+        client_max_body_size 100m;
 
         # SPA 兜底（hash 路由下深链接也可正常加载）
         location / {

digital-avatar-app/package-lock.json

@@ -8,6 +8,7 @@
       "name": "digital-avatar-app",
       "version": "1.0.0",
       "dependencies": {
+        "@dcloudio/uni-webview-js": "^0.0.3",
         "axios": "^1.6.0",
         "pinia": "^2.1.0",
         "vue": "^3.3.0",
@@ -66,6 +67,12 @@
         "node": ">=6.9.0"
       }
     },
+    "node_modules/@dcloudio/uni-webview-js": {
+      "version": "0.0.3",
+      "resolved": "https://registry.npmmirror.com/@dcloudio/uni-webview-js/-/uni-webview-js-0.0.3.tgz",
+      "integrity": "sha512-3XDnLXg22QKC/aTLtGim6zbOK+euiDYoRf98BjGDE0lqMm1U4SLFHAvOlAHa0qck91phaceKwo73ScQQgSojQg==",
+      "license": "Apache-2.0"
+    },
     "node_modules/@esbuild/aix-ppc64": {
       "version": "0.21.5",
       "resolved": "https://registry.npmmirror.com/@esbuild/aix-ppc64/-/aix-ppc64-0.21.5.tgz",

digital-avatar-app/package.json

@@ -9,6 +9,7 @@
     "preview": "vite preview"
   },
   "dependencies": {
+    "@dcloudio/uni-webview-js": "^0.0.3",
     "axios": "^1.6.0",
     "pinia": "^2.1.0",
     "vue": "^3.3.0",

digital-avatar-app/public/3HbH8vTdwX.txt

@@ -0,0 +1 @@
+8f54097b01d5b0433f8aa5ea70ca930b

digital-avatar-app/public/8f54097b01d5b0433f8aa5ea70ca930b.txt

@@ -0,0 +1 @@
+8f54097b01d5b0433f8aa5ea70ca930b

digital-avatar-app/public/favicon.svg

@@ -0,0 +1,11 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64">
+  <defs>
+    <linearGradient id="avatar-gradient" x1="8" y1="8" x2="56" y2="56" gradientUnits="userSpaceOnUse">
+      <stop stop-color="#fb923c"/>
+      <stop offset="1" stop-color="#f97316"/>
+    </linearGradient>
+  </defs>
+  <rect width="64" height="64" rx="16" fill="url(#avatar-gradient)"/>
+  <circle cx="32" cy="24" r="10" fill="#fff"/>
+  <path d="M15 52c1.8-10 8-15 17-15s15.2 5 17 15" fill="#fff"/>
+</svg>

digital-avatar-app/src/App.vue

@@ -1,73 +1,10 @@
 <template>
   <div id="app">
     <router-view />
-    <!-- 底部导航栏 -->
-    <nav class="bottom-nav" v-if="showNav">
-      <button
-        class="nav-item"
-        :class="{ active: currentRoute === '/' || currentRoute === '/avatar/manage' }"
-        @click="navigateTo('/avatar/manage')"
-      >
-        <span class="nav-icon">🤖</span>
-        <span class="nav-label">我的分身</span>
-      </button>
-      <button
-        class="nav-item"
-        :class="{ active: currentRoute === '/authorization' }"
-        @click="navigateTo('/authorization')"
-      >
-        <span class="nav-icon">🔑</span>
-        <span class="nav-label">授权管理</span>
-      </button>
-      <button
-        class="nav-item"
-        :class="{ active: currentRoute === '/token/charge' }"
-        @click="navigateTo('/token/charge')"
-      >
-        <span class="nav-icon">💰</span>
-        <span class="nav-label">Token</span>
-      </button>
-    </nav>
+    <!-- 底部菜单暂不展示，多分身能力统一从分身卡片进入。 -->
   </div>
 </template>
 
-<script setup lang="ts">
-import { ref, onMounted, watch } from 'vue'
-import { useRouter, useRoute } from 'vue-router'
-
-const router = useRouter()
-const route = useRoute()
-
-const currentRoute = ref<string>(route.path)
-const showNav = ref<boolean>(shouldShowNav(route.path))
-
-function shouldShowNav(path: string) {
-  return path !== '/'
-    && path !== '/authorization'
-    && path !== '/avatar/create'
-    && path !== '/login/sms'
-    && !path.startsWith('/avatar/edit')
-    && !path.startsWith('/avatar/chat')
-    && !path.startsWith('/share/')
-}
-
-// 监听路由变化
-watch(() => route.path, (newPath) => {
-  currentRoute.value = newPath
-  showNav.value = shouldShowNav(newPath)
-})
-
-// 导航
-const navigateTo = (path: string) => {
-  router.push(path)
-}
-
-onMounted(() => {
-  currentRoute.value = route.path
-  showNav.value = shouldShowNav(route.path)
-})
-</script>
-
 <style>
 * {
   margin: 0;
@@ -81,48 +18,4 @@ onMounted(() => {
   background: #F8F9FA;
   padding-bottom: env(safe-area-inset-bottom);
 }
-
-/* 底部导航栏 */
-.bottom-nav {
-  position: fixed;
-  bottom: 0;
-  left: 0;
-  right: 0;
-  display: flex;
-  background: white;
-  border-top: 1px solid #EDEEF1;
-  padding-bottom: env(safe-area-inset-bottom);
-  z-index: 100;
-}
-
-.nav-item {
-  flex: 1;
-  display: flex;
-  flex-direction: column;
-  align-items: center;
-  gap: 4px;
-  padding: 8px 0;
-  background: none;
-  border: none;
-  cursor: pointer;
-  transition: color 0.2s;
-}
-
-.nav-icon {
-  font-size: 20px;
-}
-
-.nav-label {
-  font-size: 11px;
-  color: #9398AE;
-  font-weight: 500;
-}
-
-.nav-item.active .nav-label {
-  color: #F97316;
-}
-
-.nav-item.active .nav-icon {
-  filter: none;
-}
 </style>

digital-avatar-app/src/api/index.ts

@@ -9,8 +9,8 @@ interface ApiClient {
 
 // API 基址：优先级 window.__APP_CONFIG__.apiBase > 环境变量 > 默认 '/api'
 //  - 开发/Vite 代理：'/api'（由 vite.config 代理到后端 :8000）
-//  - web-view 内（混合架构）：需配置为后端公网地址，例如 'https://geo.99hui.com/api'
-//  - 同域部署的构建产物：可保持 '/api'
+//  - digital.99hui.com 同域部署：保持 '/api'
+//  - 仅跨域部署时配置后端公网地址
 function resolveBaseURL(): string {
   const cfg = (window as any).__APP_CONFIG__
   if (cfg && cfg.apiBase) return cfg.apiBase as string
@@ -131,17 +131,58 @@ export const deleteAvatar = (id: string) =>
 
 // ==================== Token 管理 API ====================
 
+export interface TokenBalance {
+  balance: number
+  totalGranted: number
+  totalConsumed: number
+}
+
+export interface TokenUsageSummary {
+  avatarId: string
+  source: string
+  promptTokens: number
+  completionTokens: number
+  totalTokens: number
+  requestCount: number
+}
+
+export interface TokenPaymentOrder {
+  id: string
+  orderNo: string
+  planId: string
+  paymentMethod: 'wechat' | 'alipay'
+  payType: 'WECHAT' | 'ALIPAY'
+  payWay: 'APP' | 'LITE' | 'JSAPI'
+  pointsAmount: number
+  price: number
+  status: 'pending' | 'paid' | 'failed'
+  providerStatus: string
+  payMessage: string
+  failureReason: string
+  balance: number
+}
+
 // 获取 Token 余额
 export const getTokenBalance = () =>
-  request.get<{ balance: number }>('/token/balance')
+  request.get<TokenBalance>('/token/balance')
 
 // 获取充值套餐
 export const getRechargePlans = () =>
   request.get<Array<{ id: string; name: string; amount: number; price: number }>>('/token/plans')
 
-// 执行充值
-export const chargeToken = (planId: string) =>
-  request.post<{ balance: number; charged: number }>('/token/charge', { planId })
+// 创建会会支付订单；支付回调成功后积分到账。
+export const chargeToken = (
+  planId: string,
+  paymentMethod: 'wechat' | 'alipay',
+  payScene: 'APP' | 'LITE' | 'JSAPI'
+) => request.post<TokenPaymentOrder>('/token/charge', { planId, paymentMethod, payScene })
+
+export const getTokenPaymentStatus = (orderId: string) =>
+  request.get<TokenPaymentOrder>(`/token/payment/${orderId}`)
+
+// 按分身和使用场景汇总 Token 消耗
+export const getTokenUsage = () =>
+  request.get<TokenUsageSummary[]>('/token/usage')
 
 // ==================== 授权管理 API ====================
 
@@ -150,19 +191,28 @@ export type AvatarPermission = 'friend' | 'chat' | 'publish' | 'browse' | 'inter
 export interface AvatarPermissionSettings {
   avatarId: string
   permissions: AvatarPermission[]
+  takeoverReplyDelaySeconds: number
+  disabledAvatarIds?: string[]
 }
 
 export const getAvatarPermissionSettings = (avatarId: string) =>
   request.get<AvatarPermissionSettings>(`/avatar/${avatarId}/permission-settings`)
 
-export const updateAvatarPermissionSettings = (avatarId: string, permissions: AvatarPermission[]) =>
-  request.put<AvatarPermissionSettings>(`/avatar/${avatarId}/permission-settings`, { permissions })
+export const updateAvatarPermissionSettings = (
+  avatarId: string,
+  permissions: AvatarPermission[],
+  takeoverReplyDelaySeconds: number
+) => request.put<AvatarPermissionSettings>(`/avatar/${avatarId}/permission-settings`, {
+  permissions,
+  takeoverReplyDelaySeconds,
+})
 
 export interface TakeoverStatus {
   enabled: boolean
   status: 'disabled' | 'connecting' | 'ready' | 'needs_login' | 'error'
   message: string
   pendingCount: number
+  takeoverReplyDelaySeconds: number
   lastPolledAt: string | null
 }
 
@@ -255,6 +305,9 @@ export interface KnowledgeDoc {
   vectorized?: boolean
   embeddingModel?: string
   chunkCount?: number
+  errorMessage?: string
+  indexStage?: string
+  indexProgress?: number
   createdAt: string
 }
 
@@ -280,12 +333,83 @@ export interface SearchResult {
 export const getKnowledgeDocs = (avatarId: string) =>
   request.get<KnowledgeDoc[]>(`/avatar/${avatarId}/knowledge/docs`)
 
-// 上传文档（支持 md/txt/pdf/doc/docx/xlsx）
-export const uploadKnowledgeDoc = (avatarId: string, file: File) => {
+const KNOWLEDGE_UPLOAD_CHUNK_SIZE = 5 * 1024 * 1024
+
+const uploadKnowledgeChunk = async (
+  avatarId: string,
+  uploadId: string,
+  chunkIndex: number,
+  chunk: Blob,
+  onProgress?: (loaded: number) => void
+) => {
+  const form = new FormData()
+  form.append('file', chunk, `chunk-${chunkIndex}`)
+  let reportedLoaded = 0
+  for (let attempt = 1; attempt <= 3; attempt += 1) {
+    try {
+      await request.post(
+        `/avatar/${avatarId}/knowledge/uploads/${uploadId}/chunks/${chunkIndex}`,
+        form,
+        {
+          headers: { 'Content-Type': 'multipart/form-data' },
+          timeout: 2 * 60 * 1000,
+          onUploadProgress: (event) => {
+            reportedLoaded = Math.max(reportedLoaded, Math.min(event.loaded, chunk.size))
+            onProgress?.(reportedLoaded)
+          }
+        }
+      )
+      return
+    } catch (error: any) {
+      const status = Number(error?.response?.status || 0)
+      const retryable = !status || status === 408 || status === 429 || status >= 500
+      if (!retryable || attempt === 3) throw error
+      await new Promise((resolve) => window.setTimeout(resolve, attempt * 800))
+    }
+  }
+}
+
+// 大文件拆成 5MB 分片，避免生产代理的请求体限制拦截整个文件。
+export const uploadKnowledgeDoc = async (
+  avatarId: string,
+  file: File,
+  onUploadProgress?: (loaded: number, total: number) => void
+) => {
+  if (file.size > KNOWLEDGE_UPLOAD_CHUNK_SIZE) {
+    const totalChunks = Math.ceil(file.size / KNOWLEDGE_UPLOAD_CHUNK_SIZE)
+    const upload: any = await request.post(`/avatar/${avatarId}/knowledge/uploads`, {
+      filename: file.name,
+      fileSize: file.size,
+      totalChunks
+    })
+    let uploadedBytes = 0
+    for (let index = 0; index < totalChunks; index += 1) {
+      const start = index * KNOWLEDGE_UPLOAD_CHUNK_SIZE
+      const chunk = file.slice(start, Math.min(start + KNOWLEDGE_UPLOAD_CHUNK_SIZE, file.size))
+      await uploadKnowledgeChunk(
+        avatarId,
+        upload.uploadId,
+        index,
+        chunk,
+        (chunkLoaded) => onUploadProgress?.(uploadedBytes + chunkLoaded, file.size)
+      )
+      uploadedBytes += chunk.size
+      onUploadProgress?.(uploadedBytes, file.size)
+    }
+    return request.post<KnowledgeDoc>(
+      `/avatar/${avatarId}/knowledge/uploads/${upload.uploadId}/complete`,
+      undefined,
+      { timeout: 2 * 60 * 1000 }
+    )
+  }
+
   const form = new FormData()
   form.append('file', file)
   return request.post<KnowledgeDoc>(`/avatar/${avatarId}/knowledge/docs`, form, {
-    headers: { 'Content-Type': 'multipart/form-data' }
+    headers: { 'Content-Type': 'multipart/form-data' },
+    // A slow mobile uplink must not be mistaken for a failed upload.
+    timeout: 10 * 60 * 1000,
+    onUploadProgress: (event) => onUploadProgress?.(event.loaded, event.total || file.size)
   })
 }
 
@@ -293,6 +417,9 @@ export const uploadKnowledgeDoc = (avatarId: string, file: File) => {
 export const deleteKnowledgeDoc = (avatarId: string, docId: string) =>
   request.delete(`/avatar/${avatarId}/knowledge/docs/${docId}`)
 
+export const retryKnowledgeDoc = (avatarId: string, docId: string) =>
+  request.post<KnowledgeDoc>(`/avatar/${avatarId}/knowledge/docs/${docId}/retry`)
+
 // 标准问答对列表
 export const getQAPairs = (avatarId: string) =>
   request.get<QAPair[]>(`/avatar/${avatarId}/knowledge/qa`)
@@ -324,15 +451,35 @@ export const searchKnowledge = (avatarId: string, q: string, topK = 5) =>
 export interface ChatMessage {
   role: 'user' | 'assistant'
   content: string
+  attachmentIds?: string[]
 }
 
 export interface ChatResponse {
   answer: string
-  source: 'qa' | 'knowledge' | 'qwen'
+  source: 'qa' | 'knowledge' | 'vision' | 'qwen'
   references?: Array<{ docId?: string; filename?: string; fileType?: string; snippet?: string; score?: number }>
 }
 
-export const sendAvatarChat = (avatarId: string, payload: { message: string; history?: ChatMessage[] }) =>
+export interface ChatAttachment {
+  id: string
+  avatarId: string
+  filename: string
+  mimeType: string
+  fileSize: number
+  status: 'processing' | 'ready' | 'failed'
+  category: 'general_image' | 'document' | 'medical_document' | 'medical_image'
+  summary: string
+  warning: string
+  expiresAt: string
+}
+
+export interface ChatPayload {
+  message: string
+  attachmentIds?: string[]
+  history?: ChatMessage[]
+}
+
+export const sendAvatarChat = (avatarId: string, payload: ChatPayload) =>
   request.post<ChatResponse>(`/avatar/${avatarId}/chat`, payload)
 
 export interface PublicAvatar {
@@ -351,19 +498,41 @@ export const createAvatarShareLink = (avatarId: string) =>
 export const getPublicAvatar = (shareToken: string) =>
   request.get<PublicAvatar>(`/public/avatar/${shareToken}`)
 
-export const sendPublicAvatarChat = (shareToken: string, payload: { message: string; history?: ChatMessage[] }) =>
+export const sendPublicAvatarChat = (shareToken: string, payload: ChatPayload) =>
   request.post<ChatResponse>(`/public/avatar/${shareToken}/chat`, payload)
 
+const imageForm = (file: File) => {
+  const form = new FormData()
+  form.append('file', file)
+  return form
+}
+
+export const uploadAvatarChatImage = (avatarId: string, file: File) =>
+  request.post<ChatAttachment>(`/avatar/${avatarId}/chat/images`, imageForm(file), {
+    headers: { 'Content-Type': 'multipart/form-data' },
+    timeout: 120000
+  })
+
+export const uploadPublicAvatarChatImage = (shareToken: string, file: File) =>
+  request.post<ChatAttachment>(`/public/avatar/${shareToken}/chat/images`, imageForm(file), {
+    headers: { 'Content-Type': 'multipart/form-data' },
+    timeout: 120000
+  })
+
 type ChatStreamHandlers = {
   onMeta: (meta: Pick<ChatResponse, 'source' | 'references'>) => void
   onDelta: (content: string) => void
 }
 
-const streamChat = async (path: string, payload: { message: string; history?: ChatMessage[] }, handlers: ChatStreamHandlers) => {
+const streamChat = async (path: string, payload: ChatPayload, handlers: ChatStreamHandlers) => {
   const headers: Record<string, string> = { 'Content-Type': 'application/json', Accept: 'text/event-stream' }
   if (_authToken) headers.Authorization = `Bearer ${_authToken}`
   const response = await fetch(`${resolveBaseURL()}${path}`, { method: 'POST', headers, body: JSON.stringify(payload) })
-  if (!response.ok || !response.body) throw new Error(`对话请求失败（${response.status}）`)
+  if (!response.ok) {
+    const errorBody = await response.json().catch(() => null)
+    throw new Error(errorBody?.detail || errorBody?.message || `对话请求失败（${response.status}）`)
+  }
+  if (!response.body) throw new Error('对话响应为空，请稍后重试')
 
   const reader = response.body.getReader()
   const decoder = new TextDecoder()
@@ -386,10 +555,10 @@ const streamChat = async (path: string, payload: { message: string; history?: Ch
   }
 }
 
