feat:修复投顾agent功能

Reviewed-on: #29

agent/advisor_agent/data_query.py

@@ -13,6 +13,7 @@ from config.settings import settings
 from nl2sql.contracts import DataQueryRequest, DataQueryResult
 from nl2sql.embedding import EmbeddingError
 from nl2sql.retrieval import retrieve_metadata
+from nl2sql.cache import cache_get, cache_set, build_question_cache_key
 from nl2sql.runtime_config import runtime_config
 from nl2sql.schema import load_authoritative_schema
 from repositories.customer_relation import CustomerRelationRepo
@@ -55,7 +56,9 @@ def enrich_customer_rows(
 
 def _is_current_holdings_query(question: str) -> bool:
     text = "".join((question or "").split())
-    return any(term in text for term in ("当前持仓", "目前持仓", "现有持仓", "持仓明细", "持仓情况"))
+    if any(term in text for term in ("当前持仓", "目前持仓", "现有持仓", "持仓明细", "持仓情况")):
+        return True
+    return "持仓" in text and not any(term in text for term in ("历史", "曾经", "已卖出", "交易记录"))
 
 
 async def _query_current_holdings(
@@ -154,6 +157,7 @@ async def execute_advisor_data_query(
         pass
 
     if scope == "customer" and _is_current_holdings_query(question):
+        redis = redis or database.redis.client()
         return await _query_current_holdings(
             db,
             customer_id=customer_id,
@@ -166,6 +170,21 @@ async def execute_advisor_data_query(
 
     milvus = milvus or database.milvus.client()
     redis = redis or database.redis.client()
+    cache_key = build_question_cache_key(
+        question,
+        permission=permission,
+        data_scope={"customer_ids": customer_ids},
+        page=page,
+        page_size=page_size,
+        sort_by=sort_by,
+        sort_order=sort_order,
+    )
+    cached = await cache_get(redis, cache_key)
+    if cached is not None:
+        cached["trace_id"] = trace_id
+        cached["customer_id"] = customer_id
+        cached["warnings"] = [*cached.get("warnings", []), "cache_hit"]
+        return cached
     llm_client = llm_client or default_llm
     request = DataQueryRequest(
         question=question,
@@ -222,4 +241,5 @@ async def execute_advisor_data_query(
         payload["answer"] = f"{existing_answer.rstrip('。')}。客户姓名：{enriched['name_summary']}。"
     payload["sql"] = None
     payload["customer_id"] = customer_id
+    await cache_set(redis, cache_key, payload, ttl=runtime_config.cache_ttl)
     return payload

agent/advisor_agent/intent/classifier.py

@@ -35,20 +35,58 @@ class IntentClassification:
     reason: str = ""
 
 
-_CLASSIFIER_PROMPT = """你是基金投顾工作台的意图分类器，只负责分类，不回答用户问题。
-只能从以下分类中选择一个：
-- recommend：基金推荐、组合配置或投资方案
-- rebalance：组合偏离、调仓、再平衡、仓位调整
-- fund_analysis：单只基金分析、净值、收益、回撤、波动率、夏普比率
-- dialogue-script：给客户准备沟通话术、解释、安抚、投诉或风险提醒
-- data_query：查询客户持仓、资产、余额、收益、交易、账户明细
-- casual_chat：问候、闲聊、感谢、身份询问或无法归入业务分类的内容
+_CLASSIFIER_PROMPT = """# 任务：投顾助手意图识别
+你是投顾系统意图分类器，对客户输入文本做意图识别，**只能输出一个类别名称，禁止额外解释**。
+
+## 类别说明
+1. 基金推荐：客户希望推荐、筛选基金产品
+2. 调仓建议：客户询问是否买卖、加减仓、更换基金，寻求调仓交易建议
+3. 客户持仓基金分析：分析客户现有持仓组合、风险、收益情况，不涉及买卖操作建议
+4. 跟客户的沟通话术：投顾需要生成一段发给客户的话术文案。【注意：客户发起提问，不会是该类别】
+5. 数据查询：查询基金客观数据，如净值、基金经理、规模、持仓、费率等事实信息，不做推荐、诊断
+6. 普通聊天：日常问候、单纯情绪吐槽，无明确业务诉求
+
+## 判定优先级（同时存在多个诉求时，取优先级最高）
+调仓建议 > 基金推荐 > 客户持仓基金分析 > 数据查询 > 跟客户的沟通话术 > 普通聊天
+
+## 示例(xx表示我名下的客户名字)
+输入：帮我名下XX推荐几只适合养老的基金
+输出：基金推荐
+
+输入：XX手上的某某基金现在要不要卖出？
+输出：调仓建议
+
+输入：帮我看看xx的基金组合风险高不高
+输出：客户持仓基金分析
+
+输入：帮我查一下XX基金最新规模
+输出：数据查询
+
+输入：帮我写一段话安抚客户，解释近期回撤
+输出：跟客户的沟通话术
+
+输入：今天天气不错
+输出：普通聊天
+
+输入：xx持有的这几只基金波动很大，要不要减仓？
+输出：调仓建议
+
+输入：帮我看下xx持仓的基金，它们的基金经理是谁
+输出：数据查询
+
+现在开始分类
+输入：{{user_query}}
+输出：
 
 只输出 JSON，不要 Markdown，不要额外文字：
 {"intent":"分类值","confidence":0到1之间的数字,"reason":"不超过30字的原因"}
 """
 
 _RECOMMENDATION_TERMS = ("推荐", "组合建议", "配置建议", "买什么", "适合配置", "筛选基金", "投资方案")
+_REBALANCE_TERMS = ("调仓", "再平衡", "组合调整", "配置偏离", "偏离目标")
+_DIALOGUE_TERMS = ("话术", "沟通", "怎么跟客户说", "如何向客户解释", "安抚客户", "投诉处理")
+_CUSTOMER_DATA_TERMS = ("持仓", "资产", "余额", "交易", "账户", "份额", "市值", "盈亏", "资金")
+_REFERENCE_TERMS = ("他", "她", "它", "这个客户", "该客户", "那个客户", "刚才", "上一轮")
 _SCOPE_ERROR_MESSAGE = "投顾范围查询仅支持客户数据查询"
 
 
@@ -83,11 +121,31 @@ def _parse_model_result(raw: str) -> IntentClassification | None:
     )
 
