Merge branch 'develop' into refactor/memory_search

# Conflicts:
#	api/app/core/memory/storage_services/search/__init__.py

api/app/core/agent/langchain_agent.py

@@ -12,7 +12,7 @@ import time
 from typing import Any, AsyncGenerator, Dict, List, Optional, Sequence
 
 from langchain.agents import create_agent
-from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage
 from langchain_core.tools import BaseTool
 from langgraph.errors import GraphRecursionError
 
@@ -41,6 +41,7 @@ class LangChainAgent:
             max_tool_consecutive_calls: int = 3,  # 单个工具最大连续调用次数
             deep_thinking: bool = False,  # 是否启用深度思考模式
             thinking_budget_tokens: Optional[int] = None,  # 深度思考 token 预算
+            json_output: bool = False,  # 是否强制 JSON 输出
             capability: Optional[List[str]] = None  # 模型能力列表，用于校验是否支持深度思考
     ):
         """初始化 LangChain Agent
@@ -64,7 +65,6 @@ class LangChainAgent:
         self.streaming = streaming
         self.is_omni = is_omni
         self.max_tool_consecutive_calls = max_tool_consecutive_calls
-        self.deep_thinking = deep_thinking and ("thinking" in (capability or []))
 
         # 工具调用计数器：记录每个工具的连续调用次数
         self.tool_call_counter: Dict[str, int] = {}
@@ -80,6 +80,17 @@ class LangChainAgent:
 
         self.system_prompt = system_prompt or "你是一个专业的AI助手"
 
+        # ChatTongyi 要求 messages 含 'json' 字样才能使用 response_format
+        # 在 system prompt 中注入 JSON 要求
+        from app.models.models_model import ModelProvider
+        if json_output and (
+            (provider.lower() == ModelProvider.DASHSCOPE and not is_omni)
+            or provider.lower() == ModelProvider.VOLCANO
+            # 有工具时 response_format 会被移除，所有 provider 都需要 system prompt 注入保证 JSON 输出
+            or bool(tools)
+        ):
+            self.system_prompt += "\n请以JSON格式输出。"
+
         logger.debug(
             f"Agent 迭代次数配置: max_iterations={self.max_iterations}, "
             f"tool_count={len(self.tools)}, "
@@ -87,23 +98,17 @@ class LangChainAgent:
             f"auto_calculated={max_iterations is None}"
         )
 
-        # 根据 capability 校验是否真正支持深度思考
-        actual_deep_thinking = self.deep_thinking
-        if deep_thinking and not actual_deep_thinking:
-            logger.warning(
-                f"模型 {model_name} 不支持深度思考（capability 中无 'thinking'），已自动关闭 deep_thinking"
-            )
-
-        # 创建 RedBearLLM（支持多提供商）
+        # 创建 RedBearLLM，capability 校验由 RedBearModelConfig 统一处理
         model_config = RedBearModelConfig(
             model_name=model_name,
             provider=provider,
             api_key=api_key,
             base_url=api_base,
             is_omni=is_omni,
-            deep_thinking=actual_deep_thinking,
-            thinking_budget_tokens=thinking_budget_tokens if actual_deep_thinking else None,
-            support_thinking="thinking" in (capability or []),
+            capability=capability,
+            deep_thinking=deep_thinking,
+            thinking_budget_tokens=thinking_budget_tokens,
+            json_output=json_output,
             extra_params={
                 "temperature": temperature,
                 "max_tokens": max_tokens,
@@ -112,6 +117,9 @@ class LangChainAgent:
         )
 
         self.llm = RedBearLLM(model_config, type=ModelType.CHAT)
+        # 从经过校验的 config 读取实际生效的能力开关
+        self.deep_thinking = model_config.deep_thinking
+        self.json_output = model_config.json_output
 
         # 获取底层模型用于真正的流式调用
         self._underlying_llm = self.llm._model if hasattr(self.llm, '_model') else self.llm
@@ -237,9 +245,7 @@ class LangChainAgent:
         Returns:
             List[BaseMessage]: 消息列表
         """
-        messages:list = [SystemMessage(content=self.system_prompt)]
-
-        # 添加系统提示词
+        messages: list = []
 
         # 添加历史消息
         if history:

api/app/core/api_key_auth.py

@@ -96,6 +96,38 @@ def require_api_key(
                 resource_id=api_key_obj.resource_id,
             )
 
+            # ── Tenant 级别限速（来自套餐配额 api_ops_rate_limit）──────────
+            try:
+                from app.models.workspace_model import Workspace
+                from premium.platform_admin.package_plan_service import TenantSubscriptionService
+
+                workspace = db.query(Workspace).filter(
+                    Workspace.id == api_key_obj.workspace_id
+                ).first()
+                if workspace:
+                    quota = TenantSubscriptionService(db).get_effective_quota(workspace.tenant_id)
+                    tenant_qps_limit = quota.get("api_ops_rate_limit") if quota else None
+                    if tenant_qps_limit:
+                        rate_limiter = RateLimiterService()
+                        tenant_ok, tenant_info = await rate_limiter.check_tenant_rate_limit(
+                            workspace.tenant_id, tenant_qps_limit
+                        )
+                        if not tenant_ok:
+                            raise RateLimitException(
+                                "租户 API 调用速率超限",
+                                BizCode.API_KEY_QPS_LIMIT_EXCEEDED,
+                                rate_headers={
+                                    "X-RateLimit-Tenant-Limit": str(tenant_info["limit"]),
+                                    "X-RateLimit-Tenant-Remaining": str(tenant_info["remaining"]),
+                                    "X-RateLimit-Tenant-Reset": str(tenant_info["reset"]),
+                                }
+                            )
+            except RateLimitException:
+                raise
+            except Exception as e:
+                logger.warning(f"Tenant 限速检查异常，跳过: {e}")
+            # ─────────────────────────────────────────────────────────────
+
             rate_limiter = RateLimiterService()
             is_allowed, error_msg, rate_headers = await rate_limiter.check_all_limits(api_key_obj)
             if not is_allowed:

api/app/core/memory/agent/utils/write_tools.py

@@ -14,6 +14,7 @@ from dotenv import load_dotenv
 
 from app.core.logging_config import get_agent_logger
 from app.core.memory.agent.utils.get_dialogs import get_chunked_dialogs
+from app.core.memory.storage_services.extraction_engine.deduplication.deduped_and_disamb import _USER_PLACEHOLDER_NAMES
 from app.core.memory.storage_services.extraction_engine.extraction_orchestrator import ExtractionOrchestrator
 from app.core.memory.storage_services.extraction_engine.knowledge_extraction.memory_summary import \
     memory_summary_generation
@@ -191,15 +192,37 @@ async def write(
             if success:
                 logger.info("Successfully saved all data to Neo4j")
                 
-                # 使用 Celery 异步任务触发聚类（不阻塞主流程）
                 if all_entity_nodes:
+                    end_user_id = all_entity_nodes[0].end_user_id
+
+                    # Neo4j 写入完成后，用 PgSQL 权威 aliases 覆盖 Neo4j 用户实体
+                    try:
+                        from app.repositories.end_user_info_repository import EndUserInfoRepository
+                        if end_user_id:
+                            with get_db_context() as db_session:
+                                info = EndUserInfoRepository(db_session).get_by_end_user_id(uuid.UUID(end_user_id))
+                                pg_aliases = info.aliases if info and info.aliases else []
+                            if info is not None:
+                                # 将 Python 侧占位名集合作为参数传入，避免 Cypher 硬编码
+                                placeholder_names = list(_USER_PLACEHOLDER_NAMES)
+                                await neo4j_connector.execute_query(
+                                    """
+                                    MATCH (e:ExtractedEntity)
+                                    WHERE e.end_user_id = $end_user_id AND toLower(e.name) IN $placeholder_names
+                                    SET e.aliases = $aliases
+                                    """,
+                                    end_user_id=end_user_id, aliases=pg_aliases,
+                                    placeholder_names=placeholder_names,
+                                )
+                                logger.info(f"[AliasSync] Neo4j 用户实体 aliases 已用 PgSQL 权威源覆盖: {pg_aliases}")
+                    except Exception as sync_err:
+                        logger.warning(f"[AliasSync] PgSQL→Neo4j aliases 同步失败（不影响主流程）: {sync_err}")
+
+                    # 使用 Celery 异步任务触发聚类（不阻塞主流程）
                     try:
                         from app.tasks import run_incremental_clustering
                         
-                        end_user_id = all_entity_nodes[0].end_user_id
                         new_entity_ids = [e.id for e in all_entity_nodes]
-                        
-                        # 异步提交 Celery 任务
                         task = run_incremental_clustering.apply_async(
                             kwargs={
                                 "end_user_id": end_user_id,
@@ -207,7 +230,6 @@ async def write(
                                 "llm_model_id": str(memory_config.llm_model_id) if memory_config.llm_model_id else None,
                                 "embedding_model_id": str(memory_config.embedding_model_id) if memory_config.embedding_model_id else None,
                             },
-                            # 设置任务优先级（低优先级，不影响主业务）
                             priority=3,
                         )
                         logger.info(
@@ -215,7 +237,6 @@ async def write(
                             f"task_id={task.id}, end_user_id={end_user_id}, entity_count={len(new_entity_ids)}"
                         )
                     except Exception as e:
-                        # 聚类任务提交失败不影响主流程
                         logger.error(f"[Clustering] 提交聚类任务失败（不影响主流程）: {e}", exc_info=True)
                 
                 break

api/app/core/memory/models/__init__.py

@@ -61,9 +61,9 @@ from app.core.memory.models.triplet_models import (
 # User metadata models
 from app.core.memory.models.metadata_models import (
     UserMetadata,
-    UserMetadataBehavioralHints,
     UserMetadataProfile,
     MetadataExtractionResponse,
+    MetadataFieldChange,
 )
 
 # Ontology scenario models (LLM extracted from scenarios)
@@ -133,9 +133,9 @@ __all__ = [
     "Triplet",
     "TripletExtractionResponse",
     "UserMetadata",
-    "UserMetadataBehavioralHints",
     "UserMetadataProfile",
     "MetadataExtractionResponse",
+    "MetadataFieldChange",
     # Ontology models
     "OntologyClass",
     "OntologyExtractionResponse",

api/app/core/memory/models/metadata_models.py

@@ -4,7 +4,7 @@ Independent from triplet_models.py - these models are used by the
 standalone metadata extraction pipeline (post-dedup async Celery task).
 """
 
-from typing import List
+from typing import List, Literal, Optional
 
 from pydantic import BaseModel, ConfigDict, Field
 
@@ -13,8 +13,8 @@ class UserMetadataProfile(BaseModel):
     """用户画像信息"""
 
     model_config = ConfigDict(extra="ignore")
-    role: str = Field(default="", description="用户职业或角色")
-    domain: str = Field(default="", description="用户所在领域")
+    role: List[str] = Field(default_factory=list, description="用户职业或角色")
+    domain: List[str] = Field(default_factory=list, description="用户所在领域")
     expertise: List[str] = Field(
         default_factory=list, description="用户擅长的技能或工具"
     )
@@ -23,31 +23,37 @@ class UserMetadataProfile(BaseModel):
     )
 
 
-class UserMetadataBehavioralHints(BaseModel):
-    """行为偏好"""
-
-    model_config = ConfigDict(extra="ignore")
-    learning_stage: str = Field(default="", description="学习阶段")
-    preferred_depth: str = Field(default="", description="偏好深度")
-    tone_preference: str = Field(default="", description="语气偏好")
-
-
 class UserMetadata(BaseModel):
     """用户元数据顶层结构"""
 
     model_config = ConfigDict(extra="ignore")
     profile: UserMetadataProfile = Field(default_factory=UserMetadataProfile)
-    behavioral_hints: UserMetadataBehavioralHints = Field(
-        default_factory=UserMetadataBehavioralHints
+
+
+class MetadataFieldChange(BaseModel):
+    """单个元数据字段的变更操作"""
+
+    model_config = ConfigDict(extra="ignore")
+    field_path: str = Field(
+        description="字段路径，用点号分隔，如 'profile.role'、'profile.expertise'"
+    )
+    action: Literal["set", "remove"] = Field(
+        description="操作类型：'set' 表示新增或修改，'remove' 表示移除"
+    )
+    value: Optional[str] = Field(
+        default=None,
+        description="字段的新值（action='set' 时必填）。标量字段直接填值，列表字段填单个要新增的元素"
     )
-    knowledge_tags: List[str] = Field(default_factory=list, description="知识标签")
 
 
 class MetadataExtractionResponse(BaseModel):
-    """元数据提取 LLM 响应结构"""
+    """元数据提取 LLM 响应结构（增量模式）"""
 
     model_config = ConfigDict(extra="ignore")
-    user_metadata: UserMetadata = Field(default_factory=UserMetadata)
+    metadata_changes: List[MetadataFieldChange] = Field(
+        default_factory=list,
+        description="元数据的增量变更列表，每项描述一个字段的新增、修改或移除操作",
+    )
     aliases_to_add: List[str] = Field(
         default_factory=list,
         description="本次新发现的用户别名（用户自我介绍或他人对用户的称呼）",

api/app/core/memory/storage_services/extraction_engine/deduplication/deduped_and_disamb.py