-export const streamAvatarChat = (avatarId: string, payload: { message: string; history?: ChatMessage[] }, handlers: ChatStreamHandlers) =>
+export const streamAvatarChat = (avatarId: string, payload: ChatPayload, handlers: ChatStreamHandlers) =>
   streamChat(`/avatar/${avatarId}/chat/stream`, payload, handlers)
 
-export const streamPublicAvatarChat = (shareToken: string, payload: { message: string; history?: ChatMessage[] }, handlers: ChatStreamHandlers) =>
+export const streamPublicAvatarChat = (shareToken: string, payload: ChatPayload, handlers: ChatStreamHandlers) =>
   streamChat(`/public/avatar/${shareToken}/chat/stream`, payload, handlers)
 
 // ==================== 会会用户资料 API ====================
@@ -409,7 +578,7 @@ export const getUserProfile = () =>
 export interface SmsLoginResult {
   token: string
   user: UserProfile & { huihuiUserId: string; phone: string; createdAt?: string; lastLoginAt?: string }
-  huihui: { userId: string; nickname: string; avatarUrl: string; token: string }
+  huihui: { userId: string; nickname: string; avatarUrl: string }
 }
 
 // 发送短信验证码（演示模式会额外返回 devCode / dev 标记）
@@ -424,6 +593,10 @@ export const loginBySms = (phone: string, code: string) =>
 export const loginByPassword = (account: string, password: string) =>
   request.post<SmsLoginResult>('/huihui/pwd/login', { account, password })
 
+// Validate a production Huihui access token and exchange it for an app session.
+export const loginByHuihuiToken = (token: string) =>
+  request.post<SmsLoginResult>('/huihui/token/login', { token })
+
 // 当前登录用户
 export const getCurrentUser = () =>
   request.get<UserProfile & { huihuiUserId: string; phone: string }>('/huihui/me')

digital-avatar-app/src/assets/alipay.svg

@@ -0,0 +1,4 @@
+<svg fill="#1677FF" role="img" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg">
+  <title>支付宝</title>
+  <path d="M19.695 15.07c3.426 1.158 4.203 1.22 4.203 1.22V3.846c0-2.124-1.705-3.845-3.81-3.845H3.914C1.808.001.102 1.722.102 3.846v16.31c0 2.123 1.706 3.845 3.813 3.845h16.173c2.105 0 3.81-1.722 3.81-3.845v-.157s-6.19-2.602-9.315-4.119c-2.096 2.602-4.8 4.181-7.607 4.181-4.75 0-6.361-4.19-4.112-6.949.49-.602 1.324-1.175 2.617-1.497 2.025-.502 5.247.313 8.266 1.317a16.796 16.796 0 0 0 1.341-3.302H5.781v-.952h4.799V6.975H4.77v-.953h5.81V3.591s0-.409.411-.409h2.347v2.84h5.744v.951h-5.744v1.704h4.69a19.453 19.453 0 0 1-1.986 5.06c1.424.52 2.702 1.011 3.654 1.333m-13.81-2.032c-.596.06-1.71.325-2.321.869-1.83 1.608-.735 4.55 2.968 4.55 2.151 0 4.301-1.388 5.99-3.61-2.403-1.182-4.438-2.028-6.637-1.809"/>
+</svg>

digital-avatar-app/src/assets/wechat-pay.svg

@@ -0,0 +1,4 @@
+<svg fill="#07C160" role="img" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg">
+  <title>微信支付</title>
+  <path d="M8.691 2.188C3.891 2.188 0 5.476 0 9.53c0 2.212 1.17 4.203 3.002 5.55a.59.59 0 0 1 .213.665l-.39 1.48c-.019.07-.048.141-.048.213 0 .163.13.295.29.295a.326.326 0 0 0 .167-.054l1.903-1.114a.864.864 0 0 1 .717-.098 10.16 10.16 0 0 0 2.837.403c.276 0 .543-.027.811-.05-.857-2.578.157-4.972 1.932-6.446 1.703-1.415 3.882-1.98 5.853-1.838-.576-3.583-4.196-6.348-8.596-6.348zM5.785 5.991c.642 0 1.162.529 1.162 1.18a1.17 1.17 0 0 1-1.162 1.178A1.17 1.17 0 0 1 4.623 7.17c0-.651.52-1.18 1.162-1.18zm5.813 0c.642 0 1.162.529 1.162 1.18a1.17 1.17 0 0 1-1.162 1.178 1.17 1.17 0 0 1-1.162-1.178c0-.651.52-1.18 1.162-1.18zm5.34 2.867c-1.797-.052-3.746.512-5.28 1.786-1.72 1.428-2.687 3.72-1.78 6.22.942 2.453 3.666 4.229 6.884 4.229.826 0 1.622-.12 2.361-.336a.722.722 0 0 1 .598.082l1.584.926a.272.272 0 0 0 .14.047c.134 0 .24-.111.24-.247 0-.06-.023-.12-.038-.177l-.327-1.233a.582.582 0 0 1-.023-.156.49.49 0 0 1 .201-.398C23.024 18.48 24 16.82 24 14.98c0-3.21-2.931-5.837-6.656-6.088V8.89c-.135-.01-.27-.027-.407-.03zm-2.53 3.274c.535 0 .969.44.969.982a.976.976 0 0 1-.969.983.976.976 0 0 1-.969-.983c0-.542.434-.982.97-.982zm4.844 0c.535 0 .969.44.969.982a.976.976 0 0 1-.969.983.976.976 0 0 1-.969-.983c0-.542.434-.982.969-.982z"/>
+</svg>

digital-avatar-app/src/main.ts

@@ -1,14 +1,16 @@
 import { createApp } from 'vue'
+import uniWebView from '@dcloudio/uni-webview-js'
 import App from './App.vue'
 import router from './router'
 import pinia from './store'
-import { getLaunchParams, onNativeMessage, UniEvents } from '@/utils/uniapp-bridge'
+import { getLaunchParams, onNativeMessage, stripLaunchToken, UniEvents } from '@/utils/uniapp-bridge'
 import { useAvatarStore } from '@/store/avatar'
 import { useUserStore } from '@/store/user'
-import { setAuthToken } from '@/api'
+
+// Bundle the bridge locally so the H5 does not depend on an external CDN.
+;(window as any).uni = (window as any).uni || uniWebView
 
 const app = createApp(App)
-app.use(router)
 app.use(pinia)
 
 // —— 混合架构：在挂载前注入 uniapp 壳传入的认证与会会资料 ——
@@ -16,10 +18,8 @@ const params = getLaunchParams()
 const avatarStore = useAvatarStore(pinia)
 const userStore = useUserStore(pinia)
 
-// 恢复本地短信登录会话（会会 userId ↔ 本系统用户）
-userStore.loadFromStorage()
-if (userStore.isLogin && userStore.user) {
-  setAuthToken(userStore.token)
+function syncAvatarProfile() {
+  if (!userStore.isLogin || !userStore.user) return
   avatarStore.setNativeProfile({
     userId: (userStore.user as any).huihuiUserId || '',
     nickname: userStore.user.nickname || '',
@@ -27,9 +27,42 @@ if (userStore.isLogin && userStore.user) {
   })
 }
 
-if (params.token) {
-  setAuthToken(params.token)
+async function exchangeInjectedToken(token: string) {
+  stripLaunchToken()
+  userStore.clearSession()
+  await userStore.loginByToken(token)
+  syncAvatarProfile()
 }
+
+// 原生 → H5：注册消息处理（壳通过 web-view.evalJS 调用）
+onNativeMessage((msg) => {
+  if (!msg || !msg.type) return
+  if (msg.type === 'tokenRefresh' && msg.token) {
+    void exchangeInjectedToken(msg.token).catch(() => {
+      sessionStorage.setItem('hh_sso_error', '会会登录凭证无效或已过期，请重新进入')
+      void router.replace('/login/sms')
+    })
+  }
+  if (msg.type === 'userUpdate' && msg.user) {
+    avatarStore.setNativeProfile(msg.user)
+  }
+})
+
+async function bootstrap() {
+  // A URL token represents the current production user and must override stale storage.
+  if (params.token) {
+    try {
+      await exchangeInjectedToken(params.token)
+      await router.replace(window.location.hash.slice(1) || '/avatar/manage')
+    } catch {
+      sessionStorage.setItem('hh_sso_error', '会会登录凭证无效或已过期，请重新进入')
+      await router.replace('/login/sms')
+    }
+  } else {
+    userStore.loadFromStorage()
+    syncAvatarProfile()
+  }
+
   if (params.userId || params.nickname || params.avatar) {
     avatarStore.setNativeProfile({
       userId: params.userId || '',
@@ -38,18 +71,11 @@ if (params.userId || params.nickname || params.avatar) {
     })
   }
 
-// 原生 → H5：注册消息处理（壳通过 web-view.evalJS 调用）
-onNativeMessage((msg) => {
-  if (!msg || !msg.type) return
-  if (msg.type === 'tokenRefresh' && msg.token) {
-    setAuthToken(msg.token)
-  }
-  if (msg.type === 'userUpdate' && msg.user) {
-    avatarStore.setNativeProfile(msg.user)
-  }
-})
-
+  app.use(router)
   app.mount('#app')
 
   // 通知原生壳：H5 已就绪
   UniEvents.ready()
+}
+
+void bootstrap()

digital-avatar-app/src/router/index.ts

@@ -45,11 +45,17 @@ const routes: RouteRecordRaw[] = [
     component: () => import('@/views/AuthorizationManage.vue'),
     meta: { title: '授权管理', requiresAuth: true }
   },
+  {
+    path: '/avatar/:avatarId/authorization',
+    name: 'AvatarAuthorizationManage',
+    component: () => import('@/views/AuthorizationManage.vue'),
+    meta: { title: '授权管理', requiresAuth: true }
+  },
   {
     path: '/token/charge',
     name: 'TokenCharge',
     component: () => import('@/views/TokenCharge.vue'),
-    meta: { title: 'Token充值', requiresAuth: true }
+    meta: { title: '积分充值', requiresAuth: true }
   },
   {
     path: '/avatar/card',
@@ -81,6 +87,12 @@ const routes: RouteRecordRaw[] = [
     component: () => import('@/views/KnowledgeManage.vue'),
     meta: { title: '知识库管理', requiresAuth: true }
   },
+  {
+    path: '/avatar/:avatarId/knowledge',
+    name: 'AvatarKnowledgeManage',
+    component: () => import('@/views/KnowledgeManage.vue'),
+    meta: { title: '知识库管理', requiresAuth: true }
+  },
   {
     path: '/knowledge/qa/create',
     name: 'QaPairCreate',
@@ -93,6 +105,18 @@ const routes: RouteRecordRaw[] = [
     component: () => import('@/views/QaPairEdit.vue'),
     meta: { title: '编辑问答对', requiresAuth: true }
   },
+  {
+    path: '/avatar/:avatarId/knowledge/qa/create',
+    name: 'AvatarQaPairCreate',
+    component: () => import('@/views/QaPairEdit.vue'),
+    meta: { title: '添加问答对', requiresAuth: true }
+  },
+  {
+    path: '/avatar/:avatarId/knowledge/qa/:qaId/edit',
+    name: 'AvatarQaPairEdit',
+    component: () => import('@/views/QaPairEdit.vue'),
+    meta: { title: '编辑问答对', requiresAuth: true }
+  },
   {
     path: '/login/sms',
     name: 'SmsLogin',

digital-avatar-app/src/store/avatar.ts

@@ -1,13 +1,15 @@
 import { defineStore } from 'pinia'
 import { ref } from 'vue'
-import { getAvatarList, createAvatar as apiCreate, deleteAvatar as apiDelete, getTokenBalance, getUserProfile } from '@/api'
+import { getAvatarList, createAvatar as apiCreate, deleteAvatar as apiDelete, getTokenBalance, getTokenUsage, getUserProfile } from '@/api'
 import { unwrapListData } from '@/utils/avatar-page-data'
 
 export const useAvatarStore = defineStore('avatar', () => {
   // 已创建的分身列表（来自后端）
   const avatars = ref<any[]>([])
-  // 全局 Token 余额（来自后端）
+  // 当前用户所有分身共享的 Token 账户
   const tokenBalance = ref<number>(0)
+  const tokenConsumed = ref<number>(0)
+  const tokenUsageByAvatar = ref<Record<string, number>>({})
   // 当前选中分身 id
   const currentAvatarId = ref<string | null>(null)
   // 会会用户资料（头像/昵称，来自会会接口）
@@ -29,11 +31,24 @@ export const useAvatarStore = defineStore('avatar', () => {
     try {
       const res = await getTokenBalance()
       tokenBalance.value = (res as any)?.balance ?? 0
+      tokenConsumed.value = (res as any)?.totalConsumed ?? 0
     } catch (e) {
       console.error('加载余额失败', e)
     }
   }
 
+  const loadTokenUsage = async () => {
+    try {
+      const rows = await getTokenUsage()
+      tokenUsageByAvatar.value = rows.reduce<Record<string, number>>((result, row) => {
+        result[row.avatarId] = (result[row.avatarId] || 0) + row.totalTokens
+        return result
+      }, {})
+    } catch (e) {
+      console.error('加载 Token 用量失败', e)
+    }
+  }
+
   // 拉取会会用户资料（头像/昵称）
   const loadUserProfile = async () => {
     // 若已通过 uniapp 壳注入（混合架构），优先保留，不回退到后端 mock
@@ -82,10 +97,13 @@ export const useAvatarStore = defineStore('avatar', () => {
   return {
     avatars,
     tokenBalance,
+    tokenConsumed,
+    tokenUsageByAvatar,
     currentAvatarId,
     userProfile,
     loadAvatars,
     loadTokenBalance,
+    loadTokenUsage,
     loadUserProfile,
     setNativeProfile,
     addAvatar,

digital-avatar-app/src/store/user.ts

@@ -1,6 +1,16 @@
 import { defineStore } from 'pinia'
 import { ref } from 'vue'
-import { setAuthToken, sendSmsCode, loginBySms, loginByPassword, logoutUser, type UserProfile } from '@/api'
+import {
+  setAuthToken,
+  sendSmsCode,
+  loginBySms,
+  loginByPassword,
+  loginByHuihuiToken,
+  logoutUser,
+  type SmsLoginResult,
+  type UserProfile
+} from '@/api'
+import { clearHuihuiEmbeddedMode, markHuihuiEmbeddedMode } from '@/utils/embed-mode'
 
 const TOKEN_KEY = 'hh_app_token'
 const USER_KEY = 'hh_app_user'
@@ -37,6 +47,23 @@ export const useUserStore = defineStore('smsuser', () => {
     localStorage.removeItem(USER_KEY)
   }
 
+  const clearSession = () => {
+    token.value = ''
+    user.value = null
+    isLogin.value = false
+    setAuthToken(null)
+    clearLocal()
+  }
+
+  const acceptLogin = (res: SmsLoginResult) => {
+    token.value = res.token
+    user.value = { ...(res.user || {}), ...(res.huihui || {}) }
+    isLogin.value = true
+    setAuthToken(res.token)
+    persist()
+    return res
+  }
+
   // 发送验证码（返回结果，演示模式含 devCode）
   const sendCode = async (phone: string) => {
     return await sendSmsCode(phone)
@@ -44,24 +71,22 @@ export const useUserStore = defineStore('smsuser', () => {
 
   // 短信登录
   const login = async (phone: string, code: string) => {
-    const res: any = await loginBySms(phone, code)
-    token.value = res.token
-    user.value = { ...(res.user || {}), ...(res.huihui || {}) }
-    isLogin.value = true
-    setAuthToken(res.token)
-    persist()
-    return res
+    const result = await loginBySms(phone, code)
+    clearHuihuiEmbeddedMode()
+    return acceptLogin(result)
   }
 
   // 账号密码登录
   const loginByPwd = async (account: string, password: string) => {
-    const res: any = await loginByPassword(account, password)
-    token.value = res.token
-    user.value = { ...(res.user || {}), ...(res.huihui || {}) }
-    isLogin.value = true
-    setAuthToken(res.token)
-    persist()
-    return res
+    const result = await loginByPassword(account, password)
+    clearHuihuiEmbeddedMode()
+    return acceptLogin(result)
+  }
+
+  const loginByToken = async (huihuiToken: string) => {
+    const result = await loginByHuihuiToken(huihuiToken)
+    markHuihuiEmbeddedMode()
+    return acceptLogin(result)
   }
 
   // 退出
@@ -71,12 +96,20 @@ export const useUserStore = defineStore('smsuser', () => {
     } catch {
       /* 忽略网络错误，本地清除即可 */
     }
-    token.value = ''
-    user.value = null
-    isLogin.value = false
-    setAuthToken(null)
-    clearLocal()
+    clearHuihuiEmbeddedMode()
+    clearSession()
   }
 
-  return { token, user, isLogin, loadFromStorage, sendCode, login, loginByPwd, logout }
+  return {
+    token,
+    user,
+    isLogin,
+    loadFromStorage,
+    clearSession,
+    sendCode,
+    login,
+    loginByPwd,
+    loginByToken,
+    logout
+  }
 })

digital-avatar-app/src/types/uni-webview-js.d.ts

@@ -0,0 +1,4 @@
+declare module '@dcloudio/uni-webview-js' {
+  const uniWebView: unknown
+  export default uniWebView
+}

digital-avatar-app/src/utils/avatar-page-data.d.ts

@@ -41,5 +41,10 @@ export function pickAvatarId(
   currentAvatarId: string | null | undefined,
   avatars?: AvatarPageRecord[]
 ): string | null
+export function pickScopedAvatarId(
+  routeAvatarId: string | string[] | null | undefined,
+  currentAvatarId: string | null | undefined,
+  avatars?: AvatarPageRecord[]
+): string | null
 export function normalizeAvatarEditForm(avatar?: AvatarPageRecord): AvatarEditForm
 export function buildAvatarUpdatePayload(form: AvatarEditForm): AvatarUpdatePayload

digital-avatar-app/src/utils/avatar-page-data.js

@@ -8,6 +8,11 @@ export function pickAvatarId(currentAvatarId, avatars) {
   return currentAvatarId || avatars?.[0]?.id || null
 }
 