 
+def resolve_advisor_intent(
+    query: str | None,
+    classified_intent: str,
+    *,
+    customer_resolved: bool,
+) -> str:
+    """结合实体解析结果做最终业务路由，处理一句话中的明确优先意图。"""
+    text = re.sub(r"\s+", "", query or "")
+    if any(term in text for term in _REBALANCE_TERMS):
+        return AGENT_INTENT_REBALANCE
+    if any(term in text for term in _RECOMMENDATION_TERMS):
+        return AGENT_INTENT_RECOMMEND
+    if any(term in text for term in _DIALOGUE_TERMS):
+        return AGENT_INTENT_DIALOGUE_SCRIPT
+    if customer_resolved and any(term in text for term in _CUSTOMER_DATA_TERMS):
+        return AGENT_INTENT_DATA_QUERY
+    return classified_intent
+
+
 async def classify_advisor_intent(
     query: str | None,
     llm_client=None,
     *,
     explicit_intent: str | None = None,
+    conversation_context: str = "",
     timeout: float = 2.0,
 ) -> IntentClassification:
     """分类用户意图；显式意图兼容旧客户端，模型失败时安全回退规则。"""
@@ -102,13 +160,39 @@ async def classify_advisor_intent(
             "rule",
             "识别为系统提示文本而非业务查询",
         )
+    fast_intent = recognize_advisor_intent(query)
+    if (
+        fast_intent == AGENT_INTENT_DATA_QUERY
+        and not any(term in query for term in _RECOMMENDATION_TERMS)
+    ):
+        return IntentClassification(
+            fast_intent,
+            0.95,
+            "rule_fast",
+            "明显查询类问题",
+        )
+    normalized_query = re.sub(r"\s+", "", query)
+    if (
+        any(term in normalized_query for term in _CUSTOMER_DATA_TERMS)
+        and conversation_context
+        and any(term in normalized_query for term in _REFERENCE_TERMS)
+    ):
+        return IntentClassification(
+            AGENT_INTENT_DATA_QUERY,
+            0.9,
+            "rule_context",
+            "上下文中的客户数据追问",
+        )
     if llm_client is not None:
         try:
+            prompt = query.strip()
+            if conversation_context:
+                prompt = f"会话上下文：\n{conversation_context[:2000]}\n\n当前问题：\n{prompt}"
             raw = await asyncio.wait_for(
                 llm_client.chat(
                     [
                         {"role": "system", "content": _CLASSIFIER_PROMPT},
-                        {"role": "user", "content": query.strip()},
+                        {"role": "user", "content": prompt},
                     ],
                     temperature=0,
                     max_tokens=120,
@@ -137,4 +221,4 @@ async def classify_advisor_intent(
     return _rule_fallback(query)
 
 
-__all__ = ["IntentClassification", "classify_advisor_intent"]
+__all__ = ["IntentClassification", "classify_advisor_intent", "resolve_advisor_intent"]

agent/advisor_agent/intent/recognizer.py

@@ -26,6 +26,7 @@ _QUERY_ACTIONS = (
 _DATA_TERMS = (
     "持仓",
     "资产",
+    "资金",
     "收益",
     "交易记录",
     "申购",
@@ -41,6 +42,15 @@ _FUND_ANALYSIS_TERMS = ("基金分析", "分析基金", "基金表现", "净值
 _DIALOGUE_TERMS = ("话术", "沟通", "怎么跟客户说", "如何向客户解释", "安抚客户", "投诉处理")
 _RECOMMEND_TERMS = ("推荐", "组合建议", "配置建议", "买什么", "适合配置", "筛选基金", "投资方案")
 _CUSTOMER_IDENTITY_TERMS = ("是谁", "姓名", "实名", "基本信息", "联系方式", "手机号")
+_CUSTOMER_ROSTER_TERMS = ("名单", "列表", "几个")
+_CUSTOMER_RISK_TERMS = (
+    "风险评级", "风险等级", "风险类型", "激进型", "进取型", "平衡型", "稳健型", "保守型",
+    "C1", "C2", "C3", "C4", "C5",
+)
+_IMPLICIT_QUERY_TERMS = (
+    "谁", "哪个", "哪些", "多少", "最大", "最多", "最小", "最少", "最高", "最低",
+    "合计", "总额", "总资产", "排名", "前几", "前十", "超过", "低于", "是否", "有没有",
+)
 _SCOPE_ERROR_MESSAGE = "投顾范围查询仅支持客户数据查询"
 
 
@@ -79,7 +89,23 @@ def recognize_advisor_intent(query: str | None, explicit_intent: str | None = No
     )
     if has_customer_identity:
         return AGENT_INTENT_DATA_QUERY
-    if has_data and (has_action or "客户" in text or "近一年" in text or "本月" in text):
+    # 客户名册类提问（我名下有哪些客户/客户名单）也属于数据查询。
+    has_customer_roster = (
+        "客户" in text
+        and (has_action or _contains_any(text, _CUSTOMER_ROSTER_TERMS))
+        and not _contains_any(text, _RECOMMEND_TERMS)
+    )
+    if has_customer_roster:
+        return AGENT_INTENT_DATA_QUERY
+    if "客户" in text and _contains_any(text, _CUSTOMER_RISK_TERMS) and not _contains_any(text, _RECOMMEND_TERMS):
+        return AGENT_INTENT_DATA_QUERY
+    if has_data and (
+        has_action
+        or "客户" in text
+        or "近一年" in text
+        or "本月" in text
+        or _contains_any(text, _IMPLICIT_QUERY_TERMS)
+    ):
         return AGENT_INTENT_DATA_QUERY
     if _contains_any(text, _RECOMMEND_TERMS):
         return AGENT_INTENT_RECOMMEND

api/routers/advisor_agent.py

@@ -13,7 +13,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 from agent.advisor_agent.auth import ensure_customer_access
 from agent.advisor_agent.data_query import execute_advisor_data_query
 from agent.advisor_agent.intent.fund_analysis import build_fund_analysis
-from agent.advisor_agent.intent.classifier import classify_advisor_intent
+from agent.advisor_agent.intent.classifier import classify_advisor_intent, resolve_advisor_intent
 from agent.advisor_agent.intent.talk_script import build_talk_script
 from agent.advisor_agent.llm import generate_text
 from agent.advisor_agent.intent.generation_flow import (
@@ -69,6 +69,7 @@ from schemas.advisor_agent import (
     AdvisorDataQueryReq,
 )
 from service.nl2sql.query_service import QueryServiceError
+from nl2sql.query_rewriter import rewrite_query
 from service.advisor_agent.draft import (
     detail_draft,
     discard_draft,
@@ -248,10 +249,20 @@ async def chat_stream(
         )
 
     runtime = _advisor_runtime(request)
-    classification = await classify_advisor_intent(
+    context_store = SessionContextStore(redis_db.client())
+    conversation_context = build_conversation_context(
+        await context_store.load(user.id, chat_request.session_id)
+    )
+    effective_question = await rewrite_query(
         chat_request.query,
+        conversation_context,
+        llm_client=getattr(runtime, "llm_client", None),
+    )
+    classification = await classify_advisor_intent(
+        effective_question,
         getattr(runtime, "llm_client", None),
         explicit_intent=None,
+        conversation_context=conversation_context,
     )
     inferred_intent = classification.intent
     scope = chat_request.scope
@@ -277,11 +288,16 @@ async def chat_stream(
         await ensure_customer_access(db, advisor_id=user.id, customer_id=int(customer_id))
     else:
         resolved_customer_id, _resolve_error = await _resolve_customer_from_query(
-            db, advisor_id=user.id, query=chat_request.query
+            db, advisor_id=user.id, query=effective_question
         )
         if resolved_customer_id is not None:
             customer_id = resolved_customer_id
             scope = "customer"
+    inferred_intent = resolve_advisor_intent(
+        effective_question,
+        inferred_intent,
+        customer_resolved=customer_id is not None,
+    )
 