@@ -82,51 +82,38 @@ def _merge_attribute(canonical: ExtractedEntityNode, ent: ExtractedEntityNode):
             canonical.connect_strength = next(iter(pair))
 
     # 别名合并（去重保序，使用标准化工具）
+    # 用户实体的 aliases 由 PgSQL end_user_info 作为唯一权威源，去重合并时不修改
     try:
         canonical_name = (getattr(canonical, "name", "") or "").strip()
-        incoming_name = (getattr(ent, "name", "") or "").strip()
-        
-        # 收集所有需要合并的别名
-        all_aliases = []
-        
-        # 1. 添加canonical现有的别名
-        existing = getattr(canonical, "aliases", []) or []
-        all_aliases.extend(existing)
-        
-        # 2. 添加incoming实体的名称（如果不同于canonical的名称）
-        if incoming_name and incoming_name != canonical_name:
-            all_aliases.append(incoming_name)
-        
-        # 3. 添加incoming实体的所有别名
-        incoming = getattr(ent, "aliases", []) or []
-        all_aliases.extend(incoming)
-        
-        # 4. 标准化并去重（优先使用alias_utils工具函数）
-        try:
-            from app.core.memory.utils.alias_utils import normalize_aliases
-            canonical.aliases = normalize_aliases(canonical_name, all_aliases)
-        except Exception:
-            # 如果导入失败，使用增强的去重逻辑
-            seen_normalized = set()
-            unique_aliases = []
+        if canonical_name.lower() not in _USER_PLACEHOLDER_NAMES:
+            incoming_name = (getattr(ent, "name", "") or "").strip()
             
-            for alias in all_aliases:
-                if not alias:
-                    continue
-                
-                alias_stripped = str(alias).strip()
-                if not alias_stripped or alias_stripped == canonical_name:
-                    continue
-                
-                # 标准化：转小写用于去重判断
-                alias_normalized = alias_stripped.lower()
-                
-                if alias_normalized not in seen_normalized:
-                    seen_normalized.add(alias_normalized)
-                    unique_aliases.append(alias_stripped)
+            # 收集所有需要合并的别名，过滤掉用户占位名避免污染非用户实体
+            all_aliases = list(getattr(canonical, "aliases", []) or [])
+            if incoming_name and incoming_name != canonical_name and incoming_name.lower() not in _USER_PLACEHOLDER_NAMES:
+                all_aliases.append(incoming_name)
+            all_aliases.extend(
+                a for a in (getattr(ent, "aliases", []) or [])
+                if a and a.strip().lower() not in _USER_PLACEHOLDER_NAMES
+            )
             
-            # 排序并赋值
-            canonical.aliases = sorted(unique_aliases)
+            try:
+                from app.core.memory.utils.alias_utils import normalize_aliases
+                canonical.aliases = normalize_aliases(canonical_name, all_aliases)
+            except Exception:
+                seen_normalized = set()
+                unique_aliases = []
+                for alias in all_aliases:
+                    if not alias:
+                        continue
+                    alias_stripped = str(alias).strip()
+                    if not alias_stripped or alias_stripped == canonical_name:
+                        continue
+                    alias_normalized = alias_stripped.lower()
+                    if alias_normalized not in seen_normalized:
+                        seen_normalized.add(alias_normalized)
+                        unique_aliases.append(alias_stripped)
+                canonical.aliases = sorted(unique_aliases)
     except Exception:
         pass
 
@@ -733,66 +720,37 @@ def fuzzy_match(
 
 
     def _merge_entities_with_aliases(canonical: ExtractedEntityNode, losing: ExtractedEntityNode):
-        """ 模糊匹配中的实体合并。
+        """模糊匹配中的实体合并（别名部分）。
         
-        合并策略：
-        1. 保留canonical的主名称不变
-        2. 将losing的主名称添加为alias（如果不同）
-        3. 合并两个实体的所有aliases
-        4. 自动去重（case-insensitive）并排序
-        
-        Args:
-            canonical: 规范实体（保留）
-            losing: 被合并实体（删除）
-            
-        Note:
-            使用alias_utils.normalize_aliases进行标准化去重
+        用户实体的 aliases 由 PgSQL end_user_info 作为唯一权威源，跳过合并。
         """
-        # 获取规范实体的名称
         canonical_name = (getattr(canonical, "name", "") or "").strip()
+        if canonical_name.lower() in _USER_PLACEHOLDER_NAMES:
+            return
+
         losing_name = (getattr(losing, "name", "") or "").strip()
         
-        # 收集所有需要合并的别名
-        all_aliases = []
-        
-        # 1. 添加canonical现有的别名
-        current_aliases = getattr(canonical, "aliases", []) or []
-        all_aliases.extend(current_aliases)
-        
-        # 2. 添加losing实体的名称（如果不同于canonical的名称）
+        all_aliases = list(getattr(canonical, "aliases", []) or [])
         if losing_name and losing_name != canonical_name:
             all_aliases.append(losing_name)
+        all_aliases.extend(getattr(losing, "aliases", []) or [])
         
-        # 3. 添加losing实体的所有别名
-        losing_aliases = getattr(losing, "aliases", []) or []
-        all_aliases.extend(losing_aliases)
-        
-        # 4. 标准化并去重（使用标准化后的字符串进行去重）
         try:
             from app.core.memory.utils.alias_utils import normalize_aliases
             canonical.aliases = normalize_aliases(canonical_name, all_aliases)
         except Exception:
-            # 如果导入失败，使用增强的去重逻辑
-            # 使用标准化后的字符串作为key进行去重
             seen_normalized = set()
             unique_aliases = []
-            
             for alias in all_aliases:
                 if not alias:
                     continue
-                
                 alias_stripped = str(alias).strip()
                 if not alias_stripped or alias_stripped == canonical_name:
                     continue
-                
-                # 标准化：转小写用于去重判断
                 alias_normalized = alias_stripped.lower()
-                
                 if alias_normalized not in seen_normalized:
                     seen_normalized.add(alias_normalized)
                     unique_aliases.append(alias_stripped)
-            
-            # 排序并赋值
             canonical.aliases = sorted(unique_aliases)
     
     # ========== 主循环：遍历所有实体对进行模糊匹配 ==========

api/app/core/memory/storage_services/extraction_engine/extraction_orchestrator.py

@@ -1391,18 +1391,18 @@ class ExtractionOrchestrator:
         """
         将本轮提取的用户别名同步到 end_user 和 end_user_info 表。
 
-        注意：此方法在 Neo4j 写入之前调用，因此不能依赖 Neo4j 作为别名的权威数据源。
-        改为直接使用内存中去重后的 entity_nodes 的 aliases，与 PgSQL 已有的 aliases 合并。
+        PgSQL end_user_info.aliases 是用户别名的唯一权威源。
+        此方法仅将本轮 LLM 从对话中新提取的别名增量追加到 PgSQL，
+        不再从 Neo4j 二层去重合并历史别名，避免脏数据反向污染 PgSQL。
 
         策略：
-        1. 从内存中的 entity_nodes 提取本轮用户别名（current_aliases）
-        2. 从去重后的 entity_nodes 中提取完整别名（含 Neo4j 二层去重合并的历史别名）
-        3. 从 PgSQL end_user_info 读取已有的 aliases（db_aliases）
-        4. 合并 db_aliases + deduped_aliases + current_aliases，去重保序
-        5. 写回 PgSQL
+        1. 从本轮对话原始发言中提取用户别名（current_aliases）
+        2. 从 PgSQL end_user_info 读取已有的 aliases（db_aliases）
+        3. 合并 db_aliases + current_aliases，去重保序
+        4. 写回 PgSQL
 
         Args:
-            entity_nodes: 去重后的实体节点列表（内存中，含二层去重合并结果）
+            entity_nodes: 去重后的实体节点列表（内存中）
             dialog_data_list: 对话数据列表
         """
         try:
@@ -1418,11 +1418,6 @@ class ExtractionOrchestrator:
             # 1. 提取本轮对话的用户别名（保持 LLM 提取的原始顺序，不排序）
             current_aliases = self._extract_current_aliases(entity_nodes, dialog_data_list)
 
-            # 1.5 从去重后的 entity_nodes 中提取完整别名
-            # 二层去重会将 Neo4j 中已有的历史别名合并到 entity_nodes 中，
-            # 这里提取出来确保 PgSQL 与 Neo4j 的别名保持同步
-            deduped_aliases = self._extract_deduped_entity_aliases(entity_nodes)
-
             # 1.6 从 Neo4j 查询已有的 AI 助手别名，作为额外的排除源
             # （防止 LLM 未提取出 AI 助手实体时，AI 别名泄漏到用户别名中）
             neo4j_assistant_aliases = await self._fetch_neo4j_assistant_aliases(end_user_id)
@@ -1434,19 +1429,12 @@ class ExtractionOrchestrator:
                 ]
                 if len(current_aliases) < before_count:
                     logger.info(f"通过 Neo4j AI 助手别名排除了 {before_count - len(current_aliases)} 个误归属别名")
-                # 同样过滤 deduped_aliases
-                deduped_aliases = [
-                    a for a in deduped_aliases
-                    if a.strip().lower() not in neo4j_assistant_aliases
-                ]
 
-            if not current_aliases and not deduped_aliases:
+            if not current_aliases:
                 logger.debug(f"本轮未提取到用户别名，跳过同步: end_user_id={end_user_id}")
                 return
 
             logger.info(f"本轮对话提取的 aliases: {current_aliases}")
-            if deduped_aliases:
-                logger.info(f"去重后实体的完整 aliases（含历史）: {deduped_aliases}")
 
             # 2. 同步到数据库
             end_user_uuid = uuid.UUID(end_user_id)
@@ -1457,21 +1445,15 @@ class ExtractionOrchestrator:
                     logger.warning(f"未找到 end_user_id={end_user_id} 的用户记录")
                     return
 
-                # 3. 从 PgSQL 读取已有 aliases 并与本轮合并
+                # 3. 从 PgSQL 读取已有 aliases 并与本轮新增合并
                 info = EndUserInfoRepository(db).get_by_end_user_id(end_user_uuid)
                 db_aliases = (info.aliases if info and info.aliases else [])
                 # 过滤掉占位名称
                 db_aliases = [a for a in db_aliases if a.strip().lower() not in self.USER_PLACEHOLDER_NAMES]
 
-                # 合并：已有 + 去重后完整别名 + 本轮新增，去重保序
+                # 合并：PgSQL 已有 + 本轮新增，去重保序（不再合并 Neo4j 历史别名）
                 merged_aliases = list(db_aliases)
                 seen_lower = {a.strip().lower() for a in merged_aliases}
-                # 先合并去重后实体的完整别名（含 Neo4j 历史别名）
-                for alias in deduped_aliases:
-                    if alias.strip().lower() not in seen_lower:
-                        merged_aliases.append(alias)
-                        seen_lower.add(alias.strip().lower())
-                # 再合并本轮新提取的别名
                 for alias in current_aliases:
                     if alias.strip().lower() not in seen_lower:
                         merged_aliases.append(alias)
@@ -1505,9 +1487,7 @@ class ExtractionOrchestrator:
                         info.aliases = merged_aliases
                         logger.info(f"同步合并后 aliases 到 end_user_info: {merged_aliases}")
                 else:
-                    first_alias = current_aliases[0].strip() if current_aliases else (
-                        deduped_aliases[0].strip() if deduped_aliases else ""
-                    )
+                    first_alias = current_aliases[0].strip() if current_aliases else ""
                     # 确保 first_alias 不是占位名称
                     if first_alias and first_alias.lower() not in self.USER_PLACEHOLDER_NAMES:
                         db.add(EndUserInfo(

api/app/core/memory/storage_services/extraction_engine/knowledge_extraction/metadata_extractor.py

@@ -118,7 +118,7 @@ class MetadataExtractor:
         existing_aliases: Optional[List[str]] = None,
     ) -> Optional[tuple]:
         """
-        对筛选后的 statement 列表调用 LLM 提取元数据和用户别名。
+        对筛选后的 statement 列表调用 LLM 提取元数据增量变更和用户别名。
 
         Args:
             statements: 用户发言的 statement 文本列表
@@ -126,7 +126,8 @@ class MetadataExtractor:
             existing_aliases: 数据库已有的用户别名列表（可选）
 
         Returns:
-            (UserMetadata, List[str], List[str]) tuple: (metadata, aliases_to_add, aliases_to_remove) on success, None on failure
+            (List[MetadataFieldChange], List[str], List[str]) tuple:
+            (metadata_changes, aliases_to_add, aliases_to_remove) on success, None on failure
         """
         if not statements:
             return None
@@ -160,12 +161,12 @@ class MetadataExtractor:
             )
 
             if response:
-                metadata = response.user_metadata if response.user_metadata else None
+                changes = response.metadata_changes if response.metadata_changes else []
                 to_add = response.aliases_to_add if response.aliases_to_add else []
                 to_remove = (
                     response.aliases_to_remove if response.aliases_to_remove else []
                 )
-                return metadata, to_add, to_remove
+                return changes, to_add, to_remove
 
             logger.warning("LLM 返回的响应为空")
             return None

api/app/core/memory/utils/prompt/prompts/extract_user_metadata.jinja2

@@ -1,5 +1,5 @@
 ===Task===
-Extract user metadata from the following conversation statements spoken by the user.
+Extract user metadata changes from the following conversation statements spoken by the user.
 