+export function pickScopedAvatarId(routeAvatarId, currentAvatarId, avatars) {
+  const requested = Array.isArray(routeAvatarId) ? routeAvatarId[0] : routeAvatarId
+  return requested ? String(requested) : pickAvatarId(currentAvatarId, avatars)
+}
+
 export function normalizeAvatarEditForm(avatar = {}) {
   const config = avatar.config || {}
   return {

digital-avatar-app/src/utils/embed-mode.ts

@@ -0,0 +1,13 @@
+const HUIHUI_EMBED_MODE_KEY = 'hh_huihui_embed_mode'
+
+export function markHuihuiEmbeddedMode(): void {
+  sessionStorage.setItem(HUIHUI_EMBED_MODE_KEY, '1')
+}
+
+export function clearHuihuiEmbeddedMode(): void {
+  sessionStorage.removeItem(HUIHUI_EMBED_MODE_KEY)
+}
+
+export function isHuihuiEmbeddedMode(): boolean {
+  return sessionStorage.getItem(HUIHUI_EMBED_MODE_KEY) === '1'
+}

digital-avatar-app/src/utils/uniapp-bridge.ts

@@ -1,11 +1,10 @@
 // 会会数字分身 H5 ↔ uniapp 原生壳 桥接工具
 // 协议详见 uniapp-avatar/README.md
 //
-// 引入方式：在 index.html 中加载 uniapp web-view bridge：
-//   <script src="https://unpkg.com/@dcloudio/uni-webview-js@0.0.10/index.js"></script>
-// 引入后全局会出现 window.uni.webView，H5 即可用 postMessage 与原生通信。
+// uni-webview bridge is bundled by main.ts; no external CDN is required.
 
 const BRIDGE_HANDLER = '__uniBridgeHandle__'
+const nativeMessageHandlers = new Set<(message: any) => void>()
 
 export interface UniLaunchParams {
   token?: string
@@ -15,28 +14,64 @@ export interface UniLaunchParams {
   ts?: string
 }
 
+const PARAM_KEYS: (keyof UniLaunchParams)[] = ['token', 'userId', 'nickname', 'avatar', 'ts']
+
+function readParams(search: string, target: UniLaunchParams): void {
+  const sp = new URLSearchParams(search)
+  for (const key of PARAM_KEYS) {
+    const value = sp.get(key)
+    if (value) target[key] = value
+  }
+}
+
 // 是否运行在 uniapp web-view 环境中
 export function isInUniWebView(): boolean {
-  return !!(window as any).uni?.webView
+  const runtime = window as any
+  const userAgent = navigator.userAgent
+  const isDCloudApp = Boolean(
+    runtime.plus ||
+    runtime.__dcloud_weex_postMessage ||
+    runtime.__dcloud_weex_ ||
+    /uni-app|Html5Plus/i.test(userAgent)
+  )
+  const isMiniProgram = Boolean(
+    (/miniProgram/i.test(userAgent) && (runtime.wx?.miniProgram || runtime.qq?.miniProgram)) ||
+    runtime.my ||
+    runtime.swan?.webView ||
+    runtime.tt?.miniProgram
+  )
+  return Boolean(runtime.uni?.webView && (isDCloudApp || isMiniProgram))
 }
 
 // 解析 web-view 加载 URL 时原生注入的参数（token / 会会用户）
 export function getLaunchParams(): UniLaunchParams {
-  const sp = new URLSearchParams(window.location.search)
   const params: UniLaunchParams = {}
-  const token = sp.get('token')
-  const userId = sp.get('userId')
-  const nickname = sp.get('nickname')
-  const avatar = sp.get('avatar')
-  const ts = sp.get('ts')
-  if (token) params.token = token
-  if (userId) params.userId = userId
-  if (nickname) params.nickname = decodeURIComponent(nickname)
-  if (avatar) params.avatar = decodeURIComponent(avatar)
-  if (ts) params.ts = ts
+  readParams(window.location.search, params)
+  const hashQueryIndex = window.location.hash.indexOf('?')
+  if (hashQueryIndex >= 0) {
+    readParams(window.location.hash.slice(hashQueryIndex + 1), params)
+  }
   return params
 }
 
+// Remove the one-time login credential before any route is rendered or logged.
+export function stripLaunchToken(): void {
+  const url = new URL(window.location.href)
+  url.searchParams.delete('token')
+
+  const hash = url.hash.slice(1)
+  const queryIndex = hash.indexOf('?')
+  if (queryIndex >= 0) {
+    const path = hash.slice(0, queryIndex)
+    const hashParams = new URLSearchParams(hash.slice(queryIndex + 1))
+    hashParams.delete('token')
+    const query = hashParams.toString()
+    url.hash = `${path}${query ? `?${query}` : ''}`
+  }
+
+  window.history.replaceState(window.history.state, '', `${url.pathname}${url.search}${url.hash}`)
+}
+
 // H5 → 原生：发送事件（需引入 uniapp web-view bridge）
 export function postToNative(message: Record<string, any>): boolean {
   if (!isInUniWebView()) return false
@@ -45,15 +80,19 @@ export function postToNative(message: Record<string, any>): boolean {
 }
 
 // 原生 → H5：注册消息处理（原生通过 web-view.evalJS 调用 window.__uniBridgeHandle__）
-export function onNativeMessage(handler: (message: any) => void): void {
+export function onNativeMessage(handler: (message: any) => void): () => void {
+  nativeMessageHandlers.add(handler)
   ;(window as any)[BRIDGE_HANDLER] = (message: any) => {
+    for (const listener of nativeMessageHandlers) {
       try {
-      handler(message)
+        listener(message)
       } catch (e) {
         console.error('[uniBridge] handler error', e)
       }
     }
   }
+  return () => nativeMessageHandlers.delete(handler)
+}
 
 // 便捷事件
 export const UniEvents = {
@@ -61,5 +100,6 @@ export const UniEvents = {
   needLogin: () => postToNative({ type: 'needLogin' }),
   setTitle: (title: string) => postToNative({ type: 'setTitle', title }),
   navigate: (path: string) => postToNative({ type: 'navigate', path }),
+  payment: (payment: Record<string, any>) => postToNative({ type: 'payment', payment }),
   back: () => postToNative({ type: 'back' })
 }

digital-avatar-app/src/views/AuthorizationManage.vue

@@ -1,6 +1,6 @@
 <template>
-  <div class="authorization-page">
-    <header class="page-header">
+  <div class="authorization-page" :class="{ embedded: isEmbedded }">
+    <header v-if="!isEmbedded" class="page-header">
       <button class="back-button" type="button" aria-label="返回数字分身管理" @click="goBack">
         <svg viewBox="0 0 24 24" aria-hidden="true">
           <path d="m15 18-6-6 6-6" />
@@ -64,7 +64,7 @@
             <span class="permission-copy">
               <strong>{{ item.title }}</strong>
               <small>
-                {{ item.description }}
+                {{ item.key === 'takeover' ? takeoverDescription : item.description }}
                 <span
                   v-if="item.key === 'takeover' && takeoverConnectionLabel"
                   class="connection-state"
@@ -79,6 +79,33 @@
           </button>
         </section>
 
+        <section v-if="permissionState.takeover" class="takeover-delay-card" aria-label="自动回复等待时间">
+          <div class="delay-heading">
+            <div>
+              <strong>自动回复等待时间</strong>
+              <small>等待期间主人发言会取消本次回复，最短 3 秒</small>
+            </div>
+            <span>{{ formattedTakeoverDelay }}</span>
+          </div>
+          <div class="delay-control">
+            <input
+              v-model.number="takeoverDelayValue"
+              type="number"
+              inputmode="numeric"
+              step="1"
+              :min="takeoverDelayUnit === 'minutes' ? 1 : 3"
+              :max="takeoverDelayUnit === 'minutes' ? 1440 : 86400"
+              aria-label="等待时间"
+              :disabled="loading || saving"
+              @blur="normalizeTakeoverDelay"
+            />
+            <select v-model="takeoverDelayUnit" aria-label="等待时间单位" :disabled="loading || saving">
+              <option value="seconds">秒</option>
+              <option value="minutes">分钟</option>
+            </select>
+          </div>
+        </section>
+
         <p v-if="errorMessage" class="error-message" role="alert">{{ errorMessage }}</p>
       </template>
 
@@ -95,6 +122,9 @@
     </main>
 
     <footer v-if="activeAvatarId" class="save-area">
+      <button v-if="isEmbedded" class="footer-back-button" type="button" :disabled="saving" @click="goBack">
+        返回
+      </button>
       <button class="save-button" type="button" :disabled="loading || saving" @click="saveSettings()">
         <span v-if="saving" class="saving-spinner" aria-hidden="true"></span>
         {{ saving ? '保存中...' : '保存授权设置' }}
@@ -109,7 +139,7 @@
 
 <script setup lang="ts">
 import { computed, onBeforeUnmount, onMounted, reactive, ref } from 'vue'
-import { useRouter } from 'vue-router'
+import { useRoute, useRouter } from 'vue-router'
 import {
   getAvatarPermissionSettings,
   getTakeoverStatus,
@@ -118,12 +148,16 @@ import {
   type TakeoverStatus,
 } from '@/api'
 import { useAvatarStore } from '@/store/avatar'
-import { pickAvatarId } from '@/utils/avatar-page-data.js'
+import { pickScopedAvatarId } from '@/utils/avatar-page-data.js'
+import { isHuihuiEmbeddedMode } from '@/utils/embed-mode'
 
 type PermissionState = Record<AvatarPermission, boolean>
 
 const router = useRouter()
+const route = useRoute()
 const avatarStore = useAvatarStore()
+const avatarId = computed(() => pickScopedAvatarId(route.params.avatarId, avatarStore.currentAvatarId, avatarStore.avatars))
+const isEmbedded = isHuihuiEmbeddedMode()
 
 const permissionItems: Array<{
   key: AvatarPermission
@@ -164,7 +198,7 @@ const permissionItems: Array<{
   {
     key: 'takeover',
     title: '分身主动接管聊天回复',
-    description: '收到私聊消息 3 秒后回复，主人发言时暂停',
+    description: '收到私聊消息后按设定时间回复，主人发言时暂停',
     tone: 'cyan',
   },
 ]
@@ -183,6 +217,8 @@ const saving = ref(false)
 const errorMessage = ref('')
 const toastMessage = ref('')
 const takeoverStatus = ref<TakeoverStatus | null>(null)
+const takeoverDelayValue = ref(3)
+const takeoverDelayUnit = ref<'seconds' | 'minutes'>('minutes')
 let toastTimer: number | undefined
 let takeoverStatusTimer: number | undefined
 
@@ -203,6 +239,38 @@ const takeoverConnectionTone = computed(() => {
   return 'connecting'
 })
 
+const takeoverDelaySeconds = computed(() => {
+  const value = Math.trunc(Number(takeoverDelayValue.value) || 0)
+  return takeoverDelayUnit.value === 'minutes' ? value * 60 : value
+})
+
+const formattedTakeoverDelay = computed(() => {
+  const seconds = takeoverDelaySeconds.value
+  if (seconds > 0 && seconds % 60 === 0) return `${seconds / 60} 分钟`
+  return `${seconds} 秒`
+})
+
+const takeoverDescription = computed(() =>
+  `收到私聊消息 ${formattedTakeoverDelay.value}后回复，主人发言时暂停`
+)
+
+const applyTakeoverDelay = (seconds: number) => {
+  const normalized = Number.isFinite(seconds) && seconds >= 3 ? Math.trunc(seconds) : 180
+  if (normalized % 60 === 0) {
+    takeoverDelayUnit.value = 'minutes'
+    takeoverDelayValue.value = normalized / 60
+  } else {
+    takeoverDelayUnit.value = 'seconds'
+    takeoverDelayValue.value = normalized
+  }
+}
+
+const normalizeTakeoverDelay = () => {
+  const min = takeoverDelayUnit.value === 'minutes' ? 1 : 3
+  const max = takeoverDelayUnit.value === 'minutes' ? 1440 : 86400
+  takeoverDelayValue.value = Math.min(max, Math.max(min, Math.trunc(Number(takeoverDelayValue.value) || min)))
+}
+
 const setPermissions = (permissions: AvatarPermission[]) => {
   const enabled = new Set(permissions)
   for (const item of permissionItems) permissionState[item.key] = enabled.has(item.key)
@@ -258,6 +326,7 @@ const loadSettings = async () => {
   try {
     const settings = await getAvatarPermissionSettings(activeAvatarId.value)
     setPermissions(settings.permissions || [])
+    applyTakeoverDelay(settings.takeoverReplyDelaySeconds || 180)
     await loadTakeoverStatus()
     scheduleTakeoverStatusRefresh()
   } catch (error: any) {
@@ -284,8 +353,18 @@ const saveSettings = async (takeoverToggle = false): Promise<boolean> => {
   saving.value = true
   errorMessage.value = ''
   try {
-    const settings = await updateAvatarPermissionSettings(activeAvatarId.value, selectedPermissions())
+    normalizeTakeoverDelay()
+    if (takeoverDelaySeconds.value < 3 || takeoverDelaySeconds.value > 86400) {
+      errorMessage.value = '自动回复等待时间需在 3 秒到 24 小时之间'
+      return false
+    }
+    const settings = await updateAvatarPermissionSettings(
+      activeAvatarId.value,
+      selectedPermissions(),
+      takeoverDelaySeconds.value,
+    )
     setPermissions(settings.permissions || [])
+    applyTakeoverDelay(settings.takeoverReplyDelaySeconds || 180)
     await loadTakeoverStatus()
     scheduleTakeoverStatusRefresh()
     if (takeoverToggle) {
@@ -307,7 +386,8 @@ const goCreate = () => router.push('/avatar/create')
 
 onMounted(async () => {
   if (!avatarStore.avatars.length) await avatarStore.loadAvatars()
-  activeAvatarId.value = pickAvatarId(avatarStore.currentAvatarId, avatarStore.avatars) || ''
+  activeAvatarId.value = avatarId.value || ''
+  if (activeAvatarId.value) avatarStore.currentAvatarId = activeAvatarId.value
   await loadSettings()
 })
 
@@ -391,6 +471,10 @@ svg {
   padding: 0 20px;
 }
 
+.authorization-page.embedded .page-content {
+  padding-top: 16px;
+}
+
 .permission-intro {
   min-height: 96px;
   padding: 15px 16px 14px;
@@ -456,6 +540,75 @@ svg {
   min-height: 76px;
 }
 
+.takeover-delay-card {
+  margin-top: 12px;
+  padding: 16px;
+  border: 1px solid #dff1ef;
+  border-radius: 15px;
+  background: linear-gradient(135deg, #f5fcfb 0%, #fff 100%);
+  box-shadow: 0 8px 24px rgba(53, 166, 162, .06);
+}
+
+.delay-heading {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 12px;
+}
+
+.delay-heading strong,
+.delay-heading small {
+  display: block;
+}
+
+.delay-heading strong {
+  font-size: 14px;
+  line-height: 1.4;
+}
+
+.delay-heading small {
+  margin-top: 5px;
+  color: #8c929f;
+  font-size: 11px;
+  line-height: 1.55;
+}
+
+.delay-heading > span {
+  flex: none;
+  padding: 4px 8px;
+  border-radius: 999px;
+  color: #258e8a;
+  background: #e8f8f6;
+  font-size: 11px;
+  font-weight: 700;
+}
+
+.delay-control {
+  margin-top: 14px;
+  display: grid;
+  grid-template-columns: minmax(0, 1fr) 88px;
+  gap: 10px;
+}
+
+.delay-control input,
+.delay-control select {
+  min-width: 0;
+  height: 42px;
+  padding: 0 12px;
+  border: 1px solid #dfe5e8;
+  border-radius: 11px;
+  outline: none;
+  color: #222528;
+  background: #fff;
+  font: inherit;
+}
+
+.delay-control input:focus,
+.delay-control select:focus {
+  border-color: #35a6a2;
+  box-shadow: 0 0 0 3px rgba(53, 166, 162, .1);
+}
+
 .permission-icon {
   width: 34px;
   height: 34px;
@@ -596,12 +749,15 @@ svg {
   bottom: 0;
   width: min(100%, 390px);
   padding: 12px 20px calc(20px + env(safe-area-inset-bottom));
+  display: flex;
+  gap: 10px;
   background: linear-gradient(to bottom, rgba(250, 250, 250, 0), #fafafa 20%, #fafafa 100%);
   transform: translateX(-50%);
 }
 
 .save-button {
-  width: 100%;
+  min-width: 0;
+  flex: 1;
   height: 48px;
   display: flex;
   align-items: center;
@@ -617,6 +773,20 @@ svg {
   cursor: pointer;
 }
 
+.footer-back-button {
+  flex: 0 0 96px;
+  height: 48px;
+  border: 1px solid #eadfd6;
+  border-radius: 24px;
+  color: #6f665f;
+  background: #fff;
+  font-size: 14px;
+  font-weight: 500;
+  cursor: pointer;
+}
+
+.footer-back-button:disabled { opacity: .58; }
+
 .save-button:disabled {
   opacity: .68;
 }

digital-avatar-app/src/views/AvatarChat.vue

@@ -1,5 +1,5 @@
 <template>
-  <div class="chat-page">
+  <div class="chat-page" :class="{ 'has-pending-images': pendingImages.length }">
     <header class="chat-header">
       <button v-if="!isPublic" class="back-btn" @click="router.back()">‹</button>
       <div class="avatar-heading">
@@ -31,6 +31,12 @@
           <span v-else>{{ avatar?.emoji || '🤖' }}</span>
         </div>
         <div class="message-column">
+          <div v-if="message.attachments?.length" class="message-images">
+            <figure v-for="attachment in message.attachments" :key="attachment.id" class="message-image-card">
+              <img :src="attachment.previewUrl" :alt="attachment.filename" />
+              <figcaption v-if="attachment.warning">{{ attachment.warning }}</figcaption>
+            </figure>
+          </div>
           <div class="message-bubble" :class="{ streaming: sending && message.role === 'assistant' && index === messages.length - 1 }">
             <template v-if="message.role === 'assistant'">
               <span
@@ -70,24 +76,66 @@
     </main>
 