     if customer_id is None:
         if inferred_intent == AGENT_INTENT_DATA_QUERY:
@@ -317,20 +333,17 @@ async def chat_stream(
     if inferred_intent == AGENT_INTENT_DATA_QUERY:
         effective_scope = "customer" if customer_id is not None else "advisor"
         try:
-            context_store = SessionContextStore(redis_db.client())
-            conversation_context = build_conversation_context(
-                await context_store.load(user.id, chat_request.session_id)
-            )
             result = await execute_advisor_data_query(
                 db,
                 advisor_id=user.id,
                 customer_id=int(customer_id) if customer_id is not None else None,
                 scope=effective_scope,
-                question=chat_request.query,
+                question=effective_question,
                 trace_id=trace_id,
                 session_id=chat_request.session_id,
                 conversation_context=conversation_context,
                 llm_client=getattr(_advisor_runtime(request), "llm_client", None),
+                redis=redis_db.client(),
             )
         except QueryServiceError as exc:
             payload = agent_failure(ERR_CODE_LLM_ERROR, _data_query_error_message(exc), trace_id=trace_id)
@@ -565,6 +578,7 @@ async def advisor_data_query(
             page_size=body.page_size,
             sort_by=body.sort_by,
             sort_order=body.sort_order,
+            redis=redis_db.client(),
         )
     except QueryServiceError as exc:
         return agent_failure(

api/routers/nl2sql.py

@@ -248,6 +248,7 @@ async def query_data(
             db,
             database=settings.mysql.database,
             candidate_tables=table_names,
+            redis=redis,
         )
 
     request_contract = DataQueryRequest(

design-preview/index.html

@@ -354,16 +354,19 @@
     <div class="ca-head">
       <div class="row" style="gap:10px">
         <div class="brand-logo" style="width:34px;height:34px;font-size:15px">🤖</div>
-        <div><div class="t">华夏基金智能助手</div><div class="s">在线为您提供基金信息服务</div></div>
+        <div><div class="t">华夏基金智能助手</div><div class="s" id="ca-state">在线为您提供基金信息服务</div></div>
       </div>
-      <button class="btn btn-ghost btn-sm" style="color:#bfdbfe" onclick="toggleCA()">✕</button>
+      <button class="btn btn-ghost btn-sm" style="color:#bfdbfe" onclick="closeCA()">✕</button>
     </div>
     <div class="ca-body">
-      <div class="m bot"><div class="bl">您好，我是华夏基金智能助手。您可以咨询基金产品、风险等级和净值信息。</div></div>
-      <div class="m usr"><div class="bl">稳健型的我能买股票型基金吗？</div></div>
-      <div class="m bot"><div class="bl">风险等级 C2（保守型）通常不建议购买 R4 股票型基金，
-        两者风险等级不匹配。您可以关注 R2 债券型或 R3 混合型产品。<br><br>
-        <span class="faint xs">以上信息仅供参考，不构成投资建议。</span></div></div>
+      <div id="ca-connecting" class="xs" style="color:#94a3b8;padding:6px 0">正在建立会话...</div>
+      <div id="ca-messages" style="display:none">
+        <div class="m bot"><div class="bl">您好，我是华夏基金智能助手。您可以咨询基金产品、风险等级和净值信息。</div></div>
+        <div class="m usr"><div class="bl">稳健型的我能买股票型基金吗？</div></div>
+        <div class="m bot"><div class="bl">风险等级 C2（保守型）通常不建议购买 R4 股票型基金，
+          两者风险等级不匹配。您可以关注 R2 债券型或 R3 混合型产品。<br><br>
+          <span class="faint xs">以上信息仅供参考，不构成投资建议。</span></div></div>
+      </div>
     </div>
     <div class="ca-foot">
       <input class="input" placeholder="输入您的问题">
@@ -1020,11 +1023,35 @@ function showScreen(id){
   /* 客户/游客助手挂在全局 layout，后台不显示（后台用投顾助手悬浮球） */
   document.getElementById("client-agent").style.display =
     NO_CLIENT_AGENT.includes(id) ? "none" : "block";
-  document.getElementById("ca-win").style.display = "none";
+  closeCA();
+}
+/* 打开面板 = 先建会话（session/create），拿到会话号后才允许对话 */
+function openCA(){
+  const w = document.getElementById("ca-win");
+  if (w.style.display !== "none") return;
+  w.style.display = "flex";
+  const state = document.getElementById("ca-state");
+  document.getElementById("ca-connecting").style.display = "block";
+  document.getElementById("ca-messages").style.display = "none";
+  state.textContent = "正在建立会话...";
+  clearTimeout(window.__caTimer);
+  window.__caTimer = setTimeout(()=>{
+    state.textContent = "会话 #" + Math.random().toString(16).slice(2,8) + " · 已接入您的账户数据";
+    document.getElementById("ca-connecting").style.display = "none";
+    document.getElementById("ca-messages").style.display = "block";
+  }, 700);
+}
+/* 关闭面板 = 结束会话（session/end），丢弃服务端上下文 */
+function closeCA(){
+  const w = document.getElementById("ca-win");
+  if (w.style.display === "none") return;
+  w.style.display = "none";
+  clearTimeout(window.__caTimer);
+  document.getElementById("ca-state").textContent = "会话已结束";
 }
 function toggleCA(){
   const w = document.getElementById("ca-win");
-  w.style.display = w.style.display==="none" ? "flex" : "none";
+  w.style.display === "none" ? openCA() : closeCA();
 }
 function go(id){
   const sec = ADMIN_SECTIONS.find(s=>s.id===id);

frontend/components/admin/assistant-launcher.tsx

@@ -58,7 +58,8 @@ export function AssistantLauncher() {
   const [open, setOpen] = useState(false);
   const [tab, setTab] = useState<"chat" | "drafts">("chat");
 
-  const [sessionId, setSessionId] = useState(() => newSessionId());
+  /* 会话在「打开抽屉」时才建立（见 openAssistant），关闭时丢弃，不在挂载时就占用上下文 */
+  const [sessionId, setSessionId] = useState<string | null>(null);
   const [customers, setCustomers] = useState<Customer[]>([]);
   const [customerId, setCustomerId] = useState<number | null>(null);
   const [messages, setMessages] = useState<ChatMessage[]>([]);
@@ -87,7 +88,7 @@ export function AssistantLauncher() {
     return () => clearInterval(timer);
   }, [streaming]);
 
-  const sessionTag = useMemo(() => sessionId.slice(-6), [sessionId]);
+  const sessionTag = useMemo(() => (sessionId ? sessionId.slice(-6) : "--------"), [sessionId]);
 
   /* 客户列表只用于会话上下文选择，失败不阻断助手 */
   useEffect(() => {
@@ -101,8 +102,18 @@ export function AssistantLauncher() {
     })();
   }, []);
 