 {% if language == "zh" %}
 **"三度原则"判断标准：**
@@ -10,28 +10,36 @@ Extract user metadata from the following conversation statements spoken by the u
 **提取规则：**
 - **只提取关于"用户本人"的画像信息**，忽略用户提到的第三方人物（如朋友、同事、家人）的信息
 - 仅提取文本中明确提到的信息，不要推测
-- 如果文本中没有可提取的用户画像信息，返回空的 user_metadata 对象
 - **输出语言必须与输入文本的语言一致**（输入中文则输出中文值，输入英文则输出英文值）
 
+**增量模式（重要）：**
+你只需要输出**本次对话引起的变更操作**，不要输出完整的元数据。每个变更是一个对象，包含：
+- `field_path`：字段路径，用点号分隔（如 `profile.role`、`profile.expertise`）
+- `action`：操作类型
+  * `set`：新增或修改一个字段的值
+  * `remove`：移除一个字段的值
+- `value`：字段的新值（`action="set"` 时必填，`action="remove"` 时填要移除的元素值）
+  * 所有字段均为列表类型，每个元素一条变更记录
+
+**判断规则：**
+- 用户提到新信息 → `action="set"`，填入新值
+- 用户明确否定已有信息（如"我不再做老师了"、"我已经不学Python了"）→ `action="remove"`，`value` 填要移除的元素值
+- 如果本次对话没有任何可提取的变更，返回空的 `metadata_changes` 数组 `[]`
+- **不要为未被提及的字段生成任何变更操作**
+
 {% if existing_metadata %}
-**重要：合并已有元数据**
-下方提供了数据库中已有的用户元数据。请结合用户最新发言，输出**合并后的完整元数据**：
-- 如果用户明确否定了已有信息（如"我不再教高中物理了"），在输出中**移除**该信息
-- 如果用户提到了新信息，**添加**到对应字段中
-- 如果已有信息未被用户否定，**保留**在输出中
-- 标量字段（如 role、domain）：如果用户提到了新值，用新值替换；否则保留已有值
-- 最终输出应该是完整的、合并后的元数据，不是增量
+**已有元数据（仅供参考，用于判断是否需要变更）：**
+请对比已有数据和用户最新发言，只输出差异部分的变更操作。
+- 如果用户说的信息和已有数据一致，不需要输出变更
+- 如果用户否定了已有数据中的某个值，输出 `remove` 操作
+- 如果用户提到了新信息，输出 `set` 操作
 {% endif %}
 
 **字段说明：**
-- profile.role：用户的职业或角色，如 教师、医生、后端工程师
-- profile.domain：用户所在领域，如 教育、医疗、软件开发
-- profile.expertise：用户擅长的技能或工具（通用，不限于编程），如 Python、心理咨询、高中物理
-- profile.interests：用户主动表达兴趣的话题或领域标签
-- behavioral_hints.learning_stage：学习阶段（初学者/中级/高级）
-- behavioral_hints.preferred_depth：偏好深度（概览/技术细节/深入探讨）
-- behavioral_hints.tone_preference：语气偏好（轻松随意/专业简洁/学术严谨）
-- knowledge_tags：用户涉及的知识领域标签
+- profile.role：用户的职业或角色（列表），如 教师、医生、后端工程师，一个人可以有多个角色
+- profile.domain：用户所在领域（列表），如 教育、医疗、软件开发，一个人可以涉及多个领域
+- profile.expertise：用户擅长的技能或工具（列表），如 Python、心理咨询、高中物理
+- profile.interests：用户主动表达兴趣的话题或领域标签（列表）
 
 **用户别名变更（增量模式）：**
 - **aliases_to_add**：本次新发现的用户别名，包括：
@@ -43,7 +51,6 @@ Extract user metadata from the following conversation statements spoken by the u
 - **aliases_to_remove**：用户明确否认的别名，包括：
   * 用户说"我不叫XX了"、"别叫我XX"、"我改名了，不叫XX" → 将 XX 放入此数组
   * **严格限制**：只将用户原文中**逐字提到**的被否认名字放入，不要推断关联的其他别名
-  * 例如：用户说"我不叫陈小刀了" → 只移除"陈小刀"，不要移除"陈哥"、"老陈"等未被提及的别名
   * 如果没有要移除的别名，返回空数组 `[]`
 {% if existing_aliases %}
 - 已有别名：{{ existing_aliases | tojson }}（仅供参考，不需要在输出中重复）
@@ -57,28 +64,36 @@ Extract user metadata from the following conversation statements spoken by the u
 **Extraction rules:**
 - **Only extract profile information about the user themselves**, ignore information about third parties (friends, colleagues, family) mentioned by the user
 - Only extract information explicitly mentioned in the text, do not speculate
-- If no user profile information can be extracted, return an empty user_metadata object
 - **Output language must match the input text language**
 
+**Incremental mode (important):**
+You should only output **the change operations caused by this conversation**, not the complete metadata. Each change is an object containing:
+- `field_path`: Field path separated by dots (e.g. `profile.role`, `profile.expertise`)
+- `action`: Operation type
+  * `set`: Add or update a field value
+  * `remove`: Remove a field value
+- `value`: The new value for the field (required when `action="set"`, for `action="remove"` fill in the element value to remove)
+  * All fields are list types, one change record per element
+
+**Decision rules:**
+- User mentions new information → `action="set"`, fill in the new value
+- User explicitly negates existing info (e.g. "I'm no longer a teacher", "I stopped learning Python") → `action="remove"`, `value` is the element to remove
+- If this conversation has no extractable changes, return an empty `metadata_changes` array `[]`
+- **Do NOT generate any change operations for fields not mentioned in the conversation**
+
 {% if existing_metadata %}
-**Important: Merge with existing metadata**
-Existing user metadata from the database is provided below. Combine with the user's latest statements to output the **complete merged metadata**:
-- If the user explicitly negates existing info (e.g. "I no longer teach high school physics"), **remove** it from output
-- If the user mentions new info, **add** it to the corresponding field
-- If existing info is not negated by the user, **keep** it in the output
-- Scalar fields (e.g. role, domain): replace with new value if user mentions one; otherwise keep existing
-- The final output should be the complete, merged metadata — not an incremental update
+**Existing metadata (for reference only, to determine if changes are needed):**
+Compare existing data with the user's latest statements, and only output change operations for the differences.
+- If the user's statement matches existing data, no change is needed
+- If the user negates a value in existing data, output a `remove` operation
+- If the user mentions new information, output a `set` operation
 {% endif %}
 
 **Field descriptions:**
-- profile.role: User's occupation or role, e.g. teacher, doctor, software engineer
-- profile.domain: User's domain, e.g. education, healthcare, software development
-- profile.expertise: User's skills or tools (general, not limited to programming)
-- profile.interests: Topics or domain tags the user actively expressed interest in
-- behavioral_hints.learning_stage: Learning stage (beginner/intermediate/advanced)
-- behavioral_hints.preferred_depth: Preferred depth (overview/detailed/deep dive)
-- behavioral_hints.tone_preference: Tone preference (casual/professional/academic)
-- knowledge_tags: Knowledge domain tags related to the user
+- profile.role: User's occupation or role (list), e.g. teacher, doctor, software engineer. A person can have multiple roles
+- profile.domain: User's domain (list), e.g. education, healthcare, software development. A person can span multiple domains
+- profile.expertise: User's skills or tools (list), e.g. Python, counseling, physics
+- profile.interests: Topics or domain tags the user actively expressed interest in (list)
 
