Add V2 external service providers

ai_knowledge_base_v2/apps/backend/app/core/config.py

@@ -33,6 +33,11 @@ class Settings(BaseSettings):
     mock_sms_code: str = "123456"
     sms_code_expire_minutes: int = 5
     mock_rag_enabled: bool = True
+    mock_model_enabled: bool = True
+    feishu_mock_enabled: bool = True
+    feishu_search_url: str = ""
+    feishu_timeout_seconds: int = 20
+    feishu_retry_count: int = 2
 
     default_daily_chat_limit: int = 100
     default_user_name_prefix: str = "用户"

ai_knowledge_base_v2/apps/backend/app/services/ai_request_log_service.py

@@ -37,3 +37,32 @@ class AiRequestLogService:
                 status="SUCCESS",
             )
         )
+
+    @staticmethod
+    def write_failed(
+        db: Session,
+        *,
+        session_id: int,
+        message_id: int | None,
+        user_id: int,
+        model_name: str | None,
+        prompt: str | None,
+        knowledge_ids: str | None,
+        retrieve_count: int,
+        cost_ms: int,
+        error_message: str,
+    ) -> None:
+        db.add(
+            AiRequestLog(
+                session_id=session_id,
+                message_id=message_id,
+                user_id=user_id,
+                model_name=model_name,
+                prompt=prompt,
+                knowledge_ids=knowledge_ids,
+                retrieve_count=retrieve_count,
+                cost_ms=cost_ms,
+                status="FAILED",
+                error_message=error_message,
+            )
+        )

ai_knowledge_base_v2/apps/backend/app/services/chat_service.py

@@ -10,6 +10,7 @@ from sqlalchemy.orm import Session
 from app.models.chat import ChatMessage, ChatSession
 from app.models.user import User
 from app.services.ai_request_log_service import AiRequestLogService
+from app.services.external_errors import ExternalServiceError
 from app.services.model_service import ModelClientService
 from app.services.rag_service import RagService
 
@@ -86,8 +87,28 @@ class ChatService:
         db.flush()
 
         started_at = perf_counter()
-        rag_result = RagService.build_result(db, user, normalized_question)
-        completion = ModelClientService.complete(db, rag_result)
+        try:
+            rag_result = RagService.build_result(db, user, normalized_question)
+            completion = ModelClientService.complete(db, rag_result)
+        except ExternalServiceError as exc:
+            cost_ms = int((perf_counter() - started_at) * 1000)
+            AiRequestLogService.write_failed(
+                db,
+                session_id=session.id,
+                message_id=user_message.id,
+                user_id=user.id,
+                model_name=None,
+                prompt=normalized_question,
+                knowledge_ids=None,
+                retrieve_count=0,
+                cost_ms=cost_ms,
+                error_message=str(exc),
+            )
+            db.commit()
+            raise HTTPException(
+                status_code=status.HTTP_502_BAD_GATEWAY,
+                detail="外部服务暂时不可用，请稍后再试",
+            ) from exc
         cost_ms = int((perf_counter() - started_at) * 1000)
 
         assistant_message = ChatMessage(

ai_knowledge_base_v2/apps/backend/app/services/external_errors.py

@@ -0,0 +1,7 @@
+from __future__ import annotations
+
+
+class ExternalServiceError(RuntimeError):
+    def __init__(self, message: str, *, provider: str) -> None:
+        super().__init__(message)
+        self.provider = provider