     <form class="composer" @submit.prevent="sendMessage(inputText)">
-      <textarea v-model="inputText" rows="1" :disabled="sending" placeholder="输入你想聊的内容…" @keydown.enter.exact.prevent="sendMessage(inputText)"></textarea>
-      <button class="send-btn" type="submit" :disabled="sending || !inputText.trim()">发送</button>
+      <div v-if="pendingImages.length" class="pending-images">
+        <div v-for="image in pendingImages" :key="image.localId" class="pending-image" :class="image.status">
+          <img :src="image.previewUrl" :alt="image.filename" />
+          <div class="pending-image-copy">
+            <strong>{{ image.status === 'uploading' ? '正在识别图片…' : image.summary || image.filename }}</strong>
+            <span>{{ image.status === 'uploading' ? '正在提取图片中的可见内容' : categoryLabel(image.category) }}</span>
+          </div>
+          <button type="button" aria-label="移除图片" :disabled="sending" @click="removePendingImage(image.localId)">×</button>
+        </div>
+        <p v-if="hasPendingMedicalImage" class="medical-note">病例与医学影像识别仅供辅助，请以原始资料和医生意见为准。</p>
+      </div>
+      <div class="composer-row">
+        <button class="image-btn" type="button" :disabled="sending || uploadingImage || pendingImages.length >= 3" aria-label="选择图片" @click="imageInput?.click()">
+          <svg viewBox="0 0 24 24" aria-hidden="true"><path d="M4 5.5A2.5 2.5 0 0 1 6.5 3h11A2.5 2.5 0 0 1 20 5.5v13a2.5 2.5 0 0 1-2.5 2.5h-11A2.5 2.5 0 0 1 4 18.5v-13Zm2 12.7 3.8-4.2 2.7 2.8 1.7-1.8 3.8 3.2V5.5a.5.5 0 0 0-.5-.5h-11a.5.5 0 0 0-.5.5v12.7Zm8.3-7.8a1.7 1.7 0 1 0 0-3.4 1.7 1.7 0 0 0 0 3.4Z"/></svg>
+        </button>
+        <input ref="imageInput" class="image-input" type="file" accept="image/jpeg,image/png,image/webp" multiple @change="selectImages" />
+        <textarea v-model="inputText" rows="1" :disabled="sending" placeholder="输入问题，或选择一张图片…" @keydown.enter.exact.prevent="sendMessage(inputText)"></textarea>
+        <button class="send-btn" type="submit" :disabled="sending || uploadingImage || (!inputText.trim() && !readyPendingImages.length)">发送</button>
+      </div>
     </form>
   </div>
 </template>
 
 <script setup lang="ts">
-import { computed, nextTick, onMounted, reactive, ref } from 'vue'
+import { computed, nextTick, onBeforeUnmount, onMounted, reactive, ref } from 'vue'
 import { useRoute, useRouter } from 'vue-router'
-import { getAvatarDetail, getPublicAvatar, streamAvatarChat, streamPublicAvatarChat, type ChatMessage } from '@/api'
+import {
+  getAvatarDetail,
+  getPublicAvatar,
+  streamAvatarChat,
+  streamPublicAvatarChat,
+  uploadAvatarChatImage,
+  uploadPublicAvatarChatImage,
+  type ChatAttachment,
+  type ChatMessage
+} from '@/api'
 import { useAvatarStore } from '@/store/avatar'
 import { useUserStore } from '@/store/user'
 import { renderChatMarkdownCharacters } from '@/utils/chat-markdown.js'
 
 type DisplayMessage = ChatMessage & {
-  source?: 'qa' | 'knowledge' | 'qwen' | 'public'
+  source?: 'qa' | 'knowledge' | 'vision' | 'qwen' | 'public'
   references?: Array<{ filename?: string }>
   characters?: string[]
+  attachments?: MessageAttachment[]
+}
+
+type MessageAttachment = {
+  id: string
+  filename: string
+  previewUrl: string
+  category?: ChatAttachment['category']
+  summary?: string
+  warning?: string
+}
+
+type PendingImage = MessageAttachment & {
+  localId: string
+  attachmentId?: string
+  status: 'uploading' | 'ready'
 }
 
 const route = useRoute()
@@ -103,8 +151,11 @@ const inputText = ref('')
 const sending = ref(false)
 const thinking = ref(false)
 const errorMessage = ref('')
-const lastQuestion = ref('')
+const lastRequest = ref<{ question: string; attachments: MessageAttachment[] } | null>(null)
 const messageList = ref<HTMLElement | null>(null)
+const imageInput = ref<HTMLInputElement | null>(null)
+const pendingImages = ref<PendingImage[]>([])
+const previewUrls = new Set<string>()
 let scrollFrame: number | null = null
 
 const userAvatarUrl = computed(() => userStore.user?.avatarUrl || store.userProfile?.avatarUrl || '')
@@ -115,14 +166,25 @@ const avatarStatus = computed(() => {
   if (status === 'training') return { tone: 'training', label: '知识训练中' }
   return { tone: 'active', label: '在线，随时可以和我聊聊' }
 })
+const readyPendingImages = computed(() => pendingImages.value.filter((image) => image.status === 'ready' && image.attachmentId))
+const uploadingImage = computed(() => pendingImages.value.some((image) => image.status === 'uploading'))
+const hasPendingMedicalImage = computed(() => readyPendingImages.value.some((image) => ['medical_document', 'medical_image'].includes(image.category || '')))
 
 const sourceLabels: Record<NonNullable<DisplayMessage['source']>, string> = {
   qa: '标准问答对',
   knowledge: '参考文件知识库',
+  vision: '图片理解',
   qwen: '智能回答',
   public: ''
 }
 const sourceLabel = (source?: DisplayMessage['source']) => source ? sourceLabels[source] : ''
+const categoryLabels: Record<ChatAttachment['category'], string> = {
+  general_image: '图片内容已识别',
+  document: '文档图片已识别',
+  medical_document: '病例文字已提取，请核对原文',
+  medical_image: '医学影像已作客观描述'
+}
+const categoryLabel = (category?: ChatAttachment['category']) => category ? categoryLabels[category] : '图片内容已识别'
 
 const scrollToBottom = async () => {
   await nextTick()
@@ -214,20 +276,90 @@ const loadAvatar = async () => {
   document.title = avatar.value?.displayName || avatar.value?.name || '会会数字分身'
 }
 
-const sendMessage = async (value: string) => {
-  const question = value.trim()
-  if (!question || sending.value) return
-  lastQuestion.value = question
+const removePendingImage = (localId: string) => {
+  const target = pendingImages.value.find((image) => image.localId === localId)
+  if (target) {
+    URL.revokeObjectURL(target.previewUrl)
+    previewUrls.delete(target.previewUrl)
+  }
+  pendingImages.value = pendingImages.value.filter((image) => image.localId !== localId)
+}
+
+const selectImages = async (event: Event) => {
+  const input = event.target as HTMLInputElement
+  const slots = Math.max(0, 3 - pendingImages.value.length)
+  const files = Array.from(input.files || []).slice(0, slots)
+  input.value = ''
+  for (const file of files) {
+    if (!['image/jpeg', 'image/png', 'image/webp'].includes(file.type)) {
+      errorMessage.value = '仅支持 JPG、PNG、WebP 图片'
+      continue
+    }
+    if (file.size > 8 * 1024 * 1024) {
+      errorMessage.value = '单张图片不能超过 8MB'
+      continue
+    }
+    const previewUrl = URL.createObjectURL(file)
+    previewUrls.add(previewUrl)
+    const localId = `local-${Date.now()}-${Math.random().toString(16).slice(2)}`
+    pendingImages.value.push({
+      id: localId,
+      localId,
+      filename: file.name,
+      previewUrl,
+      status: 'uploading'
+    })
+    errorMessage.value = ''
+    try {
+      const result = isPublic
+        ? await uploadPublicAvatarChatImage(shareToken, file)
+        : await uploadAvatarChatImage(avatarId.value, file)
+      const pending = pendingImages.value.find((image) => image.localId === localId)
+      if (!pending) continue
+      Object.assign(pending, {
+        id: result.id,
+        attachmentId: result.id,
+        status: 'ready',
+        category: result.category,
+        summary: result.summary,
+        warning: result.warning
+      })
+    } catch (error: any) {
+      removePendingImage(localId)
+      errorMessage.value = error?.response?.data?.detail || error?.message || '图片识别失败，请重新选择图片'
+    }
+  }
+}
+
+const sendMessage = async (value: string, retryAttachments?: MessageAttachment[]) => {
+  const selectedAttachments = retryAttachments || readyPendingImages.value.map((image) => ({
+    id: image.attachmentId || image.id,
+    filename: image.filename,
+    previewUrl: image.previewUrl,
+    category: image.category,
+    summary: image.summary,
+    warning: image.warning
+  }))
+  const question = value.trim() || (selectedAttachments.length ? '请帮我看看这张图片。' : '')
+  if (!question || sending.value || (!retryAttachments && uploadingImage.value)) return
+  const history = messages.value.slice(-10).map(({ role, content, attachments }) => ({
+    role,
+    content,
+    attachmentIds: attachments?.map((attachment) => attachment.id) || []
+  }))
+  lastRequest.value = { question, attachments: selectedAttachments }
   inputText.value = ''
   errorMessage.value = ''
-  messages.value.push({ role: 'user', content: question })
+  if (!retryAttachments) pendingImages.value = []
+  messages.value.push({ role: 'user', content: question, attachments: selectedAttachments })
   sending.value = true
   thinking.value = true
   await scrollToBottom()
   try {
     const payload = {
       message: question,
-      history: messages.value.slice(-10).map(({ role, content }) => ({ role, content }))
+      attachmentIds: selectedAttachments.map((attachment) => attachment.id),
+      history
     }
     const streamed = createStreamReply()
     const handlers = {
@@ -256,13 +388,17 @@ const sendMessage = async (value: string) => {
 }
 
 const retryLast = () => {
-  if (!lastQuestion.value || sending.value) return
-  const last = messages.value[messages.value.length - 1]
-  if (last?.role === 'user') messages.value.pop()
-  sendMessage(lastQuestion.value)
+  if (!lastRequest.value || sending.value) return
+  while (messages.value[messages.value.length - 1]?.role === 'assistant') messages.value.pop()
+  if (messages.value[messages.value.length - 1]?.role === 'user') messages.value.pop()
+  void sendMessage(lastRequest.value.question, lastRequest.value.attachments)
 }
 
 onMounted(loadAvatar)
+onBeforeUnmount(() => {
+  previewUrls.forEach((url) => URL.revokeObjectURL(url))
+  previewUrls.clear()
+})
 </script>
 
 <style scoped>
@@ -275,6 +411,7 @@ onMounted(loadAvatar)
 .avatar-heading h1 { margin: 0; font-size: 17px; }
 .online-state { display: flex; align-items: center; gap: 4px; margin-top: 3px; font-size: 11px; opacity: .9; }.online-state i { width: 7px; height: 7px; border-radius: 50%; background: #86EFAC; box-shadow: 0 0 0 2px rgba(255,255,255,.22); }.online-state.training i { background: #FDE68A; }.online-state.inactive i { background: #FDA4AF; }
 .message-list { min-height: 0; flex: 1 1 auto; width: min(760px, 100%); box-sizing: border-box; margin: 0 auto; padding: 24px 18px 120px; overflow-y: auto; overscroll-behavior: contain; }
+.chat-page.has-pending-images .message-list { padding-bottom: min(330px, 42vh); }
 .welcome-card { padding: 28px 20px; text-align: center; background: rgba(255,255,255,.72); border: 1px solid #FFE1C2; border-radius: 22px; box-shadow: 0 10px 28px rgba(181, 99, 35, .08); }
 .welcome-avatar { width: 64px; height: 64px; display: grid; place-items: center; margin: 0 auto 14px; overflow: hidden; border: 3px solid #fff; border-radius: 50%; background: #FFE4C7; box-shadow: 0 7px 16px rgba(181, 99, 35, .18); font-size: 32px; }.welcome-avatar img { width: 100%; height: 100%; object-fit: cover; }
 .welcome-card h2 { margin: 0 0 8px; font-size: 20px; }.welcome-description { max-width: 340px; margin: 0 auto; color: #8B6B58; font-size: 14px; line-height: 1.65; }
@@ -282,6 +419,10 @@ onMounted(loadAvatar)
 .message-row.user { justify-content: flex-end; }
 .message-avatar { flex: 0 0 auto; width: 42px; height: 42px; display: grid; place-items: center; overflow: hidden; border: 2px solid rgba(255,255,255,.9); border-radius: 14px; background: #FFE4C7; box-shadow: 0 3px 10px rgba(96, 52, 21, .12); font-size: 16px; }.message-avatar img { width: 100%; height: 100%; object-fit: cover; }.user-message-face { color: #fff; background: #D97706; }
 .message-column { max-width: min(78%, 560px); }
+.message-images { display: grid; grid-template-columns: repeat(2, minmax(0, 150px)); gap: 8px; margin-bottom: 8px; }
+.message-image-card { margin: 0; overflow: hidden; border: 1px solid #F4D4B8; border-radius: 14px; background: #fff; box-shadow: 0 4px 14px rgba(96, 52, 21, .08); }
+.message-image-card img { display: block; width: 100%; max-height: 210px; object-fit: cover; }
+.message-image-card figcaption { padding: 7px 9px; color: #8A5A3B; background: #FFF6ED; font-size: 10px; line-height: 1.45; }
 .message-bubble { padding: 12px 14px; white-space: pre-wrap; line-height: 1.6; font-size: 15px; border-radius: 4px 16px 16px 16px; background: white; box-shadow: 0 3px 12px rgba(96, 52, 21, .07); }
 .message-bubble.streaming::after { content: ''; display: inline-block; width: 2px; height: 1.05em; margin-left: 3px; vertical-align: -0.16em; background: currentColor; animation: type-cursor .75s step-end infinite; }
 .typing-character { display: inline-block; animation: character-in .24s cubic-bezier(.2,.72,.25,1) both; }.typing-character.newline { display: block; height: 0; }
@@ -298,7 +439,27 @@ onMounted(loadAvatar)
 @keyframes type-cursor { 50% { opacity: 0; } }
 @keyframes character-in { from { opacity: 0; transform: translateY(3px); } to { opacity: 1; transform: translateY(0); } }
 .chat-error { margin: 4px auto; color: #B42318; font-size: 13px; }.chat-error button { border: 0; background: none; color: #C15F18; cursor: pointer; text-decoration: underline; }
-.composer { position: fixed; left: 0; right: 0; bottom: 0; display: flex; gap: 10px; padding: 12px max(18px, calc((100vw - 760px) / 2 + 18px)); background: rgba(255,255,255,.92); border-top: 1px solid #F4DCC7; backdrop-filter: blur(12px); }
+.composer { position: fixed; left: 0; right: 0; bottom: 0; display: flex; flex-direction: column; gap: 9px; padding: 10px max(18px, calc((100vw - 760px) / 2 + 18px)) 12px; background: rgba(255,255,255,.94); border-top: 1px solid #F4DCC7; backdrop-filter: blur(14px); }
+.composer-row { display: flex; align-items: flex-end; gap: 9px; }
 .composer textarea { flex: 1; resize: none; min-height: 22px; max-height: 100px; padding: 11px 13px; border: 1px solid #EED8C5; border-radius: 13px; font: inherit; color: #3B2417; outline: none; }.composer textarea:focus { border-color: #F97316; }
+.image-input { display: none; }
+.image-btn { flex: 0 0 auto; width: 44px; height: 44px; display: grid; place-items: center; border: 1px solid #EED8C5; border-radius: 13px; color: #C65A11; background: #FFF8F1; cursor: pointer; }
+.image-btn svg { width: 22px; height: 22px; fill: currentColor; }
+.image-btn:disabled { opacity: .4; cursor: not-allowed; }
+.pending-images { display: grid; gap: 7px; }
+.pending-image { display: grid; grid-template-columns: 48px minmax(0, 1fr) 30px; align-items: center; gap: 9px; min-height: 48px; padding: 6px 8px; border: 1px solid #F1D4BB; border-radius: 14px; background: #FFF9F3; }
+.pending-image img { width: 48px; height: 48px; object-fit: cover; border-radius: 10px; }
+.pending-image-copy { min-width: 0; display: flex; flex-direction: column; gap: 2px; }
+.pending-image-copy strong { overflow: hidden; color: #4B2B19; font-size: 12px; text-overflow: ellipsis; white-space: nowrap; }
+.pending-image-copy span { color: #9A7159; font-size: 10px; }
+.pending-image.uploading strong::after { content: ''; display: inline-block; width: 7px; height: 7px; margin-left: 7px; border: 2px solid #F6B889; border-top-color: #F97316; border-radius: 50%; animation: image-spin .7s linear infinite; }
+.pending-image > button { width: 28px; height: 28px; border: 0; border-radius: 9px; color: #9A7159; background: #F8E8D9; font-size: 19px; cursor: pointer; }
+.medical-note { margin: 0; padding: 0 2px; color: #9A5A2E; font-size: 10px; line-height: 1.45; }
 .send-btn { align-self: flex-end; padding: 11px 18px; border: 0; border-radius: 12px; color: white; background: #F97316; cursor: pointer; }.send-btn:disabled { opacity: .45; cursor: not-allowed; }
+@keyframes image-spin { to { transform: rotate(360deg); } }
+@media (max-width: 520px) {
+  .message-images { grid-template-columns: minmax(0, 220px); }
+  .message-column { max-width: 80%; }
+  .send-btn { padding-inline: 14px; }
+}
 </style>

digital-avatar-app/src/views/AvatarEdit.vue

@@ -1,12 +1,10 @@
 <template>
   <div class="edit-avatar-page">
     <!-- 顶部导航 -->
-    <header class="page-header">
+    <header v-if="!isEmbedded" class="page-header">
       <button class="back-btn" @click="goBack">‹</button>
       <h1 class="page-title">分身微调</h1>
-      <button class="save-btn" :disabled="loading || saving || uploadingPhoto" @click="saveChanges">
-        {{ saving ? '保存中...' : '保存' }}
-      </button>
+      <span class="header-spacer" aria-hidden="true"></span>
     </header>
 