 **User alias changes (incremental mode):**
 - **aliases_to_add**: Newly discovered user aliases from this conversation, including:
@@ -90,7 +105,6 @@ Existing user metadata from the database is provided below. Combine with the use
 - **aliases_to_remove**: Aliases the user explicitly denies, including:
   * User says "Don't call me XX anymore", "I'm not called XX", "I changed my name from XX" → put XX in this array
   * **Strict rule**: Only include the exact name the user **verbatim mentions** as denied. Do NOT infer or remove related aliases
-  * Example: User says "I'm not called John anymore" → only remove "John", do NOT remove "Johnny", "J" or other related aliases not mentioned
   * If no aliases to remove, return empty array `[]`
 {% if existing_aliases %}
 - Existing aliases: {{ existing_aliases | tojson }} (for reference only, do not repeat in output)
@@ -113,20 +127,11 @@ Existing user metadata from the database is provided below. Combine with the use
 Return a JSON object with the following structure:
 ```json
 {
-  "user_metadata": {
-    "profile": {
-      "role": "",
-      "domain": "",
-      "expertise": [],
-      "interests": []
-    },
-    "behavioral_hints": {
-      "learning_stage": "",
-      "preferred_depth": "",
-      "tone_preference": ""
-    },
-    "knowledge_tags": []
-  },
+  "metadata_changes": [
+    {"field_path": "profile.role", "action": "set", "value": "后端工程师"},
+    {"field_path": "profile.expertise", "action": "set", "value": "Python"},
+    {"field_path": "profile.expertise", "action": "remove", "value": "Java"}
+  ],
   "aliases_to_add": [],
   "aliases_to_remove": []
 }

api/app/core/models/base.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 import os
-from typing import Any, Dict, Optional, TypeVar
+from typing import Any, Dict, List, Optional, TypeVar
 
 from langchain_aws import ChatBedrock
 from langchain_community.chat_models import ChatTongyi
@@ -9,12 +9,12 @@ from langchain_core.embeddings import Embeddings
 from langchain_core.language_models import BaseLLM
 from langchain_ollama import OllamaLLM
 from langchain_openai import ChatOpenAI, OpenAI
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 from app.core.error_codes import BizCode
 from app.core.exceptions import BusinessException
 from app.models.models_model import ModelProvider, ModelType
-from app.core.models.volcano_chat import VolcanoChatOpenAI
+from app.core.models.compatible_chat import CompatibleChatOpenAI
 
 T = TypeVar("T")
 
@@ -25,10 +25,11 @@ class RedBearModelConfig(BaseModel):
     provider: str
     api_key: str
     base_url: Optional[str] = None
+    capability: List[str] = Field(default_factory=list)  # 模型能力列表，驱动所有能力开关
     is_omni: bool = False  # 是否为 Omni 模型
     deep_thinking: bool = False  # 是否启用深度思考模式
     thinking_budget_tokens: Optional[int] = None  # 深度思考 token 预算
-    support_thinking: bool = False  # 模型是否支持 enable_thinking 参数（capability 含 thinking）
+    json_output: bool = False  # 是否强制 JSON 输出
     # 请求超时时间（秒）- 默认120秒以支持复杂的LLM调用，可通过环境变量 LLM_TIMEOUT 配置
     timeout: float = Field(default_factory=lambda: float(os.getenv("LLM_TIMEOUT", "120.0")))
     # 最大重试次数 - 默认2次以避免过长等待，可通过环境变量 LLM_MAX_RETRIES 配置
@@ -36,6 +37,23 @@ class RedBearModelConfig(BaseModel):
     concurrency: int = 5  # 并发限流
     extra_params: Dict[str, Any] = {}
 
+    @model_validator(mode="after")
+    def _resolve_capabilities(self) -> "RedBearModelConfig":
+        from app.core.logging_config import get_business_logger
+        logger = get_business_logger()
+        if self.deep_thinking and "thinking" not in self.capability:
+            logger.warning(
+                f"模型 {self.model_name} 不支持深度思考（capability 中无 'thinking'），已自动关闭 deep_thinking"
+            )
+            self.deep_thinking = False
+            self.thinking_budget_tokens = None
+        if self.json_output and "json_output" not in self.capability:
+            logger.warning(
+                f"模型 {self.model_name} 不支持 JSON 输出（capability 中无 'json_output'），已自动关闭 json_output"
+            )
+            self.json_output = False
+        return self
+
 
 class RedBearModelFactory:
     """模型工厂类"""
@@ -74,18 +92,19 @@ class RedBearModelFactory:
             is_streaming = bool(config.extra_params.get("streaming"))
             if is_streaming:
                 params["stream_usage"] = True
-            # 只有支持 thinking 的模型才传 enable_thinking
-            if config.support_thinking:
-                model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {})
-                if is_streaming:
-                    model_kwargs["enable_thinking"] = config.deep_thinking
-                    if config.deep_thinking:
-                        model_kwargs["incremental_output"] = True
-                        if config.thinking_budget_tokens:
-                            model_kwargs["thinking_budget"] = config.thinking_budget_tokens
-                else:
-                    model_kwargs["enable_thinking"] = False
-                params["model_kwargs"] = model_kwargs
+            # 支持 thinking 的模型始终传 enable_thinking，关闭时显式传 False 避免模型默认开启思考
+            if "thinking" in config.capability:
+                extra_body = params.setdefault("extra_body", {})
+                if config.deep_thinking:
+                    extra_body["enable_thinking"] = False
+                    if is_streaming:
+                        extra_body["enable_thinking"] = True
+                    if config.thinking_budget_tokens:
+                        extra_body["thinking_budget"] = config.thinking_budget_tokens
+            # JSON 输出模式
+            if config.json_output:
+                model_kwargs = params.setdefault("model_kwargs", {})
+                model_kwargs["response_format"] = {"type": "json_object"}
             return params
 
         if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK, ModelProvider.OLLAMA, ModelProvider.VOLCANO]:
@@ -108,26 +127,31 @@ class RedBearModelFactory:
                 **config.extra_params
             }
             # 流式模式下启用 stream_usage 以获取 token 统计
-            if config.extra_params.get("streaming"):
-                params["stream_usage"] = True
-            # 深度思考模式
             is_streaming = bool(config.extra_params.get("streaming"))
-            if is_streaming and not config.is_omni:
+            if is_streaming:
+                params["stream_usage"] = True
+            # 支持 thinking 的模型始终传 enable_thinking，关闭时显式传 False 避免模型默认开启思考
+            if "thinking" in config.capability:
+                # VOLCANO 深度思考仅流式支持
                 if provider == ModelProvider.VOLCANO:
-                    # 火山引擎深度思考仅流式调用支持，非流式时不传 thinking 参数
-                    thinking_config: Dict[str, Any] = {
-                        "type": "enabled" if config.deep_thinking else "disabled"
-                    }
+                    thinking_config: Dict[str, Any] = {"type": "enabled" if config.deep_thinking else "disabled"}
                     if config.deep_thinking and config.thinking_budget_tokens:
                         thinking_config["budget_tokens"] = config.thinking_budget_tokens
                     params["extra_body"] = {"thinking": thinking_config}
                 else:
-                    # 始终显式传递 enable_thinking，不支持该参数的模型（如 DeepSeek-R1）会直接忽略
-                    model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {})
-                    model_kwargs["enable_thinking"] = config.deep_thinking
-                    if config.deep_thinking and config.thinking_budget_tokens:
-                        model_kwargs["thinking_budget"] = config.thinking_budget_tokens
-                    params["model_kwargs"] = model_kwargs
+                    extra_body = params.setdefault("extra_body", {})
+                    if config.deep_thinking:
+                        extra_body["enable_thinking"] = False
+                        if is_streaming:
+                            extra_body["enable_thinking"] = True
+                        if config.thinking_budget_tokens:
+                            extra_body["thinking_budget"] = config.thinking_budget_tokens
+            # JSON 输出模式
+            if config.json_output:
+                model_kwargs = params.setdefault("model_kwargs", {})
+                # VOLCANO 模型不支持 response_format，JSON 输出由 system prompt 注入实现
+                if provider != ModelProvider.VOLCANO:
+                    model_kwargs["response_format"] = {"type": "json_object"}
             return params
         elif provider == ModelProvider.DASHSCOPE:
             params = {
@@ -136,19 +160,20 @@ class RedBearModelFactory:
                 "max_retries": config.max_retries,
                 **config.extra_params
             }
-            # 只有支持 thinking 的模型才传 enable_thinking
-            if config.support_thinking:
+            # 支持 thinking 的模型始终传 enable_thinking，关闭时显式传 False 避免模型默认开启思考
+            if "thinking" in config.capability:
                 is_streaming = bool(config.extra_params.get("streaming"))
-                model_kwargs: Dict[str, Any] = config.extra_params.get("model_kwargs", {})
-                if is_streaming:
-                    model_kwargs["enable_thinking"] = config.deep_thinking
-                    if config.deep_thinking:
-                        model_kwargs["incremental_output"] = True
-                        if config.thinking_budget_tokens:
-                            model_kwargs["thinking_budget"] = config.thinking_budget_tokens
-                else:
+                model_kwargs = params.setdefault("model_kwargs", {})
+                if config.deep_thinking:
                     model_kwargs["enable_thinking"] = False
-                params["model_kwargs"] = model_kwargs
+                    if is_streaming:
+                        model_kwargs["enable_thinking"] = True
+                        model_kwargs["incremental_output"] = True
+                    if config.thinking_budget_tokens:
+                        model_kwargs["thinking_budget"] = config.thinking_budget_tokens
+            if config.json_output:
+                model_kwargs = params.setdefault("model_kwargs", {})
+                model_kwargs["response_format"] = {"type": "json_object"}
             return params
         elif provider == ModelProvider.BEDROCK:
             # Bedrock 使用 AWS 凭证
@@ -195,6 +220,10 @@ class RedBearModelFactory:
                 params["additional_model_request_fields"] = {
                     "thinking": {"type": "enabled", "budget_tokens": budget}
                 }
+            # JSON 输出模式
+            if config.json_output:
+                model_kwargs = params.setdefault("model_kwargs", {})
+                model_kwargs["response_format"] = {"type": "json_object"}
             return params
         else:
             raise BusinessException(f"不支持的提供商: {provider}", code=BizCode.PROVIDER_NOT_SUPPORTED)
@@ -223,18 +252,19 @@ def get_provider_llm_class(config: RedBearModelConfig, type: ModelType = ModelTy
     """根据模型提供商获取对应的模型类"""
     provider = config.provider.lower()
 
-    # dashscope 的 omni 模型使用 OpenAI 兼容模式
+    # dashscope的omni模型 和 volcano模型使用
     if provider == ModelProvider.DASHSCOPE and config.is_omni:
-        return ChatOpenAI
+        return CompatibleChatOpenAI
     if provider == ModelProvider.VOLCANO:
-        return VolcanoChatOpenAI
+        return CompatibleChatOpenAI
     if provider in [ModelProvider.OPENAI, ModelProvider.XINFERENCE, ModelProvider.GPUSTACK]:
-        if type == ModelType.LLM:
-            return OpenAI
-        elif type == ModelType.CHAT:
-            return ChatOpenAI
-        else:
-            raise BusinessException(f"不支持的模型提供商及类型: {provider}-{type}", code=BizCode.PROVIDER_NOT_SUPPORTED)
+        return CompatibleChatOpenAI
+        # if type == ModelType.LLM:
+        #     return OpenAI
+        # elif type == ModelType.CHAT:
+        #     return CompatibleChatOpenAI
+        # else:
+        #     raise BusinessException(f"不支持的模型提供商及类型: {provider}-{type}", code=BizCode.PROVIDER_NOT_SUPPORTED)
     elif provider == ModelProvider.DASHSCOPE:
         return ChatTongyi
     elif provider == ModelProvider.OLLAMA:

api/app/core/models/compatible_chat.py

@@ -8,12 +8,33 @@ from __future__ import annotations
 
 from typing import Any, Optional, Union
 
+from langchain_core.messages import BaseMessage
 from langchain_core.outputs import ChatGenerationChunk, ChatResult
 from langchain_openai import ChatOpenAI
 
 
-class VolcanoChatOpenAI(ChatOpenAI):
-    """火山引擎 Chat 模型，支持深度思考内容（reasoning_content）的流式和非流式透传。"""
+class CompatibleChatOpenAI(ChatOpenAI):
+    """火山和千问的omni兼容模型，支持深度思考内容（reasoning_content）的流式和非流式透传。
+
+    同时修复 json_output + tools 同时使用时 langchain_openai 强制走 .parse()/.stream()
+    导致 strict 校验报错的问题：有工具时从 payload 中移除 response_format，
+    让父类走普通 .create()/.astream() 路径，JSON 输出由 system prompt 指令保证。
+    """
+
+    def _get_request_payload(
+        self,
+        input_: list[BaseMessage],
+        *,
+        stop: list[str] | None = None,
+        **kwargs: Any,
+    ) -> dict:
+        payload = super()._get_request_payload(input_, stop=stop, **kwargs)
+        # 有工具时 langchain_openai 检测到 response_format 会切换到 .parse()/.stream()
+        # 接口，OpenAI SDK 要求此时所有工具必须 strict=True，动态生成的工具不满足。
+        # 移除 response_format，让父类走普通路径，JSON 输出由 system prompt 指令保证。
+        if payload.get("tools") and "response_format" in payload:
+            payload.pop("response_format")
+        return payload
 
     def _create_chat_result(self, response: Union[dict, Any], generation_info: Optional[dict] = None) -> ChatResult:
         result = super()._create_chat_result(response, generation_info)

api/app/core/models/scripts/bedrock_models.yaml

@@ -6,7 +6,8 @@ models:
   description: AI21 Labs大语言模型，completion生成模式，256000上下文窗口
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -20,6 +21,7 @@ models:
   is_official: true
   capability:
     - vision
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -38,6 +40,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -54,7 +57,8 @@ models:
   description: Cohere大语言模型，支持智能体思考、工具调用、流式工具调用，128000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -72,6 +76,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -87,7 +92,8 @@ models:
   description: Meta Llama大语言模型，支持智能体思考、工具调用，128000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -101,7 +107,8 @@ models:
   description: Mistral AI大语言模型，支持智能体思考、工具调用，32000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -115,7 +122,8 @@ models:
   description: OpenAI大语言模型，支持智能体思考、工具调用、流式工具调用，32768上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -130,7 +138,8 @@ models:
   description: Qwen大语言模型，支持智能体思考、工具调用、流式工具调用，32768上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型

api/app/core/models/scripts/dashscope_models.yaml

@@ -8,6 +8,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -22,6 +23,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -36,6 +38,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -48,7 +51,8 @@ models:
   description: DeepSeek-V3.1大语言模型，支持智能体思考，131072超大上下文窗口，对话模式，支持丰富生成参数调节
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -61,7 +65,8 @@ models:
   description: DeepSeek-V3.2-exp实验版大语言模型，支持智能体思考，131072超大上下文窗口，对话模式，支持丰富生成参数调节
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -74,7 +79,8 @@ models:
   description: DeepSeek-V3.2大语言模型，支持智能体思考，131072超大上下文窗口，对话模式，支持丰富生成参数调节
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -87,7 +93,8 @@ models:
   description: DeepSeek-V3大语言模型，支持智能体思考，64000上下文窗口，对话模式，支持文本与JSON格式输出