ai_knowledge_base_v2/apps/backend/app/services/feishu_service.py

@@ -0,0 +1,122 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+import httpx
+
+from app.core.config import get_settings
+from app.services.external_errors import ExternalServiceError
+from app.services.knowledge_service import KnowledgeScope
+
+if TYPE_CHECKING:
+    from app.services.rag_service import RetrievedChunk
+
+
+class FeishuKnowledgeService:
+    _mock_documents = [
+        {
+            "title": "一期产品目标",
+            "keywords": {"一期", "目标", "问答", "登录", "知识库", "飞书", "用户端", "后台"},
+            "content": "一期要交付企业飞书知识库 AI 问答系统，核心包括用户登录、AI 问答、权限内知识库回答和后台管理能力。",
+        },
+        {
+            "title": "权限过滤规则",
+            "keywords": {"权限", "授权", "越权", "用户", "知识库", "过期", "禁用"},
+            "content": "RAG 检索前必须先过滤用户授权知识库，只允许未禁用、未过期、当前用户有权限的知识库参与回答。",
+        },
+        {
+            "title": "无命中兜底规则",
+            "keywords": {"无命中", "兜底", "编造", "检索", "答案", "命中"},
+            "content": "当知识库没有命中相关内容时，系统必须固定返回无命中兜底文案。",
+        },
+        {
+            "title": "技术实现边界",
+            "keywords": {"模型", "prompt", "大模型", "日志", "sse", "流式", "追溯"},
+            "content": "后端需要组装系统 Prompt、检索片段、历史上下文和当前问题，通过 SSE 流式输出，并记录 AI 请求日志。",
+        },
+    ]
+
+    @classmethod
+    def retrieve(cls, question: str, scopes: list[KnowledgeScope]) -> list["RetrievedChunk"]:
+        if not scopes:
+            return []
+
+        settings = get_settings()
+        if settings.feishu_mock_enabled:
+            return cls._retrieve_mock(question, scopes)
+        return cls._retrieve_remote(question, scopes)
+
+    @classmethod
+    def _retrieve_mock(cls, question: str, scopes: list[KnowledgeScope]) -> list["RetrievedChunk"]:
+        from app.services.rag_service import RetrievedChunk
+
+        normalized_question = question.lower()
+        matched_documents = []
+        for document in cls._mock_documents:
+            if any(keyword.lower() in normalized_question for keyword in document["keywords"]):
+                matched_documents.append(document)
+
+        if not matched_documents:
+            return []
+
+        primary_scope = scopes[0]
+        return [
+            RetrievedChunk(
+                knowledge_id=primary_scope.id,
+                knowledge_name=primary_scope.name,
+                title=document["title"],
+                content=document["content"],
+                source_url=None,
+            )
+            for document in matched_documents[:3]
+        ]
+
+    @classmethod
+    def _retrieve_remote(cls, question: str, scopes: list[KnowledgeScope]) -> list["RetrievedChunk"]:
+        from app.services.rag_service import RetrievedChunk
+
+        settings = get_settings()
+        if not settings.feishu_search_url:
+            raise ExternalServiceError("飞书检索地址未配置", provider="feishu")
+
+        payload = {
+            "query": question,
+            "knowledgeScopes": [
+                {
+                    "id": scope.id,
+                    "spaceId": scope.feishu_space_id,
+                    "nodeId": scope.feishu_node_id,
+                    "name": scope.name,
+                }
+                for scope in scopes
+            ],
+        }
+        data = cls._post_with_retry(settings.feishu_search_url, payload)
+        chunks = data.get("chunks", [])
+        return [
+            RetrievedChunk(
+                knowledge_id=int(item.get("knowledgeId") or 0),
+                knowledge_name=str(item.get("knowledgeName") or "飞书知识库"),
+                title=str(item.get("title") or "未命名片段"),
+                content=str(item.get("content") or ""),
+                source_url=item.get("sourceUrl"),
+            )
+            for item in chunks
+            if item.get("content")
+        ]
+
+    @staticmethod
+    def _post_with_retry(url: str, payload: dict[str, Any]) -> dict[str, Any]:
+        settings = get_settings()
+        last_error: Exception | None = None
+        for _ in range(settings.feishu_retry_count + 1):
+            try:
+                response = httpx.post(url, json=payload, timeout=settings.feishu_timeout_seconds)
+                response.raise_for_status()
+                data = response.json()
+                if not isinstance(data, dict):
+                    raise ExternalServiceError("飞书检索响应格式不正确", provider="feishu")
+                return data
+            except (httpx.HTTPError, ValueError, ExternalServiceError) as exc:
+                last_error = exc
+        raise ExternalServiceError(f"飞书检索失败：{last_error}", provider="feishu")

ai_knowledge_base_v2/apps/backend/app/services/model_service.py

@@ -1,11 +1,15 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
+from typing import Any
 
+import httpx
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
+from app.core.config import get_settings
 from app.models.ai_config import ModelConfig
+from app.services.external_errors import ExternalServiceError
 from app.services.rag_service import NO_HIT_ANSWER, RagResult
 
 
@@ -27,8 +31,14 @@ class ModelClientService:
             .order_by(ModelConfig.id.desc())
             .limit(1)
         )
-        model_name = model.model_name if model is not None else "mock-model"
-        answer = _mock_answer(rag_result)
+        settings = get_settings()
+        if settings.mock_model_enabled or model is None:
+            model_name = model.model_name if model is not None else "mock-model"
+            answer = _mock_answer(rag_result)
+        else:
+            model_name = model.model_name
+            answer = _call_openai_compatible_model(model, rag_result)
+
         return ModelCompletion(
             answer=answer,
             model_id=model.id if model is not None else None,
@@ -55,3 +65,55 @@ def _mock_answer(rag_result: RagResult) -> str:
 