     <div v-if="loading" class="status-banner">加载中...</div>
@@ -159,6 +157,13 @@
         {{ deleting ? '删除中...' : '删除数字分身' }}
       </button>
     </section>
+
+    <footer class="edit-action-bar">
+      <button class="action-back-btn" type="button" :disabled="saving" @click="goBack">返回</button>
+      <button class="action-save-btn" type="button" :disabled="loading || saving || uploadingPhoto" @click="saveChanges">
+        {{ saving ? '保存中...' : '保存修改' }}
+      </button>
+    </footer>
   </div>
 </template>
 
@@ -168,11 +173,13 @@ import { useRoute, useRouter } from 'vue-router'
 import { deleteAvatar as apiDeleteAvatar, getAvatarDetail, updateAvatar, uploadAvatarPhoto } from '@/api'
 import { useAvatarStore } from '@/store/avatar'
 import { buildAvatarUpdatePayload, normalizeAvatarEditForm } from '@/utils/avatar-page-data.js'
+import { isHuihuiEmbeddedMode } from '@/utils/embed-mode'
 
 const router = useRouter()
 const route = useRoute()
 const avatarStore = useAvatarStore()
 const avatarId = route.params.id as string
+const isEmbedded = isHuihuiEmbeddedMode()
 
 // 表单数据
 const formData = reactive({
@@ -288,7 +295,7 @@ onMounted(async () => {
 .edit-avatar-page {
   min-height: 100vh;
   background: #F8F9FA;
-  padding-bottom: 40px;
+  padding-bottom: calc(104px + env(safe-area-inset-bottom));
 }
 
 /* 顶部导航 */
@@ -331,16 +338,7 @@ onMounted(async () => {
   color: #B91C1C;
 }
 
-.save-btn {
-  background: #F97316;
-  color: white;
-  border: none;
-  padding: 8px 20px;
-  border-radius: 8px;
-  font-size: 14px;
-  font-weight: 600;
-  cursor: pointer;
-}
+.header-spacer { width: 40px; }
 
 /* 头像上传 */
 .photo-section {
@@ -580,4 +578,47 @@ onMounted(async () => {
   background: #EF4444;
   color: white;
 }
+
+.edit-action-bar {
+  position: fixed;
+  z-index: 30;
+  left: 0;
+  right: 0;
+  bottom: 0;
+  display: flex;
+  gap: 12px;
+  padding: 12px 20px calc(14px + env(safe-area-inset-bottom));
+  border-top: 1px solid rgba(229, 231, 235, .9);
+  background: rgba(248, 249, 250, .96);
+  box-shadow: 0 -8px 24px rgba(56, 38, 24, .06);
+  backdrop-filter: blur(12px);
+}
+
+.action-back-btn,
+.action-save-btn {
+  height: 48px;
+  border-radius: 14px;
+  font-size: 15px;
+  font-weight: 600;
+  cursor: pointer;
+}
+
+.action-back-btn {
+  flex: 0 0 104px;
+  border: 1px solid #E4E0DC;
+  color: #655E58;
+  background: #fff;
+}
+
+.action-save-btn {
+  min-width: 0;
+  flex: 1;
+  border: 0;
+  color: #fff;
+  background: linear-gradient(105deg, #F79A38, #F97316);
+  box-shadow: 0 8px 18px rgba(249, 115, 22, .18);
+}
+
+.action-back-btn:disabled,
+.action-save-btn:disabled { opacity: .6; cursor: not-allowed; }
 </style>

digital-avatar-app/src/views/AvatarManage.vue

@@ -1,15 +1,11 @@
 <template>
   <div class="avatar-manage-page">
     <!-- 顶部导航 -->
-    <header class="page-header">
+    <header v-if="!isEmbedded" class="page-header">
       <div class="header-left">
         <button class="back-btn" @click="goBack">‹</button>
         <h1 class="page-title">数字分身管理</h1>
       </div>
-      <!-- 右上角创建入口 -->
-      <div class="header-right">
-        <button class="icon-btn" @click="goCreate" title="创建数字分身">➕</button>
-      </div>
     </header>
 
     <!-- 用户资料头（会会登录账号的头像 / 昵称） -->
@@ -24,12 +20,13 @@
       </div>
     </section>
 
-    <!-- Token 余额条 -->
-    <section class="token-section">
+    <!-- 积分余额条：暂时隐藏，保留完整实现便于后续恢复。 -->
+    <section v-if="SHOW_POINTS_BALANCE_CARD" class="token-section">
       <div class="token-card">
         <div class="token-info">
-          <span class="token-label">Token 余额</span>
+          <span class="token-label">积分余额</span>
           <span class="token-amount">{{ tokenBalance.toLocaleString() }}</span>
+          <span class="token-used">累计使用 {{ tokenConsumed.toLocaleString() }}</span>
         </div>
         <button class="recharge-btn" @click="goToRecharge">充值</button>
       </div>
@@ -38,12 +35,20 @@
     <!-- 数字分身列表（只放分身相关） -->
     <section class="avatar-list-section">
       <div class="section-head">
+        <div class="section-heading-copy">
           <h3 class="section-title">我的数字分身</h3>
           <span class="count-badge">{{ avatars.length }}</span>
         </div>
+        <button class="section-create-btn" type="button" @click="goCreate">
+          <span aria-hidden="true">＋</span> 添加分身
+        </button>
+      </div>
 
       <div v-if="avatars.length" class="avatar-list">
         <div class="avatar-card" v-for="a in avatars" :key="a.id">
+          <button class="card-edit" @click="goToEdit(a.id)" :aria-label="`编辑${a.displayName || a.name}`">
+            <span aria-hidden="true">✎</span> 编辑
+          </button>
           <div class="avatar-card-main">
             <div class="avatar-photo">
               <img v-if="a.photoUrl" :src="a.photoUrl" alt="" referrerpolicy="no-referrer" class="avatar-img" />
@@ -52,13 +57,14 @@
             <div class="avatar-details">
               <div class="avatar-name-row"><h2 class="avatar-name">{{ a.displayName || a.name }}</h2><span class="avatar-status"><i class="status-dot" :class="a.status"></i>{{ statusText(a.status) }}</span></div>
               <p class="avatar-desc">{{ a.description || '暂无描述' }}</p>
+              <span class="avatar-token-usage">累计使用 {{ avatarTokenUsage(a.id).toLocaleString() }} 积分</span>
             </div>
           </div>
           <div class="avatar-actions">
             <button class="chat-link" @click="goToChat(a.id)"><span>💬</span> 对话</button>
+            <button class="knowledge-link" @click="goToKnowledge(a.id)"><span>📚</span> 知识库</button>
+            <button class="authorization-link" @click="goToAuthorization(a.id)"><span>🔑</span> 授权</button>
             <button class="share-link" @click="shareAvatar(a)"><span>↗</span> 分享</button>
-            <button class="edit-link" @click="goToEdit(a.id)">编辑</button>
-            <button class="del-link" @click="askDelete(a)" aria-label="删除分身">删除</button>
           </div>
         </div>
       </div>
@@ -72,88 +78,7 @@
 
     <p v-if="shareToast" class="share-toast">{{ shareToast }}</p>
 
-    <!-- 分身工具入口 -->
-    <section class="tools-section">
-      <h3 class="section-title">分身工具</h3>
-      <div class="tools-grid">
-        <div class="tool-card" @click="goToKnowledge">
-          <div class="tool-icon">📚</div>
-          <div class="tool-info">
-            <span class="tool-name">知识库管理</span>
-            <span class="tool-desc">上传文档与标准问答</span>
-          </div>
-          <span class="tool-arrow">›</span>
-        </div>
-        <div class="tool-card" @click="goToAvatarCard">
-          <div class="tool-icon">🪪</div>
-          <div class="tool-info">
-            <span class="tool-name">分身名片</span>
-            <span class="tool-desc">生成并分享名片</span>
-          </div>
-          <span class="tool-arrow">›</span>
-        </div>
-        <div class="tool-card" @click="goToAvatarContacts">
-          <div class="tool-icon">🤝</div>
-          <div class="tool-info">
-            <span class="tool-name">分身人脉</span>
-            <span class="tool-desc">管理社交关系</span>
-          </div>
-          <span class="tool-arrow">›</span>
-        </div>
-        <div class="tool-card" @click="goToMyProjects">
-          <div class="tool-icon">📁</div>
-          <div class="tool-info">
-            <span class="tool-name">我的项目</span>
-            <span class="tool-desc">查看参与项目</span>
-          </div>
-          <span class="tool-arrow">›</span>
-        </div>
-        <div class="tool-card" @click="goToCreateOrg">
-          <div class="tool-icon">🏢</div>
-          <div class="tool-info">
-            <span class="tool-name">创建组织</span>
-            <span class="tool-desc">新建组织团队</span>
-          </div>
-          <span class="tool-arrow">›</span>
-        </div>
-      </div>
-    </section>
-
-    <!-- 分身动态列表 -->
-    <section class="activities-section">
-      <h3 class="section-title">分身动态</h3>
-      <div class="activity-list" v-if="activities.length > 0">
-        <div class="activity-item" v-for="activity in activities" :key="activity.id">
-          <div class="activity-icon" :class="activity.type">{{ activityIcon(activity.type) }}</div>
-          <div class="activity-content">
-            <p class="activity-text">{{ activity.text }}</p>
-            <span class="activity-time">{{ formatTime(activity.createdAt) }}</span>
-          </div>
-        </div>
-      </div>
-      <div class="empty-state" v-else>
-        <span class="empty-icon">📭</span>
-        <p class="empty-text">暂无动态</p>
-      </div>
-    </section>
-
-    <!-- 删除确认弹窗 -->
-    <div v-if="showDelete" class="modal-mask" @click.self="cancelDelete">
-      <div class="modal">
-        <div class="modal-icon">⚠️</div>
-        <h3 class="modal-title">删除数字分身</h3>
-        <p class="modal-text">
-          确认删除「{{ pendingDelete?.displayName || pendingDelete?.name }}」？<br />
-          其知识库、问答对、授权等关联数据将一并清除，且<b>不可恢复</b>。
-        </p>
-        <div class="modal-actions">
-          <button class="modal-cancel" @click="cancelDelete">取消</button>
-          <button class="modal-confirm" :disabled="deleting" @click="confirmDelete">
-            {{ deleting ? '删除中...' : '确认删除' }}
-          </button>
-        </div>
-      </div>
-    </div>
+    <!-- 分身工具和分身动态暂不展示，入口统一收拢到每张分身卡片。 -->
   </div>
 </template>
 
@@ -163,30 +88,27 @@ import { useRouter } from 'vue-router'
 import { useAvatarStore } from '@/store/avatar'
 import { useUserStore } from '@/store/user'
 import { createAvatarShareLink } from '@/api'
+import { isHuihuiEmbeddedMode } from '@/utils/embed-mode'
 
 const router = useRouter()
 const avatarStore = useAvatarStore()
 const userStore = useUserStore()
+const isEmbedded = isHuihuiEmbeddedMode()
+
+// 临时产品开关：余额卡片代码保留，后续改为 true 即可恢复展示。
+const SHOW_POINTS_BALANCE_CARD = false
 
 // 当前登录会会用户的资料（头像 / 昵称）
 const me = computed(() => userStore.user)
 
 // 状态（来自 store / 后端）
 const tokenBalance = computed(() => avatarStore.tokenBalance)
+const tokenConsumed = computed(() => avatarStore.tokenConsumed)
 const avatars = computed(() => avatarStore.avatars)
+const avatarTokenUsage = (id: string) => avatarStore.tokenUsageByAvatar[id] || 0
 
-// 删除确认弹窗状态
-const showDelete = ref(false)
-const pendingDelete = ref<any>(null)
-const deleting = ref(false)
 const shareToast = ref('')
 
-const activities = ref<Array<{ id: string; type: string; text: string; createdAt: string }>>([
-  { id: '1', type: 'create', text: '数字分身创建成功', createdAt: new Date(Date.now() - 86400000).toISOString() },
-  { id: '2', type: 'edit', text: '更新了分身描述', createdAt: new Date(Date.now() - 43200000).toISOString() },
-  { id: '3', type: 'authorize', text: '授权微信小程序访问', createdAt: new Date(Date.now() - 3600000).toISOString() }
-])
-
 // 状态文本
 const statusText = (status: string) => {
   const map: Record<string, string> = {
@@ -197,55 +119,6 @@ const statusText = (status: string) => {
   return map[status] || status
 }
 
-// 活动图标
-const activityIcon = (type: string) => {
-  const map: Record<string, string> = {
-    'create': '✨',
-    'edit': '✏️',
-    'authorize': '🔑',
-    'interact': '💬'
-  }
-  return map[type] || '📌'
-}
-
-// 格式化时间
-const formatTime = (time: string) => {
-  const date = new Date(time)
-  const now = new Date()
-  const diff = now.getTime() - date.getTime()
-  const minutes = Math.floor(diff / 60000)
-  const hours = Math.floor(diff / 3600000)
-  const days = Math.floor(diff / 86400000)
-
-  if (minutes < 60) return `${minutes}分钟前`
-  if (hours < 24) return `${hours}小时前`
-  return `${days}天前`
-}
-
-// 删除流程
-const askDelete = (a: any) => {
-  pendingDelete.value = a
-  showDelete.value = true
-}
-const cancelDelete = () => {
-  if (deleting.value) return
-  showDelete.value = false
-  pendingDelete.value = null
-}
-const confirmDelete = async () => {
-  if (!pendingDelete.value || deleting.value) return
-  deleting.value = true
-  try {
-    await avatarStore.removeAvatar(pendingDelete.value.id)
-    showDelete.value = false
-    pendingDelete.value = null
-  } catch (e: any) {
-    alert(e?.message || '删除失败，请稍后重试')
-  } finally {
-    deleting.value = false
-  }
-}
-
 // 导航
 const goBack = () => {
   router.back()
@@ -259,8 +132,12 @@ const goCreate = () => {
   router.push('/avatar/create')
 }
 
-const goToKnowledge = () => {
-  router.push('/knowledge')
+const goToKnowledge = (id: string) => {
+  router.push({ name: 'AvatarKnowledgeManage', params: { avatarId: id } })
+}
+
+const goToAuthorization = (id: string) => {
+  router.push({ name: 'AvatarAuthorizationManage', params: { avatarId: id } })
 }
 
 const goToEdit = (id: string) => {
@@ -303,26 +180,11 @@ const shareAvatar = async (avatar: any) => {
   }
 }
 
-const goToAvatarCard = () => {
-  router.push('/avatar/card')
-}
-
-const goToAvatarContacts = () => {
-  router.push('/avatar/contacts')
-}
-
-const goToMyProjects = () => {
-  router.push('/avatar/projects')
-}
-
-const goToCreateOrg = () => {
-  router.push('/avatar/org/create')
-}
-
 onMounted(() => {
   userStore.loadFromStorage()
   avatarStore.loadAvatars()
   avatarStore.loadTokenBalance()
+  avatarStore.loadTokenUsage()
 })
 </script>
 
@@ -330,7 +192,7 @@ onMounted(() => {
 .avatar-manage-page {
   min-height: 100vh;
   background: #F8F9FA;
-  padding-bottom: 80px;
+  padding-bottom: calc(28px + env(safe-area-inset-bottom));
 }
 
 /* 顶部导航 */
@@ -433,7 +295,7 @@ onMounted(() => {
   background: rgba(255, 255, 255, 0.3);
 }
 
-/* Token 余额条 */
+/* 积分余额条 */
 .token-section {
   padding: 16px 20px;
 }
@@ -465,6 +327,12 @@ onMounted(() => {
   color: #F97316;
 }
 
+.token-used {
+  margin-top: 3px;
+  color: #A0A5B4;
+  font-size: 11px;
+}
+
 .recharge-btn {
   padding: 8px 16px;
   background: #F97316;
@@ -489,10 +357,36 @@ onMounted(() => {
 .section-head {
   display: flex;
   align-items: center;
-  gap: 8px;
+  justify-content: space-between;
+  gap: 12px;
   margin: 8px 0 12px;
 }
 
+.section-heading-copy {
+  min-width: 0;
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.section-create-btn {
+  flex: 0 0 auto;
+  display: inline-flex;
+  align-items: center;
+  gap: 3px;
+  padding: 8px 12px;
+  border: 1px solid #FED7B5;
+  border-radius: 999px;
+  color: #E9650C;
+  background: #FFF7ED;
+  font-size: 12px;
+  font-weight: 650;
+  cursor: pointer;
+}
+
+.section-create-btn span { font-size: 17px; line-height: 1; }
+.section-create-btn:active { background: #FFEDD5; }
+
 .section-title {
   font-size: 16px;
   font-weight: 600;
@@ -516,13 +410,32 @@ onMounted(() => {
 }
 
 .avatar-card {
+  position: relative;
   padding: 16px;
   background: white;
   border: 1px solid #F4E5D9;
   border-radius: 18px;
   box-shadow: 0 8px 22px rgba(112, 62, 22, .07);
 }
-.avatar-card-main { display: flex; align-items: center; gap: 14px; }
+.avatar-card-main { display: flex; align-items: center; gap: 14px; padding: 4px 68px 2px 0; }
+
+.card-edit {
+  position: absolute;
+  top: 12px;
+  right: 12px;
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  padding: 7px 10px;
+  border: 1px solid #E9E2DC;
+  border-radius: 999px;
+  color: #6B625C;
+  background: #FAF8F6;
+  font-size: 12px;
+  cursor: pointer;
+}
+
+.card-edit:active { background: #F2EDE8; }
 
 .avatar-photo {
   width: 66px;
@@ -570,6 +483,13 @@ onMounted(() => {
   white-space: nowrap;
 }
 