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -100,7 +107,8 @@ models:
   description: farui-plus大语言模型，支持多工具调用、智能体思考、流式工具调用，12288上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -115,7 +123,8 @@ models:
   description: GLM-4.7大语言模型，支持多工具调用、智能体思考、流式工具调用，202752超大上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -133,6 +142,7 @@ models:
   capability:
   - vision
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -150,6 +160,7 @@ models:
   capability:
   - vision
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -180,6 +191,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -210,7 +222,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -376,6 +388,7 @@ models:
   capability:
   - vision
   - video
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -448,6 +461,7 @@ models:
   capability:
   - vision
   - video
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -466,6 +480,7 @@ models:
   capability:
   - vision
   - video
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -481,7 +496,8 @@ models:
   description: qwen2.5-0.5b-instruct大语言模型，支持多工具调用、智能体思考、流式工具调用，32768上下文窗口，对话模式，未废弃
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -498,6 +514,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -513,7 +530,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -530,6 +547,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -546,6 +564,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -561,7 +580,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -578,6 +597,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -594,6 +614,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -610,6 +631,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -626,6 +648,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -641,7 +664,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -656,7 +679,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -672,6 +695,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -687,6 +711,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -702,6 +727,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -719,6 +745,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -736,6 +763,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -752,6 +780,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -768,7 +797,7 @@ models:
   is_deprecated: false
   is_official: true
   capability:
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -785,6 +814,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -803,6 +833,8 @@ models:
   - vision
   - video
   - audio
+  - thinking
+  - json_output
   is_omni: true
   tags:
   - 大语言模型
@@ -822,7 +854,7 @@ models:
   capability:
   - vision
   - video
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -844,6 +876,7 @@ models:
   - vision
   - video
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -864,7 +897,7 @@ models:
   capability:
   - vision
   - video
-  - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -886,6 +919,7 @@ models:
   - vision
   - video
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -907,6 +941,7 @@ models:
   - vision
   - video
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -928,6 +963,7 @@ models:
   - vision
   - video
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -947,6 +983,7 @@ models:
   - vision
   - video
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -964,6 +1001,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -979,6 +1017,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -994,6 +1033,7 @@ models:
   is_official: true
   capability:
   - thinking
+  - json_output
   is_omni: false
   tags:
   - 大语言模型

api/app/core/models/scripts/openai_models.yaml

@@ -10,6 +10,7 @@ models:
     - vision
     - audio
     - video
+    - json_output
   is_omni: true
   tags:
   - 大语言模型
@@ -27,7 +28,8 @@ models:
   description: gpt-3.5-turbo-0125大语言模型，支持多工具调用、智能体思考、流式工具调用，16385上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -42,7 +44,8 @@ models:
   description: gpt-3.5-turbo-1106大语言模型，支持多工具调用、智能体思考、流式工具调用，16385上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -57,7 +60,8 @@ models:
   description: gpt-3.5-turbo-16k大语言模型，支持多工具调用、智能体思考、流式工具调用，16385上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -84,7 +88,8 @@ models:
   description: gpt-3.5-turbo大语言模型，支持多工具调用、智能体思考、流式工具调用，16385上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -99,7 +104,8 @@ models:
   description: gpt-4-0125-preview大语言模型，支持多工具调用、智能体思考、流式工具调用，128000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -114,7 +120,8 @@ models:
   description: gpt-4-1106-preview大语言模型，支持多工具调用、智能体思考、流式工具调用，128000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -131,6 +138,7 @@ models:
   is_official: true
   capability:
     - vision
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -146,7 +154,8 @@ models:
   description: gpt-4-turbo-preview大语言模型，支持多工具调用、智能体思考、流式工具调用，128000上下文窗口，对话模式
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -163,6 +172,7 @@ models:
   is_official: true
   capability:
     - vision
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -194,6 +204,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -213,6 +224,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -231,6 +243,7 @@ models:
   is_official: true
   capability:
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -248,6 +261,7 @@ models:
   is_official: true
   capability:
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -266,6 +280,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -284,6 +299,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -302,6 +318,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -321,6 +338,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -340,6 +358,7 @@ models:
   capability:
     - vision
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型

api/app/core/models/scripts/volcano_models.yaml

@@ -11,6 +11,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -26,6 +27,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -41,6 +43,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -56,6 +59,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -72,6 +76,7 @@ models:
   capability:
     - vision
     - video
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -87,6 +92,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -102,6 +108,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -117,6 +124,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -132,6 +140,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -148,6 +157,7 @@ models:
     - vision
     - video
     - thinking
+    - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -175,7 +185,8 @@ models:
   description: 全新一代主力模型，性能全面升级，在知识、代码、推理等方面表现卓越。最大支持 128k 上下文窗口，输出长度支持最大 12k tokens。
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型
@@ -187,7 +198,8 @@ models:
   description: 全新一代轻量版模型，极致响应速度，效果与时延均达到全球一流水平。支持 32k 上下文窗口，输出长度支持最大 12k tokens。
   is_deprecated: false
   is_official: true
-  capability: []
+  capability:
+  - json_output
   is_omni: false
   tags:
   - 大语言模型

api/app/core/quota_manager.py

@@ -0,0 +1,485 @@
+"""
+统一配额管理器 - 社区版和 SaaS 版共用
+
+配额来源策略：
+1. 优先从 premium 模块的 tenant_subscriptions 表读取（SaaS 版）
+2. 降级到 default_free_plan.py 配置文件（社区版兜底）
+"""
+import asyncio
+import time
+from functools import wraps
+from typing import Optional, Callable, Dict, Any
+from uuid import UUID
+
+from sqlalchemy import func
+from sqlalchemy.orm import Session
+
+from app.core.logging_config import get_auth_logger
+from app.i18n.exceptions import QuotaExceededError
+
+logger = get_auth_logger()
+
+
+def _get_user_from_kwargs(kwargs: dict):
+    """从 kwargs 中获取 user 对象"""
+    for key in ["user", "current_user"]:
+        if key in kwargs:
+            return kwargs[key]
+    return None
+
+
+def _get_tenant_id_from_kwargs(db: Session, kwargs: dict):
+    """从 kwargs 中获取 tenant_id"""
+    user = _get_user_from_kwargs(kwargs)
+    if user and hasattr(user, 'tenant_id'):
+        return user.tenant_id
+
+    workspace_id = kwargs.get("workspace_id")
+    if workspace_id:
+        from app.models.workspace_model import Workspace
+        workspace = db.query(Workspace).filter(Workspace.id == workspace_id).first()
+        if workspace:
+            return workspace.tenant_id
+
+    api_key_auth = kwargs.get("api_key_auth")
+    if api_key_auth and hasattr(api_key_auth, 'workspace_id'):
+        from app.models.workspace_model import Workspace
+        workspace = db.query(Workspace).filter(Workspace.id == api_key_auth.workspace_id).first()
+        if workspace:
+            return workspace.tenant_id
+
+    data = kwargs.get("data") or kwargs.get("body") or kwargs.get("payload")
+    if data and hasattr(data, "workspace_id"):
+        from app.models.workspace_model import Workspace
+        workspace = db.query(Workspace).filter(Workspace.id == data.workspace_id).first()
+        if workspace:
+            return workspace.tenant_id
+
+    share_data = kwargs.get("share_data")
+    if share_data and hasattr(share_data, 'share_token'):
+        from app.models.workspace_model import Workspace
+        from app.models.app_model import App
+        share_token = share_data.share_token
+        from app.models.release_share_model import ReleaseShare
+        share_record = db.query(ReleaseShare).filter(ReleaseShare.share_token == share_token).first()
+        if share_record:
+            app = db.query(App).filter(App.id == share_record.app_id, App.is_active.is_(True)).first()
+            if app:
+                return app.workspace.tenant_id
+
+    return None
+
+
+def _get_quota_config(db: Session, tenant_id: UUID) -> Optional[Dict[str, Any]]:
+    """
+    获取租户的配额配置
+    
+    优先级：
+    1. premium 模块的 tenant_subscriptions（SaaS 版）
+    2. default_free_plan.py 配置文件（社区版兜底）
+    """
+    # 尝试从 premium 模块获取
+    try:
+        from premium.platform_admin.package_plan_service import TenantSubscriptionService
+        quota_config = TenantSubscriptionService(db).get_effective_quota(tenant_id)
+        if quota_config:
+            logger.debug(f"从 premium 模块获取租户 {tenant_id} 配额配置")
+            return quota_config
+    except (ModuleNotFoundError, ImportError, Exception) as e:
+        logger.debug(f"无法从 premium 模块获取配额配置: {e}")
+
+    # 降级到配置文件
+    try:
+        from app.config.default_free_plan import DEFAULT_FREE_PLAN
+        logger.info(f"使用配置文件中的免费套餐配额: tenant={tenant_id}")
+        return DEFAULT_FREE_PLAN.get("quotas")
+    except Exception as e:
+        logger.error(f"无法从配置文件获取配额: {e}")
+        return None
+
+
+class QuotaUsageRepository:
+    """配额使用量数据访问层"""
+
+    def __init__(self, db: Session):
+        self.db = db
+
+    def count_workspaces(self, tenant_id: UUID) -> int:
+        from app.models.workspace_model import Workspace
+        return self.db.query(Workspace).filter(
+            Workspace.tenant_id == tenant_id,
+            Workspace.is_active.is_(True)
+        ).count()
+
+    def count_apps(self, tenant_id: UUID) -> int:
+        from app.models.app_model import App
+        from app.models.workspace_model import Workspace
+        return self.db.query(App).join(
+            Workspace, App.workspace_id == Workspace.id
+        ).filter(
+            Workspace.tenant_id == tenant_id,
+            App.is_active.is_(True)
+        ).count()
+
+    def count_skills(self, tenant_id: UUID) -> int:
+        from app.models.skill_model import Skill
+        return self.db.query(Skill).filter(
+            Skill.tenant_id == tenant_id,
+            Skill.is_active.is_(True)
+        ).count()
+
+    def sum_knowledge_capacity_gb(self, tenant_id: UUID) -> float:
+        from app.models.document_model import Document
+        from app.models.knowledge_model import Knowledge
+        from app.models.workspace_model import Workspace
+        result = self.db.query(func.coalesce(func.sum(Document.file_size), 0)).join(
+            Knowledge, Document.kb_id == Knowledge.id
+        ).join(
+            Workspace, Knowledge.workspace_id == Workspace.id
+        ).filter(
+            Workspace.tenant_id == tenant_id,
+            Document.status == 1,
+        ).scalar()
+        return float(result) / (1024 ** 3) if result else 0.0
+
+    def count_memory_engines(self, tenant_id: UUID) -> int:
+        from app.models.memory_config_model import MemoryConfig
+        from app.models.workspace_model import Workspace
+        return self.db.query(MemoryConfig).join(
+            Workspace, MemoryConfig.workspace_id == Workspace.id
+        ).filter(
+            Workspace.tenant_id == tenant_id
+        ).count()
+
+    def count_end_users(self, tenant_id: UUID) -> int:
+        from app.models.end_user_model import EndUser
+        from app.models.workspace_model import Workspace
+        return self.db.query(EndUser).join(
+            Workspace, EndUser.workspace_id == Workspace.id
+        ).filter(
+            Workspace.tenant_id == tenant_id
+        ).count()
+
+    def count_models(self, tenant_id: UUID) -> int:
+        from app.models.models_model import ModelConfig
+        return self.db.query(ModelConfig).filter(
+            ModelConfig.tenant_id == tenant_id,
+            ModelConfig.is_active == True
+        ).count()
+
+    def count_ontology_projects(self, tenant_id: UUID) -> int:
+        from app.models.ontology_scene import OntologyScene
+        from app.models.workspace_model import Workspace
+        return self.db.query(OntologyScene).join(
+            Workspace, OntologyScene.workspace_id == Workspace.id
+        ).filter(
+            Workspace.tenant_id == tenant_id
+        ).count()
+
+    def get_usage_by_quota_type(self, tenant_id: UUID, quota_type: str):
+        """按配额类型分发，返回当前使用量"""
+        dispatch = {
+            "workspace_quota": self.count_workspaces,
+            "app_quota": self.count_apps,
+            "skill_quota": self.count_skills,
+            "knowledge_capacity_quota": self.sum_knowledge_capacity_gb,
+            "memory_engine_quota": self.count_memory_engines,
+            "end_user_quota": self.count_end_users,
+            "model_quota": self.count_models,
+            "ontology_project_quota": self.count_ontology_projects,
+        }
+        fn = dispatch.get(quota_type)
+        return fn(tenant_id) if fn else 0
+
+
+def _check_quota(
+    db: Session,
+    tenant_id: UUID,
+    quota_type: str,
+    resource_name: str,
+    usage_func: Optional[Callable] = None,
+) -> None:
+    """核心配额检查逻辑：对比使用量和配额限制"""
+    try:
+        quota_config = _get_quota_config(db, tenant_id)
+        if not quota_config:
+            logger.warning(f"租户 {tenant_id} 无有效配额配置，跳过配额检查")
+            return
+
+        quota_limit = quota_config.get(quota_type)
+        if quota_limit is None:
+            logger.warning(f"配额配置未包含 {quota_type}，跳过配额检查")
+            return
+
+        if usage_func:
+            current_usage = usage_func(db, tenant_id)
+        else:
+            current_usage = QuotaUsageRepository(db).get_usage_by_quota_type(tenant_id, quota_type)
+
+        if current_usage >= quota_limit:
+            logger.warning(
+                f"配额不足: tenant={tenant_id}, type={quota_type}, "
+                f"usage={current_usage}, limit={quota_limit}"
+            )
+            raise QuotaExceededError(
+                resource=resource_name,
+                current_usage=current_usage,
+                quota_limit=quota_limit,
+            )
+
+        logger.debug(
+            f"配额检查通过: tenant={tenant_id}, type={quota_type}, "