 def _rough_token_count(text: str) -> int:
     return max(1, len(text.strip()) // 2)
+
+
+def _call_openai_compatible_model(model: ModelConfig, rag_result: RagResult) -> str:
+    if not rag_result.is_hit:
+        return NO_HIT_ANSWER
+    if not model.api_url or not model.api_key:
+        raise ExternalServiceError("模型 API URL 或 API Key 未配置", provider="model")
+
+    payload = {
+        "model": model.model_name,
+        "messages": [
+            {"role": "system", "content": "你是企业知识库问答助手，只能基于已提供的知识片段回答。"},
+            {"role": "user", "content": rag_result.prompt},
+        ],
+        "temperature": float(model.temperature) if model.temperature is not None else 0.2,
+        "max_tokens": model.max_token or 1024,
+        "stream": False,
+    }
+    headers = {
+        "Authorization": f"Bearer {model.api_key}",
+        "Content-Type": "application/json",
+    }
+    try:
+        response = httpx.post(
+            model.api_url,
+            json=payload,
+            headers=headers,
+            timeout=model.timeout_second,
+        )
+        response.raise_for_status()
+        return _extract_answer(response.json())
+    except (httpx.HTTPError, ValueError, KeyError, TypeError) as exc:
+        raise ExternalServiceError(f"模型调用失败：{exc}", provider="model") from exc
+
+
+def _extract_answer(data: dict[str, Any]) -> str:
+    choices = data.get("choices")
+    if not isinstance(choices, list) or not choices:
+        raise ValueError("模型响应缺少 choices")
+
+    first_choice = choices[0]
+    if not isinstance(first_choice, dict):
+        raise ValueError("模型响应 choices 格式不正确")
+
+    message = first_choice.get("message")
+    if isinstance(message, dict) and message.get("content"):
+        return str(message["content"])
+
+    if first_choice.get("text"):
+        return str(first_choice["text"])
+
+    raise ValueError("模型响应缺少回答内容")

ai_knowledge_base_v2/apps/backend/app/services/rag_service.py

@@ -7,6 +7,7 @@ from sqlalchemy.orm import Session
 
 from app.models.ai_config import Prompt
 from app.models.user import User
+from app.services.feishu_service import FeishuKnowledgeService
 from app.services.knowledge_service import KnowledgeAccessService, KnowledgeScope
 
 NO_HIT_ANSWER = "当前知识库中未检索到相关内容，请联系管理员补充相关知识。"
@@ -46,57 +47,6 @@ class RagService:
         return RagResult(question=question, knowledge_scopes=scopes, chunks=chunks, prompt=prompt)
 
 
-class FeishuKnowledgeService:
-    _mock_documents = [
-        {
-            "title": "一期产品目标",
-            "keywords": {"一期", "目标", "问答", "登录", "知识库", "飞书", "用户端", "后台"},
-            "content": "一期要交付企业飞书知识库 AI 问答系统，核心包括用户登录、AI 问答、权限内知识库回答和后台管理能力。",
-        },
-        {
-            "title": "权限过滤规则",
-            "keywords": {"权限", "授权", "越权", "用户", "知识库", "过期", "禁用"},
-            "content": "RAG 检索前必须先过滤用户授权知识库，只允许未禁用、未过期、当前用户有权限的知识库参与回答。",
-        },
-        {
-            "title": "无命中兜底规则",
-            "keywords": {"无命中", "兜底", "编造", "检索", "答案", "命中"},
-            "content": f"当知识库没有命中相关内容时，系统必须固定返回：{NO_HIT_ANSWER}",
-        },
-        {
-            "title": "技术实现边界",
-            "keywords": {"模型", "prompt", "大模型", "日志", "sse", "流式", "追溯"},
-            "content": "后端需要组装系统 Prompt、检索片段、历史上下文和当前问题，通过 SSE 流式输出，并记录 AI 请求日志。",
-        },
-    ]
-
-    @classmethod
-    def retrieve(cls, question: str, scopes: list[KnowledgeScope]) -> list[RetrievedChunk]:
-        if not scopes:
-            return []
-
-        normalized_question = question.lower()
-        matched_documents = []
-        for document in cls._mock_documents:
-            if any(keyword.lower() in normalized_question for keyword in document["keywords"]):
-                matched_documents.append(document)
-
-        if not matched_documents:
-            return []
-
-        primary_scope = scopes[0]
-        return [
-            RetrievedChunk(
-                knowledge_id=primary_scope.id,
-                knowledge_name=primary_scope.name,
-                title=document["title"],
-                content=document["content"],
-                source_url=None,
-            )
-            for document in matched_documents[:3]
-        ]
-
-
 class PromptService:
     _default_prompt = (
         "你是企业飞书知识库 AI 助手。你只能基于提供的知识片段回答，不能编造。"