+.avatar-token-usage {
+  display: inline-block;
+  margin-top: 5px;
+  color: #A0A5B4;
+  font-size: 10px;
+}
+
 .avatar-status {
   display: inline-flex;
   align-items: center;
@@ -598,56 +518,39 @@ onMounted(() => {
 }
 
 .avatar-actions {
-  display: flex;
-  align-items: center;
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
   gap: 8px;
   margin-top: 16px;
 }
 
+.avatar-actions button {
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  gap: 4px;
+  min-width: 0;
+  min-height: 58px;
+  padding: 8px 3px;
+  border-radius: 12px;
+  font-size: 12px;
+  font-weight: 600;
+  cursor: pointer;
+}
+
+.avatar-actions button span { font-size: 17px; line-height: 1; }
+
 .chat-link {
-  flex: 1;
-  padding: 10px 8px;
   background: linear-gradient(135deg, #F97316, #FB923C);
   color: #fff;
   border: none;
-  border-radius: 10px;
-  font-size: 13px;
-  cursor: pointer;
 }
-.chat-link span, .share-link span { margin-right: 3px; }
-.share-link { flex: 1; padding: 10px 8px; border: 1px solid #FFD5AF; border-radius: 10px; color: #C15F18; background: #FFF8F1; font-size: 13px; cursor: pointer; }
+.knowledge-link { border: 1px solid #F5D7B8; color: #A85418; background: #FFF8F1; }
+.authorization-link { border: 1px solid #DCE5F2; color: #47627D; background: #F6F9FC; }
+.share-link { border: 1px solid #E5E1DD; color: #675E58; background: #FAF8F6; }
 
-.edit-link {
-  padding: 10px 10px;
-  background: #F3F4F6;
-  color: #6B7280;
-  border: none;
-  border-radius: 8px;
-  font-size: 13px;
-  cursor: pointer;
-  transition: background 0.2s;
-}
-
-.edit-link:hover {
-  background: #E5E7EB;
-}
-
-.del-link {
-  padding: 10px 2px;
-  background: transparent;
-  color: #B6BCC8;
-  border: none;
-  border-radius: 8px;
-  font-size: 13px;
-  cursor: pointer;
-  transition: background 0.2s;
-}
-
-.share-toast { position: fixed; left: 50%; bottom: 92px; z-index: 300; max-width: calc(100vw - 48px); transform: translateX(-50%); padding: 10px 14px; border-radius: 10px; color: white; background: rgba(39, 32, 28, .88); font-size: 13px; text-align: center; }
-
-.del-link:hover {
-  background: #FEE2E2;
-}
+.share-toast { position: fixed; left: 50%; bottom: 24px; z-index: 300; max-width: calc(100vw - 48px); transform: translateX(-50%); padding: 10px 14px; border-radius: 10px; color: white; background: rgba(39, 32, 28, .88); font-size: 13px; text-align: center; }
 
 /* 空状态 */
 .empty-state {
@@ -684,197 +587,4 @@ onMounted(() => {
   box-shadow: 0 4px 12px rgba(249, 115, 22, 0.3);
 }
 
-/* 分身工具入口 */
-.tools-section {
-  padding: 8px 20px 16px;
-}
-
-.tools-grid {
-  display: flex;
-  flex-direction: column;
-  gap: 10px;
-}
-
-.tool-card {
-  display: flex;
-  align-items: center;
-  gap: 12px;
-  padding: 14px 16px;
-  background: white;
-  border-radius: 12px;
-  box-shadow: 0 2px 8px rgba(0, 0, 0, 0.05);
-  cursor: pointer;
-  transition: transform 0.2s, box-shadow 0.2s;
-}
-
-.tool-card:hover {
-  transform: translateY(-2px);
-  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.1);
-}
-
-.tool-icon {
-  font-size: 24px;
-  flex-shrink: 0;
-}
-
-.tool-info {
-  flex: 1;
-  display: flex;
-  flex-direction: column;
-  gap: 2px;
-  min-width: 0;
-}
-
-.tool-name {
-  font-size: 14px;
-  font-weight: 600;
-  color: #18191C;
-}
-
-.tool-desc {
-  font-size: 12px;
-  color: #9398AE;
-}
-
-.tool-arrow {
-  font-size: 18px;
-  color: #C9CDD2;
-  flex-shrink: 0;
-}
-
-/* 分身动态列表 */
-.activities-section {
-  padding: 0 20px;
-}
-
-.activity-list {
-  display: flex;
-  flex-direction: column;
-  gap: 10px;
-}
-
-.activity-item {
-  display: flex;
-  align-items: flex-start;
-  gap: 12px;
-  padding: 14px 16px;
-  background: white;
-  border-radius: 12px;
-  box-shadow: 0 2px 8px rgba(0, 0, 0, 0.05);
-}
-
-.activity-icon {
-  font-size: 20px;
-  flex-shrink: 0;
-  width: 36px;
-  height: 36px;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  border-radius: 10px;
-  background: #FFF0E6;
-}
-
-.activity-content {
-  flex: 1;
-  min-width: 0;
-}
-
-.activity-text {
-  font-size: 14px;
-  color: #18191C;
-  margin: 0 0 4px;
-  line-height: 1.4;
-}
-
-.activity-time {
-  font-size: 12px;
-  color: #9398AE;
-}
-
-/* 删除确认弹窗 */
-.modal-mask {
-  position: fixed;
-  inset: 0;
-  background: rgba(0, 0, 0, 0.45);
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  padding: 24px;
-  z-index: 50;
-  animation: fade 0.2s ease;
-}
-
-@keyframes fade { from { opacity: 0; } to { opacity: 1; } }
-
-.modal {
-  width: 100%;
-  max-width: 320px;
-  background: white;
-  border-radius: 18px;
-  padding: 24px 22px 18px;
-  text-align: center;
-  box-shadow: 0 20px 50px rgba(0, 0, 0, 0.25);
-  animation: pop 0.25s cubic-bezier(0.16, 1, 0.3, 1);
-}
-
-@keyframes pop { from { opacity: 0; transform: scale(0.94); } to { opacity: 1; transform: none; } }
-
-.modal-icon {
-  font-size: 36px;
-  margin-bottom: 8px;
-}
-
-.modal-title {
-  font-size: 17px;
-  font-weight: 700;
-  color: #18191C;
-  margin: 0 0 10px;
-}
-
-.modal-text {
-  font-size: 13px;
-  line-height: 1.6;
-  color: #6B7280;
-  margin: 0 0 20px;
-}
-
-.modal-actions {
-  display: flex;
-  gap: 12px;
-}
-
-.modal-cancel {
-  flex: 1;
-  padding: 12px;
-  background: #F3F4F6;
-  color: #6B7280;
-  border: none;
-  border-radius: 12px;
-  font-size: 15px;
-  font-weight: 600;
-  cursor: pointer;
-}
-
-.modal-confirm {
-  flex: 1;
-  padding: 12px;
-  background: #EF4444;
-  color: white;
-  border: none;
-  border-radius: 12px;
-  font-size: 15px;
-  font-weight: 600;
-  cursor: pointer;
-  transition: opacity 0.2s;
-}
-
-.modal-confirm:hover {
-  opacity: 0.92;
-}
-
-.modal-confirm:disabled {
-  opacity: 0.6;
-  cursor: not-allowed;
-}
 </style>

digital-avatar-app/src/views/KnowledgeManage.vue

@@ -1,7 +1,7 @@
 <template>
-  <div class="knowledge-page">
+  <div class="knowledge-page" :class="{ embedded: isEmbedded }">
     <!-- 顶部导航 -->
-    <header class="page-header">
+    <header v-if="!isEmbedded" class="page-header">
       <div class="header-left">
         <button class="back-btn" @click="goBack">‹</button>
         <h1 class="page-title">知识库管理</h1>
@@ -15,7 +15,7 @@
 
     <template v-else>
       <div class="tab-switcher" role="tablist" aria-label="知识库类型">
-        <button class="tab-btn" :class="{ active: activeTab === 'docs' }" role="tab" :aria-selected="activeTab === 'docs'" @click="activeTab = 'docs'">文档知识库 <b>{{ docs.length }}</b></button>
+        <button class="tab-btn" :class="{ active: activeTab === 'docs' }" role="tab" :aria-selected="activeTab === 'docs'" @click="activeTab = 'docs'">文档知识库 <b>{{ displayDocs.length }}</b></button>
         <button class="tab-btn" :class="{ active: activeTab === 'qa' }" role="tab" :aria-selected="activeTab === 'qa'" @click="activeTab = 'qa'">标准问答对 <b>{{ qaPairs.length }}</b></button>
       </div>
 
@@ -23,26 +23,37 @@
         <div class="upload-section">
           <div class="upload-zone" :class="{ 'drag-over': dragOver }" @click="triggerFile" @dragover.prevent="dragOver = true" @dragleave.prevent="dragOver = false" @drop.prevent="onDrop">
             <div class="upload-icon">📥</div>
-            <p class="upload-title">拖拽文件到此处，或<span class="upload-link">点击上传</span></p>
+            <p class="upload-title"><span class="upload-link">点击上传</span></p>
             <p class="upload-hint">支持 MD / TXT / PDF / DOC / DOCX / XLSX，上传后自动向量化</p>
-            <input ref="fileInput" type="file" accept=".md,.txt,.pdf,.doc,.docx,.xlsx" class="hidden-input" @change="onFileChange" />
+            <input ref="fileInput" type="file" multiple accept=".md,.txt,.pdf,.doc,.docx,.xlsx" class="hidden-input" @change="onFileChange" />
           </div>
-          <p v-if="uploading" class="uploading-text">上传并向量化中…</p>
+          <p v-if="uploading" class="uploading-text">{{ pendingUploads.length }} 个文件正在上传</p>
           <p v-if="uploadError" class="error-text">{{ uploadError }}</p>
         </div>
 
-        <div v-if="docs.length" class="mobile-card-list">
-          <article v-for="doc in docs" :key="doc.id" class="knowledge-card">
+        <div v-if="displayDocs.length" class="mobile-card-list">
+          <article v-for="doc in displayDocs" :key="doc.id" class="knowledge-card document-card">
             <div class="card-icon">{{ fileEmoji(doc.fileType) }}</div>
             <div class="card-content">
               <div class="card-title-row">
                 <strong>{{ doc.filename }}</strong>
-                <span class="status-pill" :class="{ pending: !doc.vectorized && doc.filePresent !== false, missing: doc.filePresent === false }">{{ doc.filePresent === false ? '文件缺失' : (doc.vectorized ? '已入库' : '处理中') }}</span>
+                <span class="status-pill" :class="documentState(doc).tone">{{ documentState(doc).label }}</span>
               </div>
               <p class="card-meta">{{ doc.fileType.toUpperCase() }} · {{ formatSize(doc.fileSize) }} · {{ formatDate(doc.createdAt) }}</p>
-              <p class="card-detail">{{ doc.filePresent === false ? '原文件不可用，请删除后重新上传' : (doc.vectorized ? `已切分 ${doc.chunkCount || 0} 段，可用于对话` : '正在解析并建立知识索引') }}</p>
+              <p class="card-detail">{{ documentState(doc).detail }}</p>
+              <div v-if="documentState(doc).progress !== undefined" class="progress-track" :aria-label="`${documentState(doc).label} ${documentState(doc).progress}%`">
+                <span class="progress-fill" :style="{ width: `${documentState(doc).progress}%` }"></span>
+              </div>
+            </div>
+            <div class="card-actions">
+              <button v-if="!doc.localUploading" class="card-delete" @click="removeDoc(doc.id)">{{ doc.localOnly ? '移除' : '删除' }}</button>
+            </div>
+            <div v-if="canRetryDoc(doc)" class="card-retry-area">
+              <span v-if="retryErrors[doc.id]" class="card-retry-error">{{ retryErrors[doc.id] }}</span>
+              <button class="card-retry" :disabled="retryingDocs[doc.id]" @click="retryDoc(doc)">
+                {{ retryingDocs[doc.id] ? '重新索引中…' : '重新索引' }}
+              </button>
             </div>
-            <button class="card-delete" @click="removeDoc(doc.id)">删除</button>
           </article>
         </div>
         <div v-else class="card-empty">📂 暂无文档，先上传一个知识文件</div>
@@ -78,14 +89,16 @@
 </template>
 
 <script setup lang="ts">
-import { ref, onMounted, computed } from 'vue'
-import { useRouter } from 'vue-router'
+import { ref, onMounted, onUnmounted, computed } from 'vue'
+import { useRoute, useRouter } from 'vue-router'
 import { useAvatarStore } from '@/store/avatar'
-import { pickAvatarId, unwrapListData } from '@/utils/avatar-page-data.js'
+import { pickScopedAvatarId, unwrapListData } from '@/utils/avatar-page-data.js'
+import { isHuihuiEmbeddedMode } from '@/utils/embed-mode'
 import {
   getKnowledgeDocs,
   uploadKnowledgeDoc,
   deleteKnowledgeDoc,
+  retryKnowledgeDoc,
   getQAPairs,
   deleteQAPair,
   searchKnowledge,
@@ -93,28 +106,80 @@ import {
 } from '@/api'
 
 const router = useRouter()
+const route = useRoute()
 const store = useAvatarStore()
+const isEmbedded = isHuihuiEmbeddedMode()
 
-const avatarId = computed(() => pickAvatarId(store.currentAvatarId, store.avatars))
+const avatarId = computed(() => pickScopedAvatarId(route.params.avatarId, store.currentAvatarId, store.avatars))
 const activeTab = ref<'docs' | 'qa'>('docs')
 
 const docs = ref<any[]>([])
+const pendingUploads = ref<any[]>([])
 const qaPairs = ref<any[]>([])
-const uploading = ref(false)
+const uploading = computed(() => pendingUploads.value.some((doc) => doc.localUploading))
 const uploadError = ref('')
 const dragOver = ref(false)
 const fileInput = ref<HTMLInputElement | null>(null)
+const retryingDocs = ref<Record<string, boolean>>({})
+const retryErrors = ref<Record<string, string>>({})
+let documentPollingTimer: ReturnType<typeof setInterval> | undefined
 
 const query = ref('')
 const searching = ref(false)
 const searched = ref(false)
 const searchResults = ref<any[]>([])
 
+const displayDocs = computed(() => [...pendingUploads.value, ...docs.value])
+
+const documentState = (doc: any) => {
+  if (doc.localUploading) {
+    return { tone: 'pending', label: '上传中', detail: `正在上传 ${doc.uploadProgress || 0}%`, progress: doc.uploadProgress || 0 }
+  }
+  if (doc.localOnly) {
+    return { tone: 'failed', label: '上传失败', detail: doc.errorMessage || '文件未上传成功，请移除后重试' }
+  }
+  if (doc.filePresent === false) {
+    return { tone: 'missing', label: '文件缺失', detail: '原文件不可用，请删除后重新上传' }
+  }
+  if (doc.vectorized) {
+    return { tone: 'ready', label: '已入库', detail: `已切分 ${doc.chunkCount || 0} 段，可用于对话` }
+  }
+  if (['uploaded', 'parsing'].includes(String(doc.status || '').toLowerCase())) {
+    const stage = String(doc.indexStage || 'queued').toLowerCase()
+    const labels: Record<string, string> = {
+      queued: '等待处理', extracting: '解析文档', chunking: '切分文本', embedding: '向量化中'
+    }
+    const progress = Math.max(0, Math.min(99, Number(doc.indexProgress || 0)))
+    return { tone: 'pending', label: labels[stage] || '处理中', detail: `${labels[stage] || '正在建立知识索引'} ${progress}%`, progress }
+  }
+  return { tone: 'failed', label: '处理失败', detail: doc.errorMessage || '未能建立知识索引，请重新索引或重新上传' }
+}
+
+const hasPendingDocuments = () => docs.value.some((doc) =>
+  ['uploaded', 'parsing'].includes(String(doc.status || '').toLowerCase())
+)
+
+const stopDocumentPolling = () => {
+  if (documentPollingTimer) {
+    clearInterval(documentPollingTimer)
+    documentPollingTimer = undefined
+  }
+}
+
+const startDocumentPolling = () => {
+  if (documentPollingTimer || !hasPendingDocuments()) return
+  documentPollingTimer = setInterval(async () => {
+    await loadDocs()
+    if (!hasPendingDocuments()) stopDocumentPolling()
+  }, 2000)
+}
+
 const loadDocs = async () => {
   if (!avatarId.value) return
   try {
     const res: any = await getKnowledgeDocs(avatarId.value)
     docs.value = unwrapListData(res)
+    startDocumentPolling()
   } catch (e) {
     console.error(e)
   }
@@ -133,40 +198,92 @@ const loadQA = async () => {
 const triggerFile = () => fileInput.value?.click()
 
 const onFileChange = (e: Event) => {
-  const f = (e.target as HTMLInputElement).files?.[0]
-  if (f) doUpload(f)
+  const files = Array.from((e.target as HTMLInputElement).files || [])
+  if (files.length) uploadFiles(files)
   ;(e.target as HTMLInputElement).value = ''
 }
 
 const onDrop = (e: DragEvent) => {
   dragOver.value = false
-  const f = e.dataTransfer?.files?.[0]
-  if (f) doUpload(f)
+  const files = Array.from(e.dataTransfer?.files || [])
+  if (files.length) uploadFiles(files)
 }
 