+            f"usage={current_usage}, limit={quota_limit}"
+        )
+
+    except QuotaExceededError:
+        raise
+    except Exception as e:
+        logger.error(
+            f"配额检查异常: tenant={tenant_id}, type={quota_type}, "
+            f"error_type={type(e).__name__}, error={str(e)}",
+            exc_info=True,
+        )
+        raise
+
+
+# ─── 具名装饰器 ────────────────────────────────────────────────────────────
+
+def check_workspace_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "workspace_quota", "workspace")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_skill_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "skill_quota", "skill")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_app_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "app_quota", "app")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_knowledge_capacity_quota(func: Callable) -> Callable:
+    @wraps(func)
+    async def async_wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        if not db:
+            logger.warning("配额检查失败：缺少 db 参数")
+            return await func(*args, **kwargs)
+        tenant_id = _get_tenant_id_from_kwargs(db, kwargs)
+        if not tenant_id:
+            logger.warning("配额检查失败：无法获取 tenant_id")
+            return await func(*args, **kwargs)
+        _check_quota(db, tenant_id, "knowledge_capacity_quota", "knowledge_capacity")
+        return await func(*args, **kwargs)
+
+    @wraps(func)
+    def sync_wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "knowledge_capacity_quota", "knowledge_capacity")
+        return func(*args, **kwargs)
+
+    return async_wrapper if asyncio.iscoroutinefunction(func) else sync_wrapper
+
+
+def check_memory_engine_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "memory_engine_quota", "memory_engine")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_end_user_quota(func: Callable) -> Callable:
+    @wraps(func)
+    async def async_wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        if not db:
+            logger.warning("配额检查失败：缺少 db 参数")
+            return await func(*args, **kwargs)
+        tenant_id = _get_tenant_id_from_kwargs(db, kwargs)
+        if not tenant_id:
+            logger.warning("配额检查失败：无法获取 tenant_id")
+            return await func(*args, **kwargs)
+        _check_quota(db, tenant_id, "end_user_quota", "end_user")
+        return await func(*args, **kwargs)
+
+    @wraps(func)
+    def sync_wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        if not db:
+            logger.warning("配额检查失败：缺少 db 参数")
+            return func(*args, **kwargs)
+        tenant_id = _get_tenant_id_from_kwargs(db, kwargs)
+        if not tenant_id:
+            logger.warning("配额检查失败：无法获取 tenant_id")
+            return func(*args, **kwargs)
+        _check_quota(db, tenant_id, "end_user_quota", "end_user")
+        return func(*args, **kwargs)
+
+    return async_wrapper if asyncio.iscoroutinefunction(func) else sync_wrapper
+
+
+def check_ontology_project_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "ontology_project_quota", "ontology_project")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_model_quota(func: Callable) -> Callable:
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        _check_quota(db, user.tenant_id, "model_quota", "model")
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_model_activation_quota(func: Callable) -> Callable:
+    """模型激活时的配额检查装饰器"""
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        db: Session = kwargs.get("db")
+        user = _get_user_from_kwargs(kwargs)
+        if not db or not user:
+            logger.warning("配额检查失败：缺少 db 或 user 参数")
+            return func(*args, **kwargs)
+        
+        model_id = kwargs.get("model_id") or (args[1] if len(args) > 1 else None)
+        model_data = kwargs.get("model_data")
+        
+        if not model_id or not model_data:
+            logger.warning("模型激活配额检查失败：缺少 model_id 或 model_data 参数")
+            return func(*args, **kwargs)
+        
+        if model_data.is_active is True:
+            try:
+                from app.models.models_model import ModelConfig
+                from app.services.model_service import ModelConfigService
+                
+                existing_model = ModelConfigService.get_model_by_id(
+                    db=db, 
+                    model_id=model_id, 
+                    tenant_id=user.tenant_id
+                )
+                
+                if not existing_model.is_active:
+                    logger.info(f"模型激活操作，检查配额: model_id={model_id}, tenant_id={user.tenant_id}")
+                    _check_quota(db, user.tenant_id, "model_quota", "model")
+            except Exception as e:
+                logger.error(f"模型激活配额检查异常: model_id={model_id}, error={str(e)}")
+                raise
+        
+        return func(*args, **kwargs)
+    return wrapper
+
+
+def check_quota(quota_type: str, resource_name: str, usage_func: Optional[Callable] = None):
+    """通用配额检查装饰器，支持自定义使用量获取函数"""
+    def decorator(func: Callable) -> Callable:
+        @wraps(func)
+        def wrapper(*args, **kwargs):
+            db: Session = kwargs.get("db")
+            user = _get_user_from_kwargs(kwargs)
+            if not db or not user:
+                logger.warning("配额检查失败：缺少 db 或 user 参数")
+                return func(*args, **kwargs)
+            _check_quota(db, user.tenant_id, quota_type, resource_name, usage_func)
+            return func(*args, **kwargs)
+        return wrapper
+    return decorator
+
+
+# ─── 配额使用统计 ────────────────────────────────────────────────────────────
+
+def get_quota_usage(db: Session, tenant_id: UUID) -> dict:
+    """获取租户所有配额的使用情况"""
+    quota_config = _get_quota_config(db, tenant_id)
+    if not quota_config:
+        return {}
+    
+    repo = QuotaUsageRepository(db)
+
+    def pct(used, limit):
+        return round(used / limit * 100, 1) if limit else None
+
+    workspace_count = repo.count_workspaces(tenant_id)
+    skill_count = repo.count_skills(tenant_id)
+    app_count = repo.count_apps(tenant_id)
+    knowledge_gb = repo.sum_knowledge_capacity_gb(tenant_id)
+    memory_count = repo.count_memory_engines(tenant_id)
+    end_user_count = repo.count_end_users(tenant_id)
+    model_count = repo.count_models(tenant_id)
+    ontology_count = repo.count_ontology_projects(tenant_id)
+
+    api_ops_current = 0
+    try:
+        from app.core.config import settings
+        import redis
+        _now = time.time()
+        _rk = f"rate_limit:tenant_qps:{tenant_id}"
+        _r = redis.StrictRedis(
+            host=settings.REDIS_HOST, port=settings.REDIS_PORT,
+            db=settings.REDIS_DB, password=settings.REDIS_PASSWORD,
+            decode_responses=True
+        )
+        api_ops_current = int(_r.zcount(_rk, _now - 1, "+inf"))
+    except Exception:
+        pass
+
+    return {
+        "workspace": {"used": workspace_count, "limit": quota_config.get("workspace_quota"), "percentage": pct(workspace_count, quota_config.get("workspace_quota"))},
+        "skill": {"used": skill_count, "limit": quota_config.get("skill_quota"), "percentage": pct(skill_count, quota_config.get("skill_quota"))},
+        "app": {"used": app_count, "limit": quota_config.get("app_quota"), "percentage": pct(app_count, quota_config.get("app_quota"))},
+        "knowledge_capacity": {"used": round(knowledge_gb, 2), "limit": quota_config.get("knowledge_capacity_quota"), "percentage": pct(knowledge_gb, quota_config.get("knowledge_capacity_quota")), "unit": "GB"},
+        "memory_engine": {"used": memory_count, "limit": quota_config.get("memory_engine_quota"), "percentage": pct(memory_count, quota_config.get("memory_engine_quota"))},
+        "end_user": {"used": end_user_count, "limit": quota_config.get("end_user_quota"), "percentage": pct(end_user_count, quota_config.get("end_user_quota"))},
+        "ontology_project": {"used": ontology_count, "limit": quota_config.get("ontology_project_quota"), "percentage": pct(ontology_count, quota_config.get("ontology_project_quota"))},
+        "model": {"used": model_count, "limit": quota_config.get("model_quota"), "percentage": pct(model_count, quota_config.get("model_quota"))},
+        "api_ops_rate_limit": {"current": api_ops_current, "limit": quota_config.get("api_ops_rate_limit"), "percentage": None, "unit": "次/秒"},
+    }

api/app/core/quota_stub.py

@@ -0,0 +1,36 @@
+"""
+配额检查 stub - 社区版和 SaaS 版统一使用 core.quota_manager 实现
+
+所有配额检查逻辑统一在 core 层实现，两个版本共用：
+- 社区版：从 default_free_plan.py 读取配额限制
+- SaaS 版：优先从 tenant_subscriptions 表读取，降级到配置文件
+"""
+from app.core.quota_manager import (
+    check_workspace_quota,
+    check_skill_quota,
+    check_app_quota,
+    check_knowledge_capacity_quota,
+    check_memory_engine_quota,
+    check_end_user_quota,
+    check_ontology_project_quota,
+    check_model_quota,
+    check_model_activation_quota,
+    get_quota_usage,
+    _check_quota,
+    QuotaUsageRepository,
+)
+
+__all__ = [
+    "check_workspace_quota",
+    "check_skill_quota",
+    "check_app_quota",
+    "check_knowledge_capacity_quota",
+    "check_memory_engine_quota",
+    "check_end_user_quota",
+    "check_ontology_project_quota",
+    "check_model_quota",
+    "check_model_activation_quota",
+    "get_quota_usage",
+    "_check_quota",
+    "QuotaUsageRepository",
+]

api/app/core/rag/app/naive.py

@@ -672,10 +672,15 @@ def chunk(filename, binary=None, from_page=0, to_page=100000,
         excel_parser = ExcelParser()
         if parser_config.get("html4excel") and parser_config.get("html4excel").lower() == "true":
             sections = [(_, "") for _ in excel_parser.html(binary, 12) if _]
-            parser_config["chunk_token_num"] = 0
         else:
             sections = [(_, "") for _ in excel_parser(binary) if _]
-        parser_config["chunk_token_num"] = 12800
+        callback(0.8, "Finish parsing.")
+        # Excel 每行直接作为一个 chunk，不经过 naive_merge 避免被 delimiter 拆分
+        chunks = [s for s, _ in sections]
+        res.extend(tokenize_chunks(chunks, doc, is_english, None))
+        res.extend(embed_res)
+        res.extend(url_res)
+        return res
 
     elif re.search(r"\.(txt|py|js|java|c|cpp|h|php|go|ts|sh|cs|kt|sql)$", filename, re.IGNORECASE):
         callback(0.1, "Start to parse.")

api/app/core/rag/deepdoc/parser/excel_parser.py

@@ -232,14 +232,14 @@ class RAGExcelParser:
                         t = str(ti[i].value) if i < len(ti) else ""
                         t += ("：" if t else "") + str(c.value)
                         fields.append(t)
-                    line = "; ".join(fields)
+                    line = "\n".join(fields)
                     if sheetname.lower().find("sheet") < 0:
-                        line += " ——" + sheetname
+                        line += "\n——" + sheetname
                     res.append(line)
             else:
                 # 只有表头的情况
                 if header_fields:
-                    line = "; ".join(header_fields)
+                    line = "\n".join(header_fields)
                     if sheetname.lower().find("sheet") < 0:
                         line += " ——" + sheetname
                     res.append(line)

api/app/core/rag/llm/embedding_model.py

@@ -50,7 +50,9 @@ class OpenAIEmbed(Base):
     def encode(self, texts: list):
         # OpenAI requires batch size <=16
         batch_size = 16
-        texts = [truncate(t, 8191) for t in texts]
+        # Use 8000 instead of 8191 to leave safety margin for tokenizer differences
+        # between cl100k_base (used by truncate) and the actual embedding model
+        texts = [truncate(t, 8000) for t in texts]
         ress = []
         total_tokens = 0
         for i in range(0, len(texts), batch_size):
@@ -63,7 +65,7 @@ class OpenAIEmbed(Base):
         return np.array(ress), total_tokens
 
     def encode_queries(self, text):
-        res = self.client.embeddings.create(input=[truncate(text, 8191)], model=self.model_name, encoding_format="float",extra_body={"drop_params": True})
+        res = self.client.embeddings.create(input=[truncate(text, 8000)], model=self.model_name, encoding_format="float",extra_body={"drop_params": True})
         return np.array(res.data[0].embedding), self.total_token_count(res)
 
 
@@ -79,6 +81,7 @@ class LocalAIEmbed(Base):
 
     def encode(self, texts: list):
         batch_size = 16
+        texts = [truncate(t, 8000) for t in texts]
         ress = []
         for i in range(0, len(texts), batch_size):
             res = self.client.embeddings.create(input=texts[i : i + batch_size], model=self.model_name)
@@ -173,6 +176,7 @@ class XinferenceEmbed(Base):
 
     def encode(self, texts: list):
         batch_size = 16
+        texts = [truncate(t, 8000) for t in texts]
         ress = []
         total_tokens = 0
         for i in range(0, len(texts), batch_size):
@@ -188,7 +192,7 @@ class XinferenceEmbed(Base):
     def encode_queries(self, text):
         res = None
         try:
-            res = self.client.embeddings.create(input=[text], model=self.model_name)
+            res = self.client.embeddings.create(input=[truncate(text, 8000)], model=self.model_name)
             return np.array(res.data[0].embedding), self.total_token_count(res)
         except Exception as _e:
             log_exception(_e, res)

api/app/core/tools/builtin/datetime_tool.py

@@ -253,9 +253,9 @@ class DateTimeTool(BuiltinTool):
         return {
             "datetime": input_value,
             "timezone": timezone_str,
-            "timestamp": int(dt.timestamp()) * 1000,
+            "timestamp": int(dt.timestamp() * 1000),
             "iso_format": dt.isoformat(),
-            "result_data": int(dt.timestamp()) * 1000
+            "result_data": int(dt.timestamp() * 1000)
         }
 
     def _calculate_datetime(self, kwargs) -> dict:

api/app/core/workflow/engine/variable_pool.py

@@ -201,12 +201,15 @@ class VariablePool:
 
     @staticmethod
     def _extract_field(struct: "VariableStruct", field: str | None) -> Any:
-        """If field is given, drill into a dict/object variable's value."""
+        """If field is given, drill into a dict/object/array[file] variable's value."""
         if field is None:
             return struct.instance.get_value()
         value = struct.instance.get_value()
+        # array[file]: extract the field from every element, return a list
+        if isinstance(value, list):
+            return [item.get(field) if isinstance(item, dict) else getattr(item, field, None) for item in value]
         if not isinstance(value, dict):
-            raise KeyError(f"Variable is not an object, cannot access field '{field}'")
+            raise KeyError(f"Variable is not an object or array, cannot access field '{field}'")
         return value.get(field)
 
     def get_instance(

api/app/core/workflow/nodes/cycle_graph/iteration.py

@@ -28,86 +28,135 @@ class IterationRuntime:
 
     def __init__(
             self,
-            start_id: str,
             stream: bool,
-            graph: CompiledStateGraph,
             node_id: str,
             config: dict[str, Any],
             state: WorkflowState,
             variable_pool: VariablePool,
-            child_variable_pool: VariablePool,
+            cycle_nodes: list,
+            cycle_edges: list,
     ):
         """
         Initialize the iteration runtime.
 
         Args:
-            graph: Compiled workflow graph capable of async invocation.