-  /* 会话 ID 变化时恢复该会话的历史；新会话为空属正常，不报错。 */
+  /*
+   * 会话握手：打开抽屉（sessionId 就绪）后拉一次该会话的记录。
+   * 投顾 Agent 没有独立的 session/create，后端按 (advisor_id, session_id) 惰性建立上下文，
+   * 所以这个请求就是「打开助手时发出的会话请求」；关闭抽屉会把 sessionId 置空、丢弃上下文。
+   */
   useEffect(() => {
+    if (!sessionId) {
+      setMessages([]);
+      setHistoryError(null);
+      setHistoryLoading(false);
+      return;
+    }
     let alive = true;
     void (async () => {
       setHistoryLoading(true);
@@ -182,7 +193,7 @@ export function AssistantLauncher() {
         const result = await assistantApi.chatOnce(
           {
             query,
-            session_id: sessionId,
+            session_id: sessionId ?? undefined,
             scope: customerId ? "customer" : "advisor",
             customer_id: customerId,
           },
@@ -241,7 +252,7 @@ export function AssistantLauncher() {
       setStreaming(true);
       try {
         if (tool === "data-query") {
-          const result = await assistantApi.dataQuery({ query, session_id: sessionId });
+          const result = await assistantApi.dataQuery({ query, session_id: sessionId ?? undefined });
           pushPair(query, result.answer ?? result.summary ?? "查询完成，无文本摘要。", "data_query");
         } else if (tool === "fund-analysis") {
           const result = await assistantApi.fundAnalysis(query);
@@ -272,7 +283,30 @@ export function AssistantLauncher() {
     [customerId, input, loadDrafts, pushPair, sessionId, toast]
   );
 
-  /** 结束会话：中止生成、清空上下文并启用新的 session_id。 */
+  /**
+   * 打开抽屉 = 建立会话。
+   * 投顾 Agent 后端只有 chat/stream 与 session/{id}/history，没有独立的 session/create，
+   * 所以这里生成会话键，由上面的 useEffect 调 history 完成握手（后端按需建立上下文）。
+   */
+  const openAssistant = useCallback(() => {
+    setOpen(true);
+    setSessionId((current) => current ?? newSessionId());
+  }, []);
+
+  /**
+   * 关闭抽屉 = 结束会话：中止生成、清空消息并丢弃上下文。
+   * 投顾侧后端未提供 session/end，因此结束只能在前端完成（下次打开是新的 session_id）。
+   */
+  const closeAssistant = useCallback(() => {
+    abortRef.current?.abort();
+    setOpen(false);
+    setMessages([]);
+    setInput("");
+    setStreaming(false);
+    setSessionId(null);
+  }, []);
+
+  /** 抽屉内「结束会话」：不等关闭，立即换一个全新会话并重新握手。 */
   const endSession = useCallback(() => {
     abortRef.current?.abort();
     setMessages([]);
@@ -328,7 +362,7 @@ export function AssistantLauncher() {
     <>
       {!open && (
         <button
-          onClick={() => setOpen(true)}
+          onClick={openAssistant}
           aria-label="打开投顾助手"
           className="fixed bottom-6 right-6 z-40 flex h-14 w-14 items-center justify-center rounded-full bg-indigo-600 text-white shadow-lg shadow-indigo-600/30 transition-transform hover:scale-105 hover:bg-indigo-700"
         >
@@ -345,7 +379,7 @@ export function AssistantLauncher() {
         <>
           <div
             className="fixed inset-0 z-40 bg-slate-900/10"
-            onClick={() => setOpen(false)}
+            onClick={closeAssistant}
             aria-hidden
           />
           <aside className="fixed inset-y-0 right-0 z-50 flex w-full max-w-[560px] flex-col bg-white shadow-2xl">
@@ -370,7 +404,7 @@ export function AssistantLauncher() {
                 结束会话
               </Button>
               <button
-                onClick={() => setOpen(false)}
+                onClick={closeAssistant}
                 aria-label="收起助手"
                 className="rounded-lg p-2 text-slate-400 transition-colors hover:bg-slate-100 hover:text-slate-600"
               >
@@ -414,7 +448,7 @@ export function AssistantLauncher() {
               <>
                 <div ref={scrollRef} className="flex-1 space-y-4 overflow-y-auto px-5 py-4">
                   {historyLoading && messages.length === 0 && (
-                    <p className="py-8 text-center text-sm text-slate-400">正在恢复会话...</p>
+                    <p className="py-8 text-center text-sm text-slate-400">正在建立会话...</p>
                   )}
                   {historyError && (
                     <p className="rounded-lg bg-red-50 px-3 py-2 text-xs text-red-600">{historyError}</p>

frontend/components/agent-chat.tsx

@@ -1,12 +1,11 @@
 "use client";
 
-import { Bot, MessageCircle, Send, X } from "lucide-react";
-import { useState } from "react";
-import { apiFetch, apiStream } from "@/lib/api";
-
-interface AgentChatProps {
-  employeeMode?: boolean;
-}
+import { Bot, MessageCircle, RotateCw, Send, X } from "lucide-react";
+import { useCallback, useEffect, useRef, useState } from "react";
+import { apiFetch } from "@/lib/api";
+import { getToken } from "@/lib/auth";
+import { chatAgentApi, type ChatAgentMode } from "@/lib/chat-agent-api";
+import { USER_TYPE } from "@/lib/roles";
 
 interface ChatMessage {
   id: number;
@@ -14,72 +13,233 @@ interface ChatMessage {
   content: string;
 }
 
-export function AgentChat({ employeeMode = false }: AgentChatProps) {
+const GREETING = "您好，我是华夏基金智能助手。您可以咨询基金产品、风险等级和净值信息。";
+
+/**
+ * 选择用哪个 Agent：
+ * - 已登录客户 → /agent/client/*（能读本人持仓与交易）
+ * - 游客 / 员工 → /agent/customer/*（匿名客服，不需要 token）
+ * 探测失败一律退回匿名客服，不阻断用户。
+ */
+async function resolveMode(): Promise<ChatAgentMode> {
+  if (!getToken()) return "customer";
+  try {
+    const result = await apiFetch<{ user?: { user_type?: string | null } }>("/auth/me");
+    return result?.user?.user_type === USER_TYPE.CUSTOMER ? "client" : "customer";
+  } catch {
+    return "customer";
+  }
+}
+
+export function AgentChat() {
   const [open, setOpen] = useState(false);
   const [input, setInput] = useState("");
-  const [loading, setLoading] = useState(false);
+  const [sending, setSending] = useState(false);
+  const [connecting, setConnecting] = useState(false);
   const [sessionId, setSessionId] = useState<string | null>(null);
+  const [mode, setMode] = useState<ChatAgentMode>("customer");
+  const [error, setError] = useState<string | null>(null);
   const [messages, setMessages] = useState<ChatMessage[]>([
-    { id: 1, role: "agent", content: "您好，我是华夏基金智能助手。您可以咨询基金产品、风险等级和净值信息。" },
+    { id: 1, role: "agent", content: GREETING },
   ]);
 
-  async function ensureSession() {
-    if (employeeMode) return "";
-    if (sessionId) return sessionId;
-    const result = await apiFetch<{ session_id: string }>("/agent/customer/session/create", { method: "POST" });
-    setSessionId(result.session_id);
-    return result.session_id;
-  }
+  const abortRef = useRef<AbortController | null>(null);
+  const listRef = useRef<HTMLDivElement | null>(null);
 
-  async function sendMessage() {
-    const content = input.trim();
-    if (!content || loading) return;
-    setInput("");
-    setMessages((current) => [...current, { id: Date.now(), role: "user", content }]);
-    setLoading(true);
+  useEffect(() => () => abortRef.current?.abort(), []);
+  useEffect(() => {
+    listRef.current?.scrollTo({ top: listRef.current.scrollHeight, behavior: "smooth" });
+  }, [messages, connecting]);
 