-const doUpload = async (file: File) => {
+const uploadFiles = (files: File[]) => {
   uploadError.value = ''
-  const ext = '.' + (file.name.split('.').pop() || '').toLowerCase()
-  if (!['.md', '.txt', '.pdf', '.doc', '.docx', '.xlsx'].includes(ext)) {
-    uploadError.value = `不支持的类型：${ext}，仅支持 md/txt/pdf/doc/docx/xlsx`
-    return
-  }
   if (!avatarId.value) {
     uploadError.value = '请先创建数字分身'
     return
   }
-  uploading.value = true
+  for (const file of files) {
+    const ext = '.' + (file.name.split('.').pop() || '').toLowerCase()
+    if (!['.md', '.txt', '.pdf', '.doc', '.docx', '.xlsx'].includes(ext)) {
+      uploadError.value = `不支持的类型：${ext}，仅支持 md/txt/pdf/doc/docx/xlsx`
+      continue
+    }
+    void uploadOne(file, ext)
+  }
+}
+
+const uploadOne = async (file: File, ext: string) => {
+  if (!avatarId.value) return
+  const localId = `upload-${Date.now()}-${Math.random().toString(16).slice(2)}`
+  const card = {
+    id: localId,
+    filename: file.name,
+    fileType: ext.slice(1),
+    fileSize: file.size,
+    createdAt: new Date().toISOString(),
+    localUploading: true,
+    localOnly: true,
+    uploadProgress: 0,
+    errorMessage: ''
+  }
+  pendingUploads.value.unshift(card)
   try {
-    await uploadKnowledgeDoc(avatarId.value, file)
-    await loadDocs()
+    const created: any = await uploadKnowledgeDoc(avatarId.value, file, (loaded, total) => {
+      const current = pendingUploads.value.find((doc) => doc.id === localId)
+      if (current) current.uploadProgress = Math.min(99, Math.round((loaded / Math.max(1, total)) * 100))
+    })
+    pendingUploads.value = pendingUploads.value.filter((doc) => doc.id !== localId)
+    docs.value = [created, ...docs.value.filter((doc) => doc.id !== created.id)]
+    startDocumentPolling()
   } catch (e: any) {
-    uploadError.value = e?.message || '上传失败'
+    const current = pendingUploads.value.find((doc) => doc.id === localId)
+    if (current) {
+      current.localUploading = false
+      current.errorMessage = e?.message || '上传失败'
+    }
+  }
+}
+
+const canRetryDoc = (doc: any) =>
+  !doc.localOnly && doc.filePresent !== false && documentState(doc).tone === 'failed'
+
+const retryDoc = async (doc: any) => {
+  if (!avatarId.value || !canRetryDoc(doc) || retryingDocs.value[doc.id]) return
+  retryingDocs.value = { ...retryingDocs.value, [doc.id]: true }
+  retryErrors.value = { ...retryErrors.value, [doc.id]: '' }
+  try {
+    const updated: any = await retryKnowledgeDoc(avatarId.value, doc.id)
+    Object.assign(doc, updated)
+    startDocumentPolling()
+  } catch (e: any) {
+    retryErrors.value = {
+      ...retryErrors.value,
+      [doc.id]: e?.response?.data?.message || e?.response?.data?.detail || e?.message || '重新索引失败'
+    }
   } finally {
-    uploading.value = false
+    retryingDocs.value = { ...retryingDocs.value, [doc.id]: false }
   }
 }
 
 const removeDoc = async (id: string) => {
+  const local = pendingUploads.value.find((doc) => doc.id === id)
+  if (local?.localOnly) {
+    pendingUploads.value = pendingUploads.value.filter((doc) => doc.id !== id)
+    return
+  }
   if (!avatarId.value) return
   await deleteKnowledgeDoc(avatarId.value, id)
   await loadDocs()
@@ -200,9 +317,21 @@ const toggleQa = async (qa: any, e: Event) => {
   }
 }
 
-const goAddQa = () => router.push('/knowledge/qa/create')
+const goAddQa = () => {
+  if (route.params.avatarId && avatarId.value) {
+    router.push({ name: 'AvatarQaPairCreate', params: { avatarId: avatarId.value } })
+    return
+  }
+  router.push('/knowledge/qa/create')
+}
 
-const goEditQa = (qa: any) => router.push(`/knowledge/qa/${qa.id}/edit`)
+const goEditQa = (qa: any) => {
+  if (route.params.avatarId && avatarId.value) {
+    router.push({ name: 'AvatarQaPairEdit', params: { avatarId: avatarId.value, qaId: qa.id } })
+    return
+  }
+  router.push(`/knowledge/qa/${qa.id}/edit`)
+}
 
 const removeQa = async (id: string) => {
   if (!avatarId.value) return
@@ -226,15 +355,18 @@ onMounted(async () => {
   if (!store.avatars.length) {
     await store.loadAvatars()
   }
+  if (avatarId.value) store.currentAvatarId = avatarId.value
   await Promise.all([loadDocs(), loadQA()])
 })
+
+onUnmounted(stopDocumentPolling)
 </script>
 
 <style scoped>
 .knowledge-page {
   min-height: 100vh;
   background: #F8F9FA;
-  padding-bottom: 80px;
+  padding-bottom: calc(28px + env(safe-area-inset-bottom));
   overflow-x: hidden;
 }
 
@@ -263,16 +395,31 @@ onMounted(async () => {
 .panel-heading p { margin: -5px 0 0; color: #9398AE; font-size: 12px; }
 .mobile-card-list { display: grid; grid-template-columns: minmax(0, 1fr); width: 100%; min-width: 0; gap: 10px; }
 .knowledge-card { display: flex; align-items: center; width: 100%; min-width: 0; box-sizing: border-box; gap: 11px; padding: 14px; background: #fff; border: 1px solid #F1E1D3; border-radius: 16px; box-shadow: 0 5px 16px rgba(112, 62, 22, .04); }
+.document-card { display: grid; grid-template-columns: 42px minmax(0, 1fr) auto; align-items: center; }
 .card-icon { flex: 0 0 auto; width: 42px; height: 42px; display: grid; place-items: center; border-radius: 13px; background: #FFF3E6; font-size: 22px; }
 .card-content { min-width: 0; flex: 1; overflow: hidden; }
 .card-title-row { display: flex; align-items: center; gap: 8px; min-width: 0; }
 .card-title-row strong { min-width: 0; flex: 1; overflow: hidden; color: #27201C; font-size: 14px; text-overflow: ellipsis; white-space: nowrap; }
 .status-pill { flex: 0 0 auto; display: inline-flex; padding: 4px 7px; border-radius: 999px; color: #15803D; background: #ECFDF3; font-size: 10px; white-space: nowrap; }.status-pill.pending { color: #B45309; background: #FFFBEB; }
 .status-pill.missing { color: #B91C1C; background: #FEF2F2; }
+.status-pill.failed { color: #B91C1C; background: #FEF2F2; }
 .card-meta, .card-detail { margin: 5px 0 0; color: #9398AE; font-size: 11px; line-height: 1.4; }.card-detail { color: #8B6B58; }
-.card-delete { flex: 0 0 auto; align-self: center; border: 0; color: #EF4444; background: #FEF2F2; border-radius: 8px; padding: 7px 9px; font-size: 12px; cursor: pointer; }
+.progress-track { width: 100%; height: 4px; margin-top: 8px; overflow: hidden; border-radius: 999px; background: #FDE7D1; }
+.progress-fill { display: block; height: 100%; border-radius: inherit; background: linear-gradient(90deg, #FB923C, #F97316); transition: width .25s ease; }
+.card-actions { flex: 0 0 auto; display: flex; align-items: center; }
+.card-delete, .card-retry { align-self: center; border: 0; border-radius: 8px; padding: 7px 9px; font-size: 12px; cursor: pointer; white-space: nowrap; }
+.card-delete { color: #EF4444; background: #FEF2F2; }
+.card-retry { color: #C15F18; background: #FFF3E6; }
+.card-retry:disabled { cursor: wait; opacity: .65; }
+.card-retry-area { grid-column: 1 / -1; display: flex; align-items: center; justify-content: flex-end; gap: 10px; min-width: 0; }
+.card-retry-error { min-width: 0; overflow: hidden; color: #DC2626; font-size: 11px; line-height: 1.35; text-overflow: ellipsis; white-space: nowrap; }
 .card-empty { padding: 42px 16px; border: 1px dashed #F1D9C3; border-radius: 16px; color: #9398AE; background: #fff; font-size: 14px; text-align: center; }
-.qa-card { align-items: stretch; }.qa-card.qa-disabled { opacity: .58; }
+.qa-card { align-items: stretch; text-align: left; }.qa-card.qa-disabled { opacity: .58; }
+.qa-card .card-content,
+.qa-card .qa-question,
+.qa-card .qa-answer,
+.qa-card .card-meta,
+.qa-card .qa-card-actions { text-align: left; }
 .qa-card-head { display: flex; align-items: center; justify-content: space-between; margin-bottom: 8px; }.qa-label { color: #C15F18; font-size: 11px; font-weight: 700; }
 .qa-question { display: block; color: #27201C; font-size: 15px; line-height: 1.5; }.qa-answer { display: -webkit-box; margin: 7px 0 0; overflow: hidden; color: #6B7280; font-size: 13px; line-height: 1.55; -webkit-box-orient: vertical; -webkit-line-clamp: 3; }
 .qa-card-actions { display: flex; gap: 8px; margin-top: 11px; }
@@ -491,8 +638,13 @@ onMounted(async () => {
 @media (max-width: 520px) {
   .knowledge-panel { padding: 0 12px; }
   .knowledge-card { display: grid; grid-template-columns: 42px minmax(0, 1fr); align-items: start; gap: 10px; padding: 13px; }
+  .document-card { grid-template-columns: 42px minmax(0, 1fr) auto; }
   .card-content { grid-column: 2; }
-  .card-delete { grid-column: 2; justify-self: end; margin-top: -2px; }
+  .card-actions { grid-column: 3; grid-row: 1; }
+  .card-delete { justify-self: end; margin-top: -2px; }
+  .card-retry-area { grid-column: 1 / -1; }
+  .qa-card { display: block; }
+  .qa-card .card-content { width: 100%; grid-column: 1; }
   .card-title-row { align-items: flex-start; flex-wrap: wrap; gap: 5px 7px; }
   .status-pill { order: 2; }
   .search-bar { gap: 8px; }

digital-avatar-app/src/views/QaPairEdit.vue

@@ -10,10 +10,12 @@
     <section class="form-section">
       <label class="field-label">问题</label>
       <textarea
+        ref="questionInput"
         v-model="form.question"
-        class="field-input"
-        rows="3"
+        class="field-input question-input"
+        rows="1"
         placeholder="例如：你们的退款政策是什么？"
+        @input="resizeQuestion"
       ></textarea>
 
       <label class="field-label">标准答案</label>
@@ -46,23 +48,31 @@
 </template>
 
 <script setup lang="ts">
-import { ref, reactive, computed, onMounted } from 'vue'
+import { ref, reactive, computed, nextTick, onMounted } from 'vue'
 import { useRouter, useRoute } from 'vue-router'
 import { useAvatarStore } from '@/store/avatar'
-import { pickAvatarId, unwrapListData } from '@/utils/avatar-page-data.js'
+import { pickScopedAvatarId, unwrapListData } from '@/utils/avatar-page-data.js'
 import { getQAPairs, createQAPair, updateQAPair } from '@/api'
 
 const router = useRouter()
 const route = useRoute()
 const store = useAvatarStore()
 
-const avatarId = computed(() => pickAvatarId(store.currentAvatarId, store.avatars))
+const avatarId = computed(() => pickScopedAvatarId(route.params.avatarId, store.currentAvatarId, store.avatars))
 const qaId = computed(() => (route.params.qaId as string) || null)
 const isEdit = computed(() => !!qaId.value)
 
 const form = reactive({ question: '', answer: '', enabled: true })
 const saving = ref(false)
 const error = ref('')
+const questionInput = ref<HTMLTextAreaElement | null>(null)
+
+const resizeQuestion = (event?: Event) => {
+  const element = (event?.target as HTMLTextAreaElement | null) || questionInput.value
+  if (!element) return
+  element.style.height = 'auto'
+  element.style.height = `${element.scrollHeight}px`
+}
 
 const goBack = () => router.back()
 
@@ -104,8 +114,11 @@ const save = async () => {
     } else {
       await createQAPair(avatarId.value, payload)
     }
-    // 保存成功返回知识库管理页
+    if (route.params.avatarId) {
+      router.replace({ name: 'AvatarKnowledgeManage', params: { avatarId: avatarId.value } })
+    } else {
       router.replace('/knowledge')
+    }
   } catch (e: any) {
     error.value = e?.message || '保存失败'
   } finally {
@@ -117,9 +130,12 @@ onMounted(async () => {
   if (!store.avatars.length) {
     await store.loadAvatars()
   }
+  if (avatarId.value) store.currentAvatarId = avatarId.value
   if (isEdit.value) {
     await loadForEdit()
   }
+  await nextTick()
+  resizeQuestion()
 })
 </script>
 
@@ -190,6 +206,13 @@ onMounted(async () => {
   border-color: #F97316;
 }
 
+.question-input {
+  min-height: 44px;
+  overflow: hidden;
+  resize: none;
+  line-height: 1.55;
+}
+
 .switch-row {
   display: flex;
   align-items: center;

digital-avatar-app/src/views/SmsLogin.vue

@@ -138,7 +138,7 @@
 </template>
 
 <script setup lang="ts">
-import { reactive, ref, computed, onUnmounted } from 'vue'
+import { reactive, ref, computed, onMounted, onUnmounted } from 'vue'
 import { useRouter, useRoute } from 'vue-router'
 import { useUserStore } from '@/store/user'
 import { useAvatarStore } from '@/store/avatar'
@@ -161,6 +161,14 @@ const password = ref('')
 const loading = ref(false)
 const errorMsg = ref('')
 
+onMounted(() => {
+  const ssoError = sessionStorage.getItem('hh_sso_error')
+  if (ssoError) {
+    errorMsg.value = ssoError
+    sessionStorage.removeItem('hh_sso_error')
+  }
+})
+
 const counting = ref(false)
 const countdown = ref(60)
 let timer: any = null

digital-avatar-app/src/views/TokenCharge.vue

@@ -3,7 +3,7 @@
     <!-- 顶部导航 -->
     <header class="page-header">
       <button class="back-btn" @click="goBack">‹</button>
-      <h1 class="page-title">Token 充值</h1>
+      <h1 class="page-title">积分充值</h1>
       <div class="header-right"></div>
     </header>
 
@@ -12,7 +12,8 @@
       <div class="balance-card">
         <span class="balance-label">当前余额</span>
         <span class="balance-amount">{{ currentBalance.toLocaleString() }}</span>
-        <span class="balance-unit">Token</span>
+        <span class="balance-unit">积分</span>
+        <span class="balance-used">累计使用 {{ totalConsumed.toLocaleString() }} 积分</span>
       </div>
     </section>
 
@@ -28,10 +29,10 @@
           @click="selectedPlan = plan"
         >
           <div class="plan-badge" v-if="plan.badge">{{ plan.badge }}</div>
-          <div class="plan-amount">{{ plan.amount.toLocaleString() }}</div>
-          <div class="plan-unit">Token</div>
+          <div class="plan-amount">{{ formatTokenAmount(plan.amount) }}</div>
+          <div class="plan-unit">积分</div>
           <div class="plan-price">¥{{ plan.price }}</div>
-          <div class="plan-desc" v-if="plan.desc">{{ plan.desc }}</div>
+          <div class="plan-desc" v-if="plan.desc">{{ formatPointsCopy(plan.desc) }}</div>
         </div>
       </div>
     </section>
@@ -45,7 +46,7 @@
           :class="{ selected: paymentMethod === 'wechat' }"
           @click="paymentMethod = 'wechat'"
         >
-          <span class="payment-icon">💚</span>
+          <span class="payment-icon wechat-icon"><img :src="wechatPayIcon" alt="" /></span>
           <span class="payment-name">微信支付</span>
           <span class="payment-check" v-if="paymentMethod === 'wechat'">✓</span>
         </div>
@@ -54,36 +55,47 @@
           :class="{ selected: paymentMethod === 'alipay' }"
           @click="paymentMethod = 'alipay'"
         >
-          <span class="payment-icon">💙</span>
+          <span class="payment-icon alipay-icon"><img :src="alipayIcon" alt="" /></span>
           <span class="payment-name">支付宝</span>
           <span class="payment-check" v-if="paymentMethod === 'alipay'">✓</span>
         </div>
       </div>
+      <p v-if="paymentNotice" class="payment-notice" :class="paymentNoticeTone">{{ paymentNotice }}</p>
     </section>
 
     <!-- 充值按钮 -->
     <section class="checkout-section">
       <button
         class="checkout-btn"
-        :class="{ disabled: !selectedPlan }"
-        :disabled="!selectedPlan"
+        :class="{ disabled: !selectedPlan || charging || Boolean(pendingOrderId) }"
+        :disabled="!selectedPlan || charging || Boolean(pendingOrderId)"
         @click="doCharge"
       >
-        {{ selectedPlan ? `立即支付 ¥${selectedPlan.price}` : '请选择充值套餐' }}
+        {{ checkoutLabel }}
       </button>
     </section>
   </div>
 </template>
 
 <script setup lang="ts">
-import { ref, onMounted } from 'vue'
+import { computed, onMounted, onUnmounted, ref } from 'vue'
 import { useRouter } from 'vue-router'
-import { getTokenBalance, getRechargePlans, chargeToken } from '@/api'
+import {
+  chargeToken,
+  getRechargePlans,
+  getTokenBalance,
+  getTokenPaymentStatus,
+  type TokenPaymentOrder
+} from '@/api'
+import { isInUniWebView, onNativeMessage, UniEvents } from '@/utils/uniapp-bridge'
+import wechatPayIcon from '@/assets/wechat-pay.svg'
+import alipayIcon from '@/assets/alipay.svg'
 
 const router = useRouter()
 
 // 当前余额
-const currentBalance = ref<number>(1250)
+const currentBalance = ref<number>(0)
+const totalConsumed = ref<number>(0)
 
 // 充值套餐
 const plans = ref<Array<{
@@ -99,12 +111,19 @@ const selectedPlan = ref<any>(null)
 
 // 支付方式
 const paymentMethod = ref<'wechat' | 'alipay'>('wechat')
+const paymentNotice = ref('')
+const paymentNoticeTone = ref<'pending' | 'success' | 'error'>('pending')
+const pendingOrderId = ref(sessionStorage.getItem('hh_pending_payment_order') || '')
+let pollTimer: number | undefined
+let pollDeadline = 0
+let removeNativeListener: (() => void) | undefined
 
 // 从后端加载余额与套餐
 const loadData = async () => {
   try {
     const b: any = await getTokenBalance()
     currentBalance.value = b?.balance ?? 0
+    totalConsumed.value = b?.totalConsumed ?? 0
   } catch (e) {
     console.error('加载余额失败', e)
   }
@@ -116,18 +135,144 @@ const loadData = async () => {
   }
 }
 