-            node_id: Unique identifier of the loop node.
-            config: Dictionary containing iteration node configuration.
-            state: Current workflow state at the point of iteration.
+            stream:       Whether to run in streaming mode. When True, each iteration
+                          uses graph.astream and emits cycle_item events in real time.
+                          When False, graph.ainvoke is used instead.
+            node_id:      The unique identifier of the iteration node in the workflow.
+                          Also used as the variable namespace for item/index inside
+                          the subgraph (e.g. {{ node_id.item }}).
+            config:       Raw configuration dict for the iteration node, parsed into
+                          IterationNodeConfig. Controls input/output variable selectors,
+                          parallel execution settings, and output flattening.
+            state:        The parent workflow state at the point the iteration node is
+                          entered. Each task receives a copy of this state as its
+                          starting point.
+            variable_pool: The parent VariablePool containing all variables available
+                           at the time the iteration node executes, including sys.*,
+                           conv.*, and outputs from upstream nodes. Used as the source
+                           for deep-copying into each task's independent child pool.
+            cycle_nodes:  List of node config dicts belonging to this iteration's
+                          subgraph (i.e. nodes whose cycle field equals node_id).
+                          Passed to GraphBuilder when constructing each task's subgraph.
+            cycle_edges:  List of edge config dicts connecting nodes within the subgraph.
+                          Passed to GraphBuilder alongside cycle_nodes.
         """
-        self.start_id = start_id
         self.stream = stream
-        self.graph = graph
         self.state = state
         self.node_id = node_id
         self.typed_config = IterationNodeConfig(**config)
         self.looping = True
         self.variable_pool = variable_pool
-        self.child_variable_pool = child_variable_pool
+        self.cycle_nodes = cycle_nodes
+        self.cycle_edges = cycle_edges
         self.event_write = get_stream_writer()
-        self.checkpoint = RunnableConfig(
-            configurable={
-                "thread_id": uuid.uuid4()
-            }
-        )
 
         self.output_value = None
         self.result: list = []
 
-    async def _init_iteration_state(self, item, idx):
+    def _build_child_graph(self) -> tuple[CompiledStateGraph, VariablePool, str]:
         """
-        Initialize a per-iteration copy of the workflow state.
+        Build an independent compiled subgraph for a single iteration task.
 
-        Args:
-            item: Current element from the input array for this iteration.
-            idx: Index of the element in the input array.
+        Each call creates a brand-new VariablePool by deep-copying the parent pool,
+        then passes it to GraphBuilder. GraphBuilder binds this pool to every node's
+        execution closure at build time, so the pool and the subgraph always reference
+        the same object. This is the key design invariant: item/index written into the
+        pool after build will be visible to all nodes inside the subgraph.
 
         Returns:
-            A copy of the workflow state with iteration-specific variables set.
+            graph:      The compiled LangGraph subgraph ready for invocation.
+            child_pool: The VariablePool bound to this subgraph's node closures.
+                        Callers must write item/index into this pool before invoking
+                        the graph, and read output from it after invocation.
+            start_node_id: The ID of the CYCLE_START node inside the subgraph,
+                           used to set the initial activation signal in workflow state.
         """
-        loopstate = WorkflowState(
-            **self.state
+        from app.core.workflow.engine.graph_builder import GraphBuilder
+        child_pool = VariablePool()
+        child_pool.copy(self.variable_pool)
+        builder = GraphBuilder(
+            {"nodes": self.cycle_nodes, "edges": self.cycle_edges},
+            stream=self.stream,
+            variable_pool=child_pool,
+            cycle=self.node_id,
         )
-        self.child_variable_pool.copy(self.variable_pool)
-        await self.child_variable_pool.new(self.node_id, "item", item, VariableType.type_map(item), mut=True)
-        await self.child_variable_pool.new(self.node_id, "index", item, VariableType.type_map(item), mut=True)
-        loopstate["node_outputs"][self.node_id] = {
-            "item": item,
-            "index": idx,
-        }
+        graph = builder.build()
+        return graph, builder.variable_pool, builder.start_node_id
+
+    async def _init_iteration_state(self, item, idx, child_pool: VariablePool, start_id: str):
+        """
+        Initialize the workflow state for a single iteration.
+
+        Writes the current item and its index into child_pool under the iteration
+        node's namespace (e.g. iteration_xxx.item, iteration_xxx.index), making them
+        accessible to downstream nodes inside the subgraph via variable selectors.
+
+        Also prepares a copy of the parent workflow state with:
+        - node_outputs[node_id] set to {item, index} so the state snapshot is consistent
+          with the pool values.
+        - looping flag set to 1 (active) to signal the subgraph is inside a cycle.
+        - activate[start_id] set to True to trigger the CYCLE_START node.
+
+        Args:
+            item:       The current element from the input array.
+            idx:        The zero-based index of this element in the input array.
+            child_pool: The VariablePool bound to this iteration's subgraph.
+                        Must be the same object returned by _build_child_graph.
+            start_id:   The ID of the CYCLE_START node inside the subgraph.
+
+        Returns:
+            A WorkflowState instance ready to be passed to graph.ainvoke or graph.astream.
+        """
+        loopstate = WorkflowState(**self.state)
+        await child_pool.new(self.node_id, "item", item, VariableType.type_map(item), mut=True)
+        await child_pool.new(self.node_id, "index", idx, VariableType.type_map(idx), mut=True)
+        loopstate["node_outputs"][self.node_id] = {"item": item, "index": idx}
         loopstate["looping"] = 1
-        loopstate["activate"][self.start_id] = True
+        loopstate["activate"][start_id] = True
         return loopstate
 
-    def merge_conv_vars(self):
-        self.variable_pool.variables["conv"].update(
-            self.child_variable_pool.variables["conv"]
-        )
+    def _merge_conv_vars(self, child_pool: VariablePool):
+        self.variable_pool.variables["conv"].update(child_pool.variables["conv"])
 
     async def run_task(self, item, idx):
         """
         Execute a single iteration asynchronously.
+        Each task builds its own subgraph so the variable pool closure is independent.
 
-        Args:
-            item: The input element for this iteration.
-            idx: The index of this iteration.
+        Returns:
+            Tuple of (idx, output, result, child_pool, stopped)
         """
+        graph, child_pool, start_id = self._build_child_graph()
+        checkpoint = RunnableConfig(configurable={"thread_id": uuid.uuid4()})
+        init_state = await self._init_iteration_state(item, idx, child_pool, start_id)
+
         if self.stream:
-            async for event in self.graph.astream(
-                    await self._init_iteration_state(item, idx),
+            async for event in graph.astream(
+                    init_state,
                     stream_mode=["debug"],
-                    config=self.checkpoint
+                    config=checkpoint
             ):
                 if isinstance(event, tuple) and len(event) == 2:
                     mode, data = event
@@ -117,7 +166,6 @@ class IterationRuntime:
                     event_type = data.get("type")
                     payload = data.get("payload", {})
                     node_name = payload.get("name")
-
                     if node_name and node_name.startswith("nop"):
                         continue
                     if event_type == "task_result":
@@ -140,17 +188,13 @@ class IterationRuntime:
                                 "token_usage": result.get("node_outputs", {}).get(node_name, {}).get("token_usage")
                             }
                         })
-            result = self.graph.get_state(config=self.checkpoint).values
+            result = graph.get_state(config=checkpoint).values
         else:
-            result = await self.graph.ainvoke(await self._init_iteration_state(item, idx))
-        output = self.child_variable_pool.get_value(self.output_value)
-        if isinstance(output, list) and self.typed_config.flatten:
-            self.result.extend(output)
-        else:
-            self.result.append(output)
-        if result["looping"] == 2:
-            self.looping = False
-        return result
+            result = await graph.ainvoke(init_state)
+
+        output = child_pool.get_value(self.output_value)
+        stopped = result["looping"] == 2
+        return idx, output, result, child_pool, stopped
 
     def _create_iteration_tasks(self, array_obj, idx):
         """
@@ -196,16 +240,32 @@ class IterationRuntime:
                 tasks = self._create_iteration_tasks(array_obj, idx)
                 logger.info(f"Iteration node {self.node_id}: running, concurrency {len(tasks)}")
                 idx += self.typed_config.parallel_count
-                child_state.extend(await asyncio.gather(*tasks))
-                self.merge_conv_vars()
+                batch = await asyncio.gather(*tasks)
+                # Sort by idx to preserve order, then collect results
+                batch_sorted = sorted(batch, key=lambda x: x[0])
+                for _, output, result, child_pool, stopped in batch_sorted:
+                    if isinstance(output, list) and self.typed_config.flatten:
+                        self.result.extend(output)
+                    else:
+                        self.result.append(output)
+                    child_state.append(result)
+                    self._merge_conv_vars(child_pool)
+                    if stopped:
+                        self.looping = False
         else:
             # Execute iterations sequentially
             while idx < len(array_obj) and self.looping:
                 logger.info(f"Iteration node {self.node_id}: running")
                 item = array_obj[idx]
-                result = await self.run_task(item, idx)
-                self.merge_conv_vars()
+                _, output, result, child_pool, stopped = await self.run_task(item, idx)
+                if isinstance(output, list) and self.typed_config.flatten:
+                    self.result.extend(output)
+                else:
+                    self.result.append(output)
+                self._merge_conv_vars(child_pool)
                 child_state.append(result)
+                if stopped:
+                    self.looping = False
                 idx += 1
         logger.info(f"Iteration node {self.node_id}: execution completed")
         return {

api/app/core/workflow/nodes/cycle_graph/node.py

@@ -123,7 +123,7 @@ class CycleGraphNode(BaseNode):
 
         return cycle_nodes, cycle_edges
 
-    def build_graph(self):
+    def build_graph(self, variable_pool: VariablePool):
         """
         Build and compile the internal subgraph for this cycle node.
 
@@ -135,6 +135,7 @@ class CycleGraphNode(BaseNode):
         from app.core.workflow.engine.graph_builder import GraphBuilder
 
         self.child_variable_pool = VariablePool()
+        self.child_variable_pool.copy(variable_pool)
         builder = GraphBuilder(
             {
                 "nodes": self.cycle_nodes,
@@ -165,8 +166,8 @@ class CycleGraphNode(BaseNode):
         Raises:
             RuntimeError: If the node type is unsupported.
         """
-        self.build_graph()
         if self.node_type == NodeType.LOOP:
+            self.build_graph(variable_pool)
             return await LoopRuntime(
                 start_id=self.start_node_id,
                 stream=False,
@@ -179,20 +180,19 @@ class CycleGraphNode(BaseNode):
             ).run()
         if self.node_type == NodeType.ITERATION:
             return await IterationRuntime(
-                start_id=self.start_node_id,
                 stream=False,
-                graph=self.graph,
                 node_id=self.node_id,
                 config=self.config,
                 state=state,
                 variable_pool=variable_pool,
-                child_variable_pool=self.child_variable_pool
+                cycle_nodes=self.cycle_nodes,
+                cycle_edges=self.cycle_edges,
             ).run()
         raise RuntimeError("Unknown cycle node type")
 
     async def execute_stream(self, state: WorkflowState, variable_pool: VariablePool):
-        self.build_graph()
         if self.node_type == NodeType.LOOP:
+            self.build_graph(variable_pool)
             yield {
                 "__final__": True,
                 "result": await LoopRuntime(
@@ -211,14 +211,13 @@ class CycleGraphNode(BaseNode):
             yield {
                 "__final__": True,
                 "result": await IterationRuntime(
-                    start_id=self.start_node_id,
                     stream=True,
-                    graph=self.graph,
                     node_id=self.node_id,
                     config=self.config,
                     state=state,
                     variable_pool=variable_pool,
-                    child_variable_pool=self.child_variable_pool
+                    cycle_nodes=self.cycle_nodes,
+                    cycle_edges=self.cycle_edges,
                 ).run()
             }
             return

api/app/core/workflow/nodes/http_request/config.py

@@ -72,8 +72,9 @@ class HttpContentTypeConfig(BaseModel):
     @classmethod
     def validate_data(cls, v, info):
         content_type = info.data.get("content_type")
-        if content_type == HttpContentType.FROM_DATA and not isinstance(v, HttpFormData):
-            raise ValueError("When content_type is 'form-data', data must be of type HttpFormData")
+        if content_type == HttpContentType.FROM_DATA and (
+                not isinstance(v, list) or not all(isinstance(item, HttpFormData) for item in v)):
+            raise ValueError("When content_type is 'form-data', data must be a list of HttpFormData")
         elif content_type in [HttpContentType.JSON] and not isinstance(v, str):
             raise ValueError("When content_type is JSON, data must be of type str")
         elif content_type in [HttpContentType.WWW_FORM] and not isinstance(v, dict):

api/app/core/workflow/nodes/http_request/node.py

@@ -260,17 +260,22 @@ class HttpRequestNode(BaseNode):
                 ))
             case HttpContentType.FROM_DATA:
                 data = {}
-                content["files"] = {}
+                files = []