+  /** 建立会话：探测 Agent 类型 → session/create 换取后端签发的 session_id。 */
+  const connect = useCallback(async () => {
+    setConnecting(true);
+    setError(null);
     try {
-      const activeSessionId = await ensureSession();
-      const response = employeeMode
-        ? await apiStream("/advisor-agent/chat/stream", { query: content })
-        : await apiStream("/agent/customer/chat", { session_id: activeSessionId, query: content });
-      setMessages((current) => [...current, { id: Date.now() + 1, role: "agent", content: response || "暂时没有找到合适的回答，请稍后再试。" }]);
-    } catch {
-      setMessages((current) => [...current, { id: Date.now() + 1, role: "agent", content: "当前服务暂时不可用，请稍后再试。" }]);
+      const resolved = await resolveMode();
+      const session = await chatAgentApi.createSession(resolved);
+      setMode(resolved);
+      setSessionId(session.session_id);
+    } catch (err) {
+      setError(err instanceof Error ? err.message : "会话建立失败，请重试");
     } finally {
-      setLoading(false);
+      setConnecting(false);
     }
-  }
+  }, []);
 
-  async function closeChat() {
+  /** 打开面板即建立会话（不是等用户第一次提问才建）。 */
+  const openChat = useCallback(() => {
+    setOpen(true);
+    if (!sessionId && !connecting) void connect();
+  }, [connect, connecting, sessionId]);
+
+  /** 关闭面板即结束会话：通知后端清理上下文与限流键，本地一并重置。 */
+  const closeChat = useCallback(async () => {
+    abortRef.current?.abort();
     setOpen(false);
-    if (!sessionId || employeeMode) return;
-    try { await apiFetch("/agent/customer/session/end", { method: "POST", body: JSON.stringify({ session_id: sessionId }) }); } catch { /* session cleanup is best effort */ }
+    setInput("");
+    setError(null);
+    setMessages([{ id: Date.now(), role: "agent", content: GREETING }]);
+    const current = sessionId;
     setSessionId(null);
-  }
+    if (!current) return;
+    try {
+      await chatAgentApi.endSession(mode, current);
+    } catch {
+      /* 会话清理失败不阻断用户 */
+    }
+  }, [mode, sessionId]);
+
+  const sendMessage = useCallback(async () => {
+    const content = input.trim();
+    if (!content || sending || connecting) return;
+    if (!sessionId) {
+      setError("会话尚未建立，请点击重试。");
+      return;
+    }
+    setInput("");
+    setError(null);
+    setMessages((prev) => [...prev, { id: Date.now(), role: "user", content }]);
+    setSending(true);
+
+    const controller = new AbortController();
+    abortRef.current = controller;
+    try {
+      const result = await chatAgentApi.chat(
+        mode,
+        { session_id: sessionId, query: content },
+        controller.signal
+      );
+      setMessages((prev) => [
+        ...prev,
+        {
+          id: Date.now() + 1,
+          role: "agent",
+          content: result.error || result.text || "暂时没有找到合适的回答，请稍后再试。",
+        },
+      ]);
+    } catch (err) {
+      if ((err as Error)?.name === "AbortError") return;
+      setMessages((prev) => [
+        ...prev,
+        {
+          id: Date.now() + 1,
+          role: "agent",
+          content: err instanceof Error ? err.message : "当前服务暂时不可用，请稍后再试。",
+        },
+      ]);
+    } finally {
+      setSending(false);
+      abortRef.current = null;
+    }
+  }, [connecting, input, mode, sending, sessionId]);
+
+  const ready = !!sessionId;
 
   return (
     <div className="fixed bottom-3 right-7 z-50">
       {open && (
-        <div role="dialog" aria-label="华夏基金智能助手" className="mb-4 flex h-[520px] w-[380px] flex-col overflow-hidden rounded-2xl border border-slate-200 bg-white shadow-2xl">
+        <div
+          role="dialog"
+          aria-label="华夏基金智能助手"
+          className="mb-4 flex h-[520px] w-[380px] flex-col overflow-hidden rounded-2xl border border-slate-200 bg-white shadow-2xl"
+        >
           <div className="flex items-center justify-between bg-[var(--brand-navy)] px-5 py-4 text-white">
             <div className="flex items-center gap-3">
-              <div className="flex h-9 w-9 items-center justify-center rounded-xl bg-[var(--brand-primary)]"><Bot className="h-5 w-5" /></div>
-              <div><p className="text-sm font-semibold">华夏基金智能助手</p><p className="mt-0.5 text-xs text-blue-100">在线为您提供基金信息服务</p></div>
+              <div className="flex h-9 w-9 items-center justify-center rounded-xl bg-[var(--brand-primary)]">
+                <Bot className="h-5 w-5" />
+              </div>
+              <div>
+                <p className="text-sm font-semibold">华夏基金智能助手</p>
+                <p className="mt-0.5 text-xs text-blue-100">
+                  {mode === "client" ? "已接入您的账户数据" : "在线为您提供基金信息服务"}
+                </p>
+              </div>
             </div>
-            <button aria-label="关闭聊天框" onClick={() => void closeChat()} className="rounded-lg p-2 text-blue-100 transition hover:bg-white/10 hover:text-white"><X className="h-4 w-4" /></button>
+            <button
+              aria-label="关闭聊天框"
+              onClick={() => void closeChat()}
+              className="rounded-lg p-2 text-blue-100 transition hover:bg-white/10 hover:text-white"
+            >
+              <X className="h-4 w-4" />
+            </button>
           </div>
-          <div className="flex-1 space-y-4 overflow-y-auto bg-slate-50 p-4">
-            {messages.map((message) => <div key={message.id} className={`flex ${message.role === "user" ? "justify-end" : "justify-start"}`}><div className={`max-w-[82%] rounded-2xl px-4 py-3 text-sm leading-6 ${message.role === "user" ? "rounded-br-md bg-[var(--brand-primary)] text-white" : "rounded-bl-md border border-slate-200 bg-white text-slate-700"}`}>{message.content}</div></div>)}
-            {loading && <div className="text-xs text-slate-400">正在整理信息...</div>}
+
+          <div ref={listRef} className="flex-1 space-y-4 overflow-y-auto bg-slate-50 p-4">
+            {messages.map((message) => (
+              <div
+                key={message.id}
+                className={`flex ${message.role === "user" ? "justify-end" : "justify-start"}`}
+              >
+                <div
+                  className={`max-w-[82%] rounded-2xl px-4 py-3 text-sm leading-6 ${
+                    message.role === "user"
+                      ? "rounded-br-md bg-[var(--brand-primary)] text-white"
+                      : "rounded-bl-md border border-slate-200 bg-white text-slate-700"
+                  }`}
+                >
+                  {message.content}
+                </div>
+              </div>
+            ))}
+
+            {connecting && (
+              <div className="flex items-center gap-2 text-xs text-slate-400">
+                <RotateCw className="h-3 w-3 animate-spin" />
+                正在建立会话...
+              </div>
+            )}
+            {sending && !connecting && <div className="text-xs text-slate-400">正在整理信息...</div>}
           </div>
-          <form onSubmit={(event) => { event.preventDefault(); void sendMessage(); }} className="flex gap-2 border-t border-slate-200 bg-white p-3">
-            <input value={input} onChange={(event) => setInput(event.target.value)} placeholder="输入您的问题" className="min-w-0 flex-1 rounded-xl border border-slate-200 px-3 py-2.5 text-sm outline-none transition focus:border-[var(--brand-primary)]" />
-            <button type="submit" aria-label="发送消息" className="flex h-10 w-10 shrink-0 items-center justify-center rounded-xl bg-[var(--brand-primary)] text-white transition hover:bg-[var(--brand-primary-dark)] disabled:cursor-not-allowed disabled:opacity-50" disabled={loading || !input.trim()}><Send className="h-4 w-4" /></button>
+
+          {error && (
+            <div className="flex items-center justify-between gap-3 border-t border-rose-100 bg-rose-50 px-4 py-2.5">
+              <span className="text-xs text-rose-600">{error}</span>
+              {!ready && (
+                <button
+                  onClick={() => void connect()}
+                  disabled={connecting}
+                  className="shrink-0 rounded-lg border border-rose-200 bg-white px-2.5 py-1 text-xs text-rose-600 transition hover:bg-rose-100 disabled:opacity-50"
+                >
+                  重试
+                </button>
+              )}
+            </div>
+          )}
+
+          <form
+            onSubmit={(event) => {
+              event.preventDefault();
+              void sendMessage();
+            }}
+            className="flex gap-2 border-t border-slate-200 bg-white p-3"
+          >
+            <input
+              value={input}
+              onChange={(event) => setInput(event.target.value)}
+              placeholder={ready ? "输入您的问题" : "正在准备会话..."}
+              disabled={connecting}
+              className="min-w-0 flex-1 rounded-xl border border-slate-200 px-3 py-2.5 text-sm outline-none transition focus:border-[var(--brand-primary)] disabled:bg-slate-50"
+            />
+            <button
+              type="submit"
+              aria-label="发送消息"
+              className="flex h-10 w-10 shrink-0 items-center justify-center rounded-xl bg-[var(--brand-primary)] text-white transition hover:bg-[var(--brand-primary-dark)] disabled:cursor-not-allowed disabled:opacity-50"
+              disabled={sending || connecting || !input.trim()}
+            >
+              <Send className="h-4 w-4" />
+            </button>
           </form>
         </div>
       )}
-      <button aria-label={open ? "收起智能助手" : "打开智能助手"} onClick={() => setOpen((value) => !value)} className="flex h-14 w-14 items-center justify-center rounded-full bg-[var(--brand-primary)] text-white shadow-lg shadow-indigo-900/25 transition hover:-translate-y-0.5 hover:bg-[var(--brand-primary-dark)]">{open ? <X className="h-5 w-5" /> : <MessageCircle className="h-5 w-5" />}</button>
+
+      <button
+        aria-label={open ? "收起智能助手" : "打开智能助手"}
+        onClick={() => (open ? void closeChat() : openChat())}
+        className="flex h-14 w-14 items-center justify-center rounded-full bg-[var(--brand-primary)] text-white shadow-lg shadow-indigo-900/25 transition hover:-translate-y-0.5 hover:bg-[var(--brand-primary-dark)]"
+      >
+        {open ? <X className="h-5 w-5" /> : <MessageCircle className="h-5 w-5" />}
+      </button>
     </div>
   );
 }