-// 执行充值（写入后端）
+// 会会支付订单创建与到账确认
 const charging = ref(false)
+const formatTokenAmount = (amount: number) => {
+  if (amount >= 1_000_000 && amount % 1_000_000 === 0) return `${amount / 1_000_000}M`
+  return amount.toLocaleString()
+}
+const formatPointsCopy = (value: string) => value.replace(/\bToken\b/gi, '积分')
+
+const checkoutLabel = computed(() => {
+  if (!selectedPlan.value) return '请选择充值套餐'
+  if (charging.value) return '正在创建支付订单…'
+  if (pendingOrderId.value) return '等待支付结果…'
+  return `立即支付 ¥${selectedPlan.value.price}`
+})
+
+const payScene = () => {
+  if (paymentMethod.value === 'wechat' && /MicroMessenger/i.test(navigator.userAgent)) return 'JSAPI' as const
+  return 'APP' as const
+}
+
+const parsePayMessage = (message: string) => {
+  const text = (message || '').trim()
+  if (!text) return null
+  try {
+    return JSON.parse(text)
+  } catch {
+    return text
+  }
+}
+
+const findPaymentUrl = (value: any): string => {
+  if (typeof value === 'string' && /^https?:\/\//i.test(value)) return value
+  if (!value || typeof value !== 'object') return ''
+  for (const key of ['url', 'payUrl', 'mwebUrl', 'mweb_url', 'redirectUrl']) {
+    if (typeof value[key] === 'string' && /^https?:\/\//i.test(value[key])) return value[key]
+  }
+  return ''
+}
+
+const invokeWechatJsapi = (params: Record<string, any>) => new Promise<void>((resolve, reject) => {
+  const invoke = () => {
+    const bridge = (window as any).WeixinJSBridge
+    if (!bridge?.invoke) {
+      reject(new Error('当前微信环境无法拉起支付'))
+      return
+    }
+    bridge.invoke('getBrandWCPayRequest', params, (result: any) => {
+      const message = String(result?.err_msg || result?.errMsg || '')
+      if (/\bok\b/i.test(message)) resolve()
+      else reject(new Error(/cancel/i.test(message) ? '支付已取消' : '微信支付未完成'))
+    })
+  }
+  if ((window as any).WeixinJSBridge) invoke()
+  else document.addEventListener('WeixinJSBridgeReady', invoke, { once: true })
+})
+
+const launchPayment = async (order: TokenPaymentOrder) => {
+  const parsed = parsePayMessage(order.payMessage)
+  if (order.payWay === 'JSAPI' && order.paymentMethod === 'wechat' && parsed && typeof parsed === 'object') {
+    await invokeWechatJsapi(parsed)
+    return
+  }
+  if (isInUniWebView() && UniEvents.payment({
+    orderId: order.id,
+    orderNo: order.orderNo,
+    paymentMethod: order.paymentMethod,
+    payType: order.payType,
+    payWay: order.payWay,
+    payMessage: order.payMessage,
+    paymentParams: parsed
+  })) return
+
+  const paymentUrl = findPaymentUrl(parsed)
+  if (paymentUrl) {
+    window.location.assign(paymentUrl)
+    return
+  }
+  throw new Error('请在会会 App 内完成支付')
+}
+
+const clearPendingOrder = () => {
+  pendingOrderId.value = ''
+  sessionStorage.removeItem('hh_pending_payment_order')
+  if (pollTimer) window.clearTimeout(pollTimer)
+  pollTimer = undefined
+}
+
+const pollPayment = async () => {
+  if (!pendingOrderId.value) return
+  try {
+    const order = await getTokenPaymentStatus(pendingOrderId.value)
+    if (order.status === 'paid') {
+      currentBalance.value = order.balance
+      paymentNoticeTone.value = 'success'
+      paymentNotice.value = `支付成功，${order.pointsAmount.toLocaleString()} 积分已到账`
+      clearPendingOrder()
+      return
+    }
+    if (order.status === 'failed') {
+      paymentNoticeTone.value = 'error'
+      paymentNotice.value = order.failureReason || '支付未完成，请重新发起'
+      clearPendingOrder()
+      return
+    }
+  } catch (error) {
+    console.error('查询支付结果失败', error)
+  }
+  if (Date.now() < pollDeadline) pollTimer = window.setTimeout(pollPayment, 2000)
+  else {
+    paymentNoticeTone.value = 'pending'
+    paymentNotice.value = '支付结果仍在确认中，可稍后刷新余额或重新发起'
+    clearPendingOrder()
+  }
+}
+
+const startPolling = (orderId: string) => {
+  pendingOrderId.value = orderId
+  sessionStorage.setItem('hh_pending_payment_order', orderId)
+  pollDeadline = Date.now() + 2 * 60 * 1000
+  if (pollTimer) window.clearTimeout(pollTimer)
+  void pollPayment()
+}
+
 const doCharge = async () => {
   if (!selectedPlan.value || charging.value) return
   charging.value = true
+  paymentNotice.value = ''
   try {
-    const methodText = paymentMethod.value === 'wechat' ? '微信支付' : '支付宝'
-    const res: any = await chargeToken(selectedPlan.value.id)
-    currentBalance.value = res?.balance ?? currentBalance.value
-    alert(`已通过${methodText}成功充值，当前余额：${currentBalance.value} Token`)
-  } catch (e) {
-    alert('充值失败，请重试')
+    const order = await chargeToken(selectedPlan.value.id, paymentMethod.value, payScene())
+    if (order.status === 'failed') throw new Error(order.failureReason || '支付订单创建失败')
+    paymentNoticeTone.value = 'pending'
+    paymentNotice.value = '支付订单已创建，正在等待付款结果'
+    startPolling(order.id)
+    await launchPayment(order)
+  } catch (error: any) {
+    paymentNoticeTone.value = 'error'
+    paymentNotice.value = error?.message || '充值失败，请重试'
+    clearPendingOrder()
   } finally {
     charging.value = false
   }
@@ -139,7 +284,33 @@ const goBack = () => {
 }
 
 onMounted(() => {
-  loadData()
+  void loadData()
+  removeNativeListener = onNativeMessage((message) => {
+    if (message?.type !== 'paymentResult') return
+    if (message.orderId && message.orderId !== pendingOrderId.value) return
+    if (['cancelled', 'canceled'].includes(String(message.status).toLowerCase())) {
+      paymentNoticeTone.value = 'error'
+      paymentNotice.value = '支付已取消'
+      clearPendingOrder()
+      return
+    }
+    if (String(message.status).toLowerCase() === 'failed') {
+      paymentNoticeTone.value = 'error'
+      paymentNotice.value = message.message || '支付未完成，请重新发起'
+      clearPendingOrder()
+      return
+    }
+    void pollPayment()
+  })
+  if (pendingOrderId.value) {
+    paymentNotice.value = '正在确认上一笔支付结果'
+    startPolling(pendingOrderId.value)
+  }
+})
+
+onUnmounted(() => {
+  if (pollTimer) window.clearTimeout(pollTimer)
+  removeNativeListener?.()
 })
 </script>
 
@@ -187,6 +358,7 @@ onMounted(() => {
 
 .balance-card {
   display: flex;
+  flex-wrap: wrap;
   align-items: baseline;
   gap: 8px;
   padding: 20px;
@@ -211,6 +383,12 @@ onMounted(() => {
   opacity: 0.9;
 }
 
+.balance-used {
+  flex-basis: 100%;
+  font-size: 12px;
+  opacity: 0.82;
+}
+
 /* 充值套餐 */
 .plans-section {
   padding: 0 20px 20px;
@@ -319,7 +497,24 @@ onMounted(() => {
 }
 
 .payment-icon {
-  font-size: 24px;
+  width: 38px;
+  height: 38px;
+  display: grid;
+  place-items: center;
+  border-radius: 11px;
+}
+
+.payment-icon img {
+  width: 25px;
+  height: 25px;
+}
+
+.wechat-icon {
+  background: #EAF9F0;
+}
+
+.alipay-icon {
+  background: #EAF3FF;
 }
 
 .payment-name {
@@ -342,6 +537,26 @@ onMounted(() => {
   font-weight: 600;
 }
 
+.payment-notice {
+  margin: 12px 2px 0;
+  padding: 10px 12px;
+  border-radius: 10px;
+  color: #9A5B16;
+  background: #FFF5E8;
+  font-size: 13px;
+  line-height: 1.5;
+}
+
+.payment-notice.success {
+  color: #08783E;
+  background: #EAF9F0;
+}
+
+.payment-notice.error {
+  color: #B42318;
+  background: #FFF0EE;
+}
+
 /* 充值按钮 */
 .checkout-section {
   padding: 0 20px;

docker-compose.yml

@@ -17,6 +17,7 @@ services:
       - REDIS_PORT=6379
       - SECRET_KEY=your-secret-key-change-in-production
       - AES_KEY=your-aes-key-32-chars-change-now!
+      - AVATAR_MODEL_CONFIG_TOKEN=${AVATAR_MODEL_CONFIG_TOKEN:-}
       - TZ=Asia/Shanghai
       - AVATAR_DB_PATH=/app/avatar.db
     volumes:

docker/mysql/init.sql

@@ -94,6 +94,7 @@ CREATE TABLE IF NOT EXISTS `ai_model_configs` (
   `id` bigint NOT NULL AUTO_INCREMENT,
   `model_name` varchar(64) NOT NULL COMMENT '模型名称',
   `provider` varchar(32) NOT NULL COMMENT 'openai/zhipu/wenxin/qianwen/local',
+  `usage_scope` varchar(16) NOT NULL DEFAULT 'general' COMMENT '用途:general/digital_avatar',
   `api_base_url` varchar(256) DEFAULT NULL COMMENT 'API地址',
   `api_key_enc` varchar(512) DEFAULT NULL COMMENT '加密API Key',
   `model_version` varchar(64) DEFAULT NULL COMMENT '模型版本',

frontend/Dockerfile

@@ -6,7 +6,8 @@ RUN npm install
 COPY . .
 RUN npm run build
 
-FROM nginx:alpine
+# Nginx 1.31 uses syscalls that are blocked by the test server's legacy kernel.
+FROM nginx:1.28.3-alpine
 COPY --from=build /app/dist /usr/share/nginx/html
 COPY nginx.conf /etc/nginx/conf.d/default.conf
 EXPOSE 80

frontend/src/views/AIModels.vue

@@ -15,12 +15,17 @@
             <span class="model-title">{{ m.model_name }}</span>
           </div>
           <div style="display:flex;gap:6px;align-items:center">
+            <el-tag :type="m.usage_scope === 'digital_avatar' ? 'warning' : 'info'" size="small">
+              {{ scopeLabels[m.usage_scope] || '通用业务' }}
+            </el-tag>
             <el-tag v-if="m.is_default" type="success" size="small">默认</el-tag>
             <el-tag v-if="!m.is_enabled" type="danger" size="small">禁用</el-tag>
           </div>
         </div>
         <div class="model-meta">
           <span>版本: {{ m.model_version || '--' }}</span>
+          <span v-if="m.usage_scope === 'digital_avatar'">视觉: {{ m.vision_model_version || 'qwen3.6-flash' }}</span>
+          <span v-if="m.usage_scope === 'digital_avatar'">病例OCR: {{ m.ocr_model_version || 'qwen-vl-ocr' }}</span>
           <span>温度: {{ m.temperature }}</span>
           <span>Max Tokens: {{ m.max_tokens }}</span>
           <span>超时: {{ m.timeout_seconds }}s</span>
@@ -49,6 +54,13 @@
             <el-option v-for="(l,v) in providerLabels" :key="v" :label="l" :value="v" />
           </el-select>
         </el-form-item>
+        <el-form-item label="使用场景" prop="usage_scope">
+          <el-radio-group v-model="form.usage_scope">
+            <el-radio-button value="general">通用业务</el-radio-button>
+            <el-radio-button value="digital_avatar">数字分身专用</el-radio-button>
+          </el-radio-group>
+          <div class="scope-tip">数字分身专用模型仅用于分身对话和主动接管回复</div>
+        </el-form-item>
         <el-form-item label="API地址">
           <el-input v-model="form.api_base_url" placeholder="留空使用默认地址" />
         </el-form-item>
@@ -58,6 +70,15 @@
         <el-form-item label="模型版本">
           <el-input v-model="form.model_version" placeholder="如: gpt-4-turbo, glm-4" />
         </el-form-item>
+        <template v-if="form.usage_scope === 'digital_avatar'">
+          <el-form-item label="视觉模型">
+            <el-input v-model="form.vision_model_version" placeholder="如: qwen3.6-flash" />
+          </el-form-item>
+          <el-form-item label="病例OCR模型">
+            <el-input v-model="form.ocr_model_version" placeholder="如: qwen-vl-ocr" />
+            <div class="scope-tip">识别为病例、处方、检查单后自动调用，普通图片不会重复调用。</div>
+          </el-form-item>
+        </template>
         <el-row :gutter="16">
           <el-col :span="12">
             <el-form-item label="温度">
@@ -131,8 +152,9 @@ const testResult = ref(null)
 const testing = ref(false)
 
 const providerLabels = { openai: 'OpenAI', zhipu: '智谱GLM', wenxin: '文心一言', qianwen: '通义千问', local: '本地模型' }
-const form = reactive({ model_name: '', provider: 'openai', api_base_url: '', api_key: '', model_version: '', temperature: 0.7, max_tokens: 1000, timeout_seconds: 30, is_default: 0 })
-const rules = { model_name: [{ required: true, message: '请输入模型名称' }], provider: [{ required: true }] }
+const scopeLabels = { general: '通用业务', digital_avatar: '数字分身专用' }
+const form = reactive({ model_name: '', provider: 'openai', usage_scope: 'general', api_base_url: '', api_key: '', model_version: '', vision_model_version: 'qwen3.6-flash', ocr_model_version: 'qwen-vl-ocr', temperature: 0.7, max_tokens: 1000, timeout_seconds: 30, is_default: 0 })
+const rules = { model_name: [{ required: true, message: '请输入模型名称' }], provider: [{ required: true }], usage_scope: [{ required: true }] }
 
 async function load() {
   const res = await getAIModels()
@@ -155,13 +177,13 @@ function onProviderChange(provider) {
 
 function openCreate() {
   editModel.value = null
-  Object.assign(form, { model_name: '', provider: 'openai', api_base_url: PROVIDER_DEFAULTS.openai.api_base_url, api_key: '', model_version: PROVIDER_DEFAULTS.openai.model_version, temperature: 0.7, max_tokens: 1000, timeout_seconds: 30, is_default: 0 })
+  Object.assign(form, { model_name: '', provider: 'openai', usage_scope: 'general', api_base_url: PROVIDER_DEFAULTS.openai.api_base_url, api_key: '', model_version: PROVIDER_DEFAULTS.openai.model_version, vision_model_version: 'qwen3.6-flash', ocr_model_version: 'qwen-vl-ocr', temperature: 0.7, max_tokens: 1000, timeout_seconds: 30, is_default: 0 })
   dialogVisible.value = true
 }
 
 function openEdit(m) {
   editModel.value = m
-  Object.assign(form, { model_name: m.model_name, provider: m.provider, api_base_url: m.api_base_url || '', api_key: '', model_version: m.model_version || '', temperature: m.temperature, max_tokens: m.max_tokens, timeout_seconds: m.timeout_seconds, is_default: m.is_default })
+  Object.assign(form, { model_name: m.model_name, provider: m.provider, usage_scope: m.usage_scope || 'general', api_base_url: m.api_base_url || '', api_key: '', model_version: m.model_version || '', vision_model_version: m.vision_model_version || 'qwen3.6-flash', ocr_model_version: m.ocr_model_version || 'qwen-vl-ocr', temperature: m.temperature, max_tokens: m.max_tokens, timeout_seconds: m.timeout_seconds, is_default: m.is_default })
   dialogVisible.value = true
 }
 
@@ -236,4 +258,5 @@ onMounted(load)
 .result-meta { display: flex; align-items: center; gap: 10px; margin-bottom: 10px; }
 .result-content { background: var(--color-bg); border: 1px solid var(--color-border); border-radius: 8px; padding: 12px; font-size: 13px; line-height: 1.6; white-space: pre-wrap; max-height: 200px; overflow-y: auto; }
 .empty-state { grid-column: 1/-1; padding: 40px; }
+.scope-tip { margin-top: 6px; color: var(--color-text-muted); font-size: 12px; line-height: 1.5; }
 </style>

uniapp-avatar/README.md

@@ -54,6 +54,7 @@ H5 引入 uniapp web-view bridge 后调用：
 | `needLogin` | — | token 失效，请求壳重新登录 |
 | `setTitle` | `title` | 设置原生导航栏标题 |
 | `navigate` | `path` | 请求原生跳转（打开原生页/新 web-view） |
+| `payment` | `payment` | 拉起会会原生支付；包含 `orderId/orderNo/payType/payWay/payMessage/paymentParams` |
 | `back` | — | 请求原生返回 |
 
 ### 2.3 原生 → H5（壳主动推送）
@@ -63,9 +64,12 @@ H5 引入 uniapp web-view bridge 后调用：
 | `context` | `platform, version` | 注入运行环境信息 |
 | `tokenRefresh` | `token` | 登录刷新后下发新 token |
 | `userUpdate` | `user` | 会会资料变更 |
+| `paymentResult` | `orderId,status` | 原生支付结束通知；`status` 为 `success/cancelled/failed` |
 
 > H5 侧用 `onNativeMessage(cb)` 注册 `window.__uniBridgeHandle__`，见 `digital-avatar-app/src/utils/uniapp-bridge.ts`。
 
+壳收到 `payment` 后应调用会会 App 已有的微信/支付宝支付能力（或 `uni.requestPayment`），把 `payMessage/paymentParams` 原样交给对应渠道。原生 SDK 返回后再发送 `paymentResult`；H5 不以原生返回作为到账依据，只会轮询本地订单，最终由会会服务端支付回调确认并增加积分。
+
 ---
 
 ## 3. 项目结构（uni CLI / src 布局，已验证可编译）