                 for item in self.typed_config.body.data:
+                    key = self._render_template(item.key, variable_pool)
                     if item.type == "text":
-                        data[self._render_template(item.key, variable_pool)] = self._render_template(item.value,
-                                                                                                     variable_pool)
+                        data[key] = self._render_template(item.value, variable_pool)
                     elif item.type == "file":
-                        content["files"][self._render_template(item.key, variable_pool)] = (
-                            uuid.uuid4().hex,
-                            await variable_pool.get_instance(item.value).get_content()
-                        )
+                        file_instance = variable_pool.get_instance(item.value)
+                        if isinstance(file_instance, ArrayVariable):
+                            for v in file_instance.value:
+                                if isinstance(v, FileVariable):
+                                    files.append((key, (uuid.uuid4().hex, await v.get_content())))
+                        elif isinstance(file_instance, FileVariable):
+                            files.append((key, (uuid.uuid4().hex, await file_instance.get_content())))
                 content["data"] = data
+                if files:
+                    content["files"] = files
             case HttpContentType.BINARY:
                 content["files"] = []
                 file_instence = variable_pool.get_instance(self.typed_config.body.data)

api/app/core/workflow/nodes/if_else/config.py

@@ -6,6 +6,30 @@ from app.core.workflow.nodes.base_config import BaseNodeConfig
 from app.core.workflow.nodes.enums import ComparisonOperator, LogicOperator, ValueInputType
 
 
+class SubVariableConditionItem(BaseModel):
+    """A single condition on a file object's field, used inside sub_variable_condition."""
+    key: str = Field(..., description="Field name of the file object, e.g. type, size, name")
+    operator: ComparisonOperator = Field(..., description="Comparison operator")
+    value: Any = Field(default=None, description="Value to compare with, or variable selector when input_type=variable")
+    input_type: ValueInputType = Field(default=ValueInputType.CONSTANT, description="constant or variable")
+
+    @field_validator("input_type", mode="before")
+    @classmethod
+    def lower_input_type(cls, v):
+        if isinstance(v, str):
+            try:
+                return ValueInputType(v.lower())
+            except ValueError:
+                raise ValueError(f"Invalid input_type: {v}")
+        return v
+
+
+class SubVariableCondition(BaseModel):
+    """Sub-conditions applied to each file element in an array[file] variable."""
+    logical_operator: LogicOperator = Field(default=LogicOperator.AND)
+    conditions: list[SubVariableConditionItem] = Field(default_factory=list)
+
+
 class ConditionDetail(BaseModel):
     operator: ComparisonOperator = Field(
         ...,
@@ -14,12 +38,12 @@ class ConditionDetail(BaseModel):
 
     left: str = Field(
         ...,
-        description="Value to compare against"
+        description="Variable selector, e.g. {{sys.files}}"
     )
 
     right: Any = Field(
         default=None,
-        description="Value to compare with"
+        description="Value to compare with (unused when sub_variable_condition is set)"
     )
 
     input_type: ValueInputType = Field(
@@ -27,6 +51,11 @@ class ConditionDetail(BaseModel):
         description="Value input type for comparison"
     )
 
+    sub_variable_condition: SubVariableCondition | None = Field(
+        default=None,
+        description="Sub-conditions for array[file] fields. When set, operator must be contains/not_contains."
+    )
+
     @field_validator("input_type", mode="before")
     @classmethod
     def lower_input_type(cls, v):
@@ -39,16 +68,19 @@ class ConditionDetail(BaseModel):
 
 
 class ConditionBranchConfig(BaseModel):
-    """Configuration for a conditional branch"""
+    """Configuration for a conditional branch.
+
+    logical_operator controls how all expressions are combined (AND/OR).
+    """
 
     logical_operator: LogicOperator = Field(
         default=LogicOperator.AND,
-        description="Logical operator used to combine multiple condition expressions"
+        description="Logical operator used to combine all conditions"
     )
 
     expressions: list[ConditionDetail] = Field(
-        ...,
-        description="List of condition expressions within this branch"
+        default_factory=list,
+        description="List of conditions within this branch"
     )
 
 

api/app/core/workflow/nodes/if_else/node.py

@@ -7,7 +7,7 @@ from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.base_node import BaseNode
 from app.core.workflow.nodes.enums import ComparisonOperator, LogicOperator, ValueInputType
 from app.core.workflow.nodes.if_else import IfElseNodeConfig
-from app.core.workflow.nodes.operators import ConditionExpressionResolver, CompareOperatorInstance
+from app.core.workflow.nodes.operators import ConditionExpressionResolver, CompareOperatorInstance, ArrayFileContainsOperator
 from app.core.workflow.variable.base_variable import VariableType
 
 logger = logging.getLogger(__name__)
@@ -90,11 +90,9 @@ class IfElseNode(BaseNode):
             list[str]: A list of Python boolean expression strings,
             ordered by branch priority.
         """
-        branch_index = 0
         conditions = []
 
         for case_branch in self.typed_config.cases:
-            branch_index += 1
             branch_result = []
             for expression in case_branch.expressions:
                 pattern = r"\{\{\s*(.*?)\s*\}\}"
@@ -103,13 +101,18 @@ class IfElseNode(BaseNode):
                     left_value = self.get_variable(left_string, variable_pool)
                 except KeyError:
                     left_value = None
-                evaluator = ConditionExpressionResolver.resolve_by_value(left_value)(
-                    variable_pool,
-                    expression.left,
-                    expression.right,
-                    expression.input_type
-                )
+
+                if expression.sub_variable_condition is not None and isinstance(left_value, list):
+                    evaluator = ArrayFileContainsOperator(left_value, expression.sub_variable_condition, variable_pool)
+                else:
+                    evaluator = ConditionExpressionResolver.resolve_by_value(left_value)(
+                        variable_pool,
+                        expression.left,
+                        expression.right,
+                        expression.input_type
+                    )
                 branch_result.append(self._evaluate(expression.operator, evaluator))
+
             if case_branch.logical_operator == LogicOperator.AND:
                 conditions.append(all(branch_result))
             else:

api/app/core/workflow/nodes/llm/config.py

@@ -116,6 +116,11 @@ class LLMNodeConfig(BaseNodeConfig):
         description="Top-p 采样参数"
     )
 
+    json_output: bool = Field(
+        default=False,
+        description="是否以 JSON 格式输出"
+    )
+
     frequency_penalty: float | None = Field(
         default=None,
         ge=-2.0,

api/app/core/workflow/nodes/llm/node.py

@@ -22,6 +22,7 @@ from app.db import get_db_context
 from app.models import ModelType
 from app.schemas.model_schema import ModelInfo
 from app.services.model_service import ModelConfigService
+from app.models.models_model import ModelProvider
 
 logger = logging.getLogger(__name__)
 
@@ -126,7 +127,11 @@ class LLMNode(BaseNode):
 
         # 4. 创建 LLM 实例（使用已提取的数据）
         # 注意：对于流式输出，需要在模型初始化时设置 streaming=True
-        extra_params = {"streaming": stream} if stream else {}
+        extra_params: dict[str, Any] = {"streaming": stream} if stream else {}
+        if self.typed_config.temperature is not None:
+            extra_params["temperature"] = self.typed_config.temperature
+        if self.typed_config.max_tokens is not None:
+            extra_params["max_tokens"] = self.typed_config.max_tokens
 
         llm = RedBearLLM(
             RedBearModelConfig(
@@ -135,7 +140,9 @@ class LLMNode(BaseNode):
                 api_key=model_info.api_key,
                 base_url=model_info.api_base,
                 extra_params=extra_params,
-                is_omni=model_info.is_omni
+                is_omni=model_info.is_omni,
+                capability=model_info.capability,
+                json_output=self.typed_config.json_output,
             ),
             type=model_info.model_type
         )
@@ -218,6 +225,19 @@ class LLMNode(BaseNode):
             rendered = self._render_template(prompt_template, variable_pool)
             self.messages = [{"role": "user", "content": rendered}]
 
+        # ChatTongyi 要求 messages 含 'json' 字样才能使用 response_format，在 system prompt 中注入
+        # VOLCANO 模型不支持 response_format，同样需要 system prompt 注入
+        need_json_prompt = self.typed_config.json_output and (
+            (model_info.provider.lower() == ModelProvider.DASHSCOPE and not model_info.is_omni)
+            or model_info.provider.lower() == ModelProvider.VOLCANO
+        )
+        if need_json_prompt:
+            system_msg = next((m for m in self.messages if m["role"] == "system"), None)
+            if system_msg:
+                system_msg["content"] += "\n请以JSON格式输出。"
+            else:
+                self.messages.insert(0, {"role": "system", "content": "请以JSON格式输出。"})
+
         return llm
 
     async def execute(self, state: WorkflowState, variable_pool: VariablePool) -> AIMessage:

api/app/core/workflow/nodes/operators.py

@@ -395,11 +395,73 @@ class NoneObjectComparisonOperator:
         return lambda *args, **kwargs: False
 
 
+class ArrayFileContainsOperator:
+    """Handles contains/not_contains on array[file] with sub_variable_condition."""
+
+    def __init__(self, left_value: list[dict], sub_variable_condition: Any, pool: VariablePool | None = None):
+        self.left_value = left_value
+        self.sub_variable_condition = sub_variable_condition
+        self.pool = pool
+
+    def _resolve_value(self, cond: Any) -> Any:
+        if cond.input_type == ValueInputType.VARIABLE and self.pool is not None:
+            pattern = r"\{\{\s*(.*?)\s*\}\}"
+            selector = re.sub(pattern, r"\1", str(cond.value)).strip()
+            return self.pool.get_value(selector, default=None, strict=False)
+        return cond.value
+
+    def _match_item(self, file_item: dict) -> bool:
+        results = []
+        for cond in self.sub_variable_condition.conditions:
+            field_val = file_item.get(cond.key)
+            expected = self._resolve_value(cond)
+            result = self._eval_sub(field_val, cond.operator.value, expected)
+            results.append(result)
+        if self.sub_variable_condition.logical_operator.value == "and":
+            return all(results)
+        return any(results)
+
+    @staticmethod
+    def _eval_sub(field_val: Any, op: str, expected: Any) -> bool:
+        if field_val is None:
+            return op == "empty"
+        match op:
+            case "eq":           return str(field_val) == str(expected)
+            case "ne":           return str(field_val) != str(expected)
+            case "contains":     return isinstance(field_val, str) and str(expected) in field_val
+            case "not_contains": return isinstance(field_val, str) and str(expected) not in field_val
+            case "in":           return field_val in (expected if isinstance(expected, list) else [expected])
+            case "not_in":       return field_val not in (expected if isinstance(expected, list) else [expected])
+            case "gt":           return isinstance(field_val, (int, float)) and field_val > float(expected)
+            case "ge":           return isinstance(field_val, (int, float)) and field_val >= float(expected)
+            case "lt":           return isinstance(field_val, (int, float)) and field_val < float(expected)
+            case "le":           return isinstance(field_val, (int, float)) and field_val <= float(expected)
+            case "empty":        return field_val in (None, "", 0)
+            case "not_empty":    return field_val not in (None, "", 0)
+            case _:              return False
+
+    def contains(self) -> bool:
+        return any(self._match_item(f) for f in self.left_value if isinstance(f, dict))
+
+    def not_contains(self) -> bool:
+        return not self.contains()
+
+    def empty(self) -> bool:
+        return not self.left_value
+
+    def not_empty(self) -> bool:
+        return bool(self.left_value)
+
+    def __getattr__(self, name):
+        return lambda *args, **kwargs: False
+
+
 CompareOperatorInstance = Union[
     StringComparisonOperator,
     NumberComparisonOperator,
     BooleanComparisonOperator,
     ArrayComparisonOperator,
+    ArrayFileContainsOperator,
     ObjectComparisonOperator
 ]
 CompareOperatorType = Type[CompareOperatorInstance]

api/app/core/workflow/nodes/tool/node.py

@@ -15,6 +15,7 @@ from app.services.tool_service import ToolService
 logger = logging.getLogger(__name__)
 
 TEMPLATE_PATTERN = re.compile(r"\{\{.*?}}")
+PURE_VARIABLE_PATTERN = re.compile(r"^\{\{\s*([\w.]+)\s*}}$")
 
 
 class ToolNode(BaseNode):
@@ -52,13 +53,21 @@ class ToolNode(BaseNode):
         # 渲染工具参数
         rendered_parameters = {}
         for param_name, param_template in self.typed_config.tool_parameters.items():
-            if isinstance(param_template, str) and TEMPLATE_PATTERN.search(param_template):
-                try:
-                    rendered_value = self._render_template(param_template, variable_pool)
-                except Exception as e:
-                    raise ValueError(f"模板渲染失败：参数 {param_name} 的模板 {param_template} 解析错误") from e
+            if isinstance(param_template, str):
+                pure_match = PURE_VARIABLE_PATTERN.match(param_template)
+                if pure_match:
+                    # 纯单变量引用直接取原始值，保留 int/bool/float 等类型
+                    rendered_value = self.get_variable(pure_match.group(1), variable_pool, strict=False)
+                    if rendered_value is None:
+                        rendered_value = self._render_template(param_template, variable_pool)
+                elif TEMPLATE_PATTERN.search(param_template):
+                    try:
+                        rendered_value = self._render_template(param_template, variable_pool)
+                    except Exception as e:
+                        raise ValueError(f"模板渲染失败：参数 {param_name} 的模板 {param_template} 解析错误") from e
+                else:
+                    rendered_value = param_template
             else:
-                # 非模板参数（数字/布尔/普通字符串）直接保留原值
                 rendered_value = param_template
             rendered_parameters[param_name] = rendered_value
 

api/app/core/workflow/variable/variable_objects.py

@@ -84,7 +84,7 @@ class FileVariable(BaseVariable):
         total_bytes = 0
         chunks = []
 
-        async with httpx.AsyncClient() as client:
+        async with httpx.AsyncClient(follow_redirects=True) as client:
             async with client.stream("GET", self.value.url) as resp:
                 resp.raise_for_status()
                 async for chunk in resp.aiter_bytes(8192):