frontend/lib/api.ts

@@ -29,6 +29,22 @@ function handleUnauthorized(status: number) {
   if (!window.location.pathname.startsWith("/login")) window.location.href = "/login";
 }
 
+/**
+ * 从错误响应体里取后端给的原因（如「会话不存在或已过期」）。
+ * 非 2xx 时后端返回的是 JSON（utils/response.fail），部分场景也会包成 SSE。
+ */
+function extractErrorMessage(body: string): string | undefined {
+  const text = body.trim();
+  if (!text) return undefined;
+  const jsonText = text.startsWith("data:") ? text.split("\n")[0].slice(5).trim() : text;
+  try {
+    const payload = JSON.parse(jsonText) as { message?: string; error?: string };
+    return payload?.message ?? payload?.error;
+  } catch {
+    return undefined;
+  }
+}
+
 /**
  * 后端存在两套成功码，必须都认：
  * - utils/response.success → code 200（工作台、风控、工单等）
@@ -126,6 +142,8 @@ export interface SSESummary {
   traceId?: string;
   /** error 事件里的提示文案。 */
   error?: string;
+  /** 客服 Agent 命中的知识来源（裸 JSON 响应里的 sources）。 */
+  sources?: unknown[];
 }
 
 /**
@@ -152,7 +170,13 @@ export async function readSSE(
   });
 
   handleUnauthorized(response.status);
-  if (!response.ok) throw new ApiError("请求失败，请稍后重试", response.status);
+  if (!response.ok) {
+    const body = await response.text();
+    throw new ApiError(
+      extractErrorMessage(body) ?? "请求失败，请稍后重试",
+      response.status
+    );
+  }
 
   const summary: SSESummary = { text: "" };
   const raw = await response.text();
@@ -170,29 +194,45 @@ export async function readSSE(
       continue; /* 忽略心跳与非 JSON 块 */
     }
 
-    if (event.type === "text") {
-      summary.text += event.content ?? "";
-    } else if (event.type === "meta") {
-      summary.intent = event.intent;
-      summary.draftId = event.draft_id;
-      summary.queryId = event.query_id;
-      summary.traceId = event.trace_id;
-    } else if (event.type === "done") {
-      if (event.draft_id) summary.draftId = event.draft_id;
-      if (event.query_id) summary.queryId = event.query_id;
-    } else if (event.type === "error") {
-      summary.error = event.message ?? "助手返回错误，请稍后重试";
+    const record = event as Record<string, unknown>;
+    const type = typeof record.type === "string" ? record.type : undefined;
+
+    if (type === "text") {
+      summary.text += (record.content as string) ?? "";
+    } else if (type === "meta") {
+      summary.intent = record.intent as string | undefined;
+      summary.draftId = record.draft_id as string | undefined;
+      summary.queryId = record.query_id as string | undefined;
+      summary.traceId = record.trace_id as string | undefined;
+    } else if (type === "done") {
+      if (record.draft_id) summary.draftId = record.draft_id as string;
+      if (record.query_id) summary.queryId = record.query_id as string;
+    } else if (type === "error") {
+      summary.error = (record.message as string) ?? "助手返回错误，请稍后重试";
+    } else if (!type) {
+      /*
+       * 无 type 的响应——客服 Agent / 客户 Agent 不用事件包装，直接 dump 业务对象：
+       * service/customer_agent/chat.py → {answer, sources, intent, rewritten_query, trace_id}
+       * 另兼容 success() 包装的 {code, message, data}（取值时自动下钻一层 data）。
+       * 旧实现只认 type=text，导致这两个 Agent 的回复恒为空串（聊天框永远回「暂时没有找到合适的回答」）。
+       */
+      const node =
+        record.data && typeof record.data === "object"
+          ? (record.data as Record<string, unknown>)
+          : record;
+      const answer = node.answer ?? node.content ?? node.summary;
+      if (typeof answer === "string" && answer) {
+        summary.text += answer;
+      } else if (typeof record.message === "string" && record.message) {
+        summary.error = record.message;
+      }
+      if (typeof node.intent === "string") summary.intent = node.intent;
+      if (typeof node.query_id === "string") summary.queryId = node.query_id;
+      const trace = node.trace_id ?? record.trace_id;
+      if (typeof trace === "string") summary.traceId = trace;
+      if (Array.isArray(node.sources)) summary.sources = node.sources;
     }
   }
 
   return summary;
 }
-
-/**
- * 取 SSE 响应的正文文本（客服/客户 Agent 聊天用）。
- * 旧实现只取第一个 data 行，命中 meta 事件时会拿到空串，这里改为聚合全部 text 事件。
- */
-export async function apiStream(path: string, body: unknown): Promise<string> {
-  const summary = await readSSE(path, body);
-  return summary.error || summary.text;
-}

frontend/lib/assistant-api.ts

@@ -72,10 +72,7 @@ export const assistantApi = {
   sessionHistory: (sessionId: string) =>
     apiFetch<AgentMessage[]>(`/advisor-agent/session/${encodeURIComponent(sessionId)}/history`),
 
-  /**
-   * POST /chat/stream —— 响应是 text/event-stream，但后端整包输出，
-   * 因此读完整条响应后一次性返回聚合结果（非流式渲染）。
-   */
+  /** POST /chat/stream —— 读取 SSE，并支持调用方实时消费事件。 */
   chatOnce(
     body: {
       query: string;

frontend/lib/chat-agent-api.ts

@@ -0,0 +1,44 @@
+import { apiFetch, readSSE, type SSESummary } from "@/lib/api";
+
+/**
+ * 客户端两个 Agent 的会话适配层。
+ *
+ * 两者都是「打开建会话 → 带 session_id 对话 → 关闭结束会话」的三段式，
+ * 后端的会话归属校验会拒绝未创建或不属于自己的 session_id（SessionOwnershipError），
+ * 所以 session_id 必须来自 session/create，不能在前端自己造。
+ *
+ * - customer：匿名客服 Agent（/api/agent/customer/*），游客与登录用户都可用，无需 token
+ * - client  ：登录客户 Agent（/api/agent/client/*），需 require_customer，
+ *             能读到本人持仓 / 交易（匿名版没有 customer_id，数据查询会被引导去登录）
+ */
+export type ChatAgentMode = "customer" | "client";
+
+const PREFIX: Record<ChatAgentMode, string> = {
+  customer: "/agent/customer",
+  client: "/agent/client",
+};
+
+export interface AgentSession {
+  session_id: string;
+  customer_id: number | null;
+}
+
+export const chatAgentApi = {
+  /** POST /session/create —— 建立会话，返回后端签发的 session_id。 */
+  createSession: (mode: ChatAgentMode) =>
+    apiFetch<AgentSession>(`${PREFIX[mode]}/session/create`, { method: "POST" }),
+
+  /** POST /chat —— 响应是 SSE 包装，但内容为 {answer, sources, intent...}。 */
+  chat: (
+    mode: ChatAgentMode,
+    body: { session_id: string; query: string },
+    signal?: AbortSignal
+  ): Promise<SSESummary> => readSSE(`${PREFIX[mode]}/chat`, body, signal),
+
+  /** POST /session/end —— 关闭会话并清理服务端上下文与限流键。 */
+  endSession: (mode: ChatAgentMode, sessionId: string) =>
+    apiFetch<{ session_id: string; archived?: boolean }>(`${PREFIX[mode]}/session/end`, {
+      method: "POST",
+      body: JSON.stringify({ session_id: sessionId }),
+    }),
+};

frontend/next-env.d.ts

@@ -1,7 +1,7 @@
 /// <reference types="next" />
 /// <reference types="next/image-types/global" />
-import "./.next/dev/types/routes.d.ts";
-import "./.next/dev/types/root-params.d.ts";
+import "./.next/types/routes.d.ts";
+import "./.next/types/root-params.d.ts";
 
 // NOTE: This file should not be edited
 // see https://nextjs.org/docs/app/api-reference/config/typescript for more information.

main.py

@@ -90,6 +90,7 @@ async def lifespan(app: FastAPI):
         await app.state.advisor_event_consumer.stop()
     if app.state.advisor_scheduler is not None:
         app.state.advisor_scheduler.shutdown()
+    await llm_client.aclose()
     await database.dispose()
 
 

nl2sql/cache.py

@@ -49,6 +49,32 @@ def build_cache_key(
     return f"nl2sql:{digest}"
 
 
+def build_question_cache_key(
+    question: str,
+    *,
+    permission: dict[str, Any],
+    data_scope: dict[str, Any] | None = None,
+    page: int = 1,
+    page_size: int = 100,
+    sort_by: str | None = None,
+    sort_order: str = "asc",
+) -> str:
+    """生成自然语言查询结果缓存键，覆盖问题、权限和数据范围。"""
+    payload = {
+        "question": " ".join((question or "").split()),
+        "permission": _permission_payload(permission),
+        "data_scope": data_scope or {},
+        "page": page,
+        "page_size": page_size,
+        "sort_by": sort_by or "",
+        "sort_order": sort_order,
+    }
+    digest = hashlib.sha256(
+        json.dumps(payload, ensure_ascii=False, sort_keys=True, separators=(",", ":")).encode()
+    ).hexdigest()
+    return f"nl2sql:question:{digest}"
+
+
 async def cache_get(redis, key: str) -> dict[str, Any] | None:
     """读取缓存，Redis 异常或内容损坏时返回空结果。"""
     try:

nl2sql/query_rewriter.py

@@ -16,6 +16,29 @@ _SYSTEM_PROMPT = """你是基金平台 NL2SQL 的问题改写器。
 不能扩大权限范围，不能臆造查询结果；上下文不足时保留原问题。
 """
 
+_REFERENCE_TERMS = (
+    "他", "她", "它", "他们", "她们", "它们", "这个客户", "该客户", "那个客户",
+    "这个基金", "该基金", "那只基金", "这只基金", "这个结果", "上一轮", "刚才", "前面",
+)
+
+_CUSTOMER_RISK_LABELS = {
+    "保守型": "C1",
+    "稳健型": "C2",
+    "平衡型": "C3",
+    "进取型": "C4",
+    "激进型": "C5",
+}
+
+
+def _normalize_customer_risk_label(question: str) -> str:
+    """将展示层客户风险标签规范化为数据库 C1-C5 编码。"""
+    normalized = re.sub(r"\s+", "", question)
+    if "客户" not in normalized and "风险" not in normalized:
+        return normalized
+    for label, code in _CUSTOMER_RISK_LABELS.items():
+        normalized = normalized.replace(label, code)
+    return normalized
+
 
 def _rewrite_explicit_customer_identity(question: str) -> str | None:
     normalized = re.sub(r"\s+", "", question)
@@ -58,7 +81,7 @@ async def rewrite_query(
     llm_client=llm,
 ) -> str:
     """使用有限会话上下文补全问题；改写失败时安全返回原问题。"""
-    original = (question or "").strip()
+    original = _normalize_customer_risk_label((question or "").strip())
     context = (conversation_context or "").strip()
     if not original:
         return original
@@ -68,7 +91,7 @@ async def rewrite_query(
     contextual_identity = _rewrite_customer_name_follow_up(original, context)
     if contextual_identity:
         return contextual_identity
-    if not context or llm_client is None:
+    if not context or llm_client is None or not any(term in original for term in _REFERENCE_TERMS):
         return original
 
     prompt = (

nl2sql/result.py

@@ -67,6 +67,7 @@ async def render_answer(
                 {"role": "user", "content": prompt},
             ],
             temperature=0,
+            max_tokens=256,
         )
     except Exception:  # noqa: BLE001 摘要失败回退固定文本
         return fallback

nl2sql/retrieval.py

@@ -1,6 +1,7 @@
 """NL2SQL 元数据召回与授权过滤。"""
 from __future__ import annotations
 
+import asyncio
 from collections.abc import Awaitable, Callable, Iterable
 from typing import Any
 
@@ -70,15 +71,21 @@ async def retrieve_metadata(
         "is_valid",
         "is_deprecated",
     ]
-    results: list[dict[str, Any]] = []
-    for chunk_type in ("table_meta", "field_meta"):
-        hits = await milvus_client.search(
+    async def search_chunk(chunk_type: str):
+        return await milvus_client.search(
             collection_name=NL2SQL_COLLECTION,
             data=[vector],
             limit=top_k,
             filter=f'chunk_type == "{chunk_type}" and is_valid == true',
             output_fields=output_fields,
         )
+
+    table_hits, field_hits = await asyncio.gather(
+        search_chunk("table_meta"),
+        search_chunk("field_meta"),
+    )
+    results: list[dict[str, Any]] = []
+    for hits in (table_hits, field_hits):
         for hit in _flatten_hits(hits):
             entity = hit.get("entity") or hit
             if not entity.get("is_valid", True) or entity.get("is_deprecated", False):

nl2sql/schema.py

@@ -1,12 +1,23 @@
 """从 MySQL information_schema 加载并校验 NL2SQL 权威 Schema。"""
 from __future__ import annotations
 
+import hashlib
+import json
+import logging
 from typing import Any
 
 from sqlalchemy import bindparam, text
 
 from nl2sql.metadata import normalize_column_row, normalize_table_row
 
+logger = logging.getLogger("nl2sql.schema")
+SCHEMA_CACHE_TTL = 60
+
+
+def _schema_cache_key(database: str, candidates: list[str]) -> str:
+    payload = json.dumps([database, candidates], ensure_ascii=False, separators=(",", ":"))
+    return "nl2sql:schema:" + hashlib.sha256(payload.encode()).hexdigest()
+
 
 TABLES_SQL = text(
     """
@@ -34,13 +45,25 @@ async def load_authoritative_schema(
     *,
     database: str,
     candidate_tables: set[str] | list[str],
+    redis=None,
+    cache_ttl: int = SCHEMA_CACHE_TTL,
 ) -> dict[str, list[dict[str, Any]]]:
     """从权威元数据源加载候选表，并剔除不存在的表和孤立字段。"""
     candidates = {str(name).strip() for name in candidate_tables if str(name).strip()}
     if not candidates:
         return {"tables": [], "columns": []}
 
-    params = {"database": database, "candidate_tables": sorted(candidates)}
+    candidate_list = sorted(candidates)
+    cache_key = _schema_cache_key(database, candidate_list)
+    if redis is not None:
+        try:
+            cached = await redis.get(cache_key)
+            if cached:
+                return json.loads(cached)
+        except Exception:  # noqa: BLE001 缓存异常回退权威查询
+            logger.warning("Schema 缓存读取失败", exc_info=True)
+
+    params = {"database": database, "candidate_tables": candidate_list}
     table_result = await session.execute(TABLES_SQL, params)
     column_result = await session.execute(COLUMNS_SQL, params)
 
@@ -56,4 +79,10 @@ async def load_authoritative_schema(
         normalized = normalize_column_row(dict(row))
         if normalized["table_name"] in valid_table_names and normalized["field_name"]:
             columns.append(normalized)
-    return {"tables": tables, "columns": columns}
+    schema = {"tables": tables, "columns": columns}
+    if redis is not None:
+        try:
+            await redis.set(cache_key, json.dumps(schema, ensure_ascii=False), ex=cache_ttl)
+        except Exception:  # noqa: BLE001 缓存异常不阻断查询
+            logger.warning("Schema 缓存写入失败", exc_info=True)
+    return schema

nl2sql/semantic_catalog.json

@@ -18,12 +18,12 @@
       "value_hint": "fin_product.product_type 实际枚举值为：货币型/债券型/混合型/股票型/指数型/QDII；股票型基金必须用 product_type = '股票型'，不要使用 Schema 注释里的'股票基金'"
     },
     {
-      "term": "客户",
+      "term": "客户风险等级",
       "enabled": true,
-      "aliases": ["客户", "投资人", "持有人"],
+      "aliases": ["客户", "投资人", "持有人", "风险评级", "风险等级", "风险类型", "激进型", "进取型", "平衡型", "稳健型", "保守型", "C1", "C2", "C3", "C4", "C5"],
       "tables": ["fin_customer_profile", "fin_holdings", "fin_risk_assessment"],
       "fields": ["customer_id", "risk_level", "customer_level"],
-      "value_hint": "fin_customer_profile.risk_level / fin_risk_assessment.risk_level 实际枚举值为 R1-R5（R1 最保守，R5 最激进，与产品风险等级同一口径）；查'保守型/稳健型客户'对应 R1/R2，'激进型客户'对应 R5"
+      "value_hint": "客户风险字段实际枚举值为 C1-C5（C1 最保守，C5 最激进）；展示层映射为保守型=C1、稳健型=C2、平衡型=C3、进取型=C4、激进型=C5。产品风险等级才使用 R1-R5"
     },
     {
       "term": "持仓",

nl2sql/sql_agent.py

@@ -74,7 +74,7 @@ async def generate_sql(
         },
     ]
     try:
-        output = await llm_client.chat(messages, temperature=0)
+        output = await llm_client.chat(messages, temperature=0, max_tokens=256)
     except Exception as exc:  # noqa: BLE001 统一收敛模型异常
         raise SqlGenerationError("SQL 模型调用失败") from exc
     sql = _clean_model_output(output)

service/client_agent/runtime.py

@@ -304,7 +304,7 @@ def _build_data_query(*, db_session_factory, milvus_client, llm_client, config_g
 
         async def schema_loader(table_names: set[str], _permission: dict):
             return await load_authoritative_schema(
-                db, database=schema_database, candidate_tables=table_names
+                db, database=schema_database, candidate_tables=table_names, redis=redis
             )
 
         request = DataQueryRequest(

tool/llm.py

@@ -103,6 +103,21 @@ class LLMClient:
     def __init__(self, cfg: LLMCfg | None = None):
         self.cfg = cfg or settings.llm
         self.backends = resolve_backends(self.cfg)
+        self._clients: dict[str, httpx.AsyncClient] = {}
+
+    def _client_for(self, backend: Backend) -> httpx.AsyncClient:
+        clients = getattr(self, "_clients", None)
+        if clients is None:
+            clients = {}
+            self._clients = clients
+        if backend.name not in clients:
+            clients[backend.name] = backend.client(self.cfg.timeout)
+        return clients[backend.name]
+
+    async def aclose(self) -> None:
+        for client in getattr(self, "_clients", {}).values():
+            await client.aclose()
+        getattr(self, "_clients", {}).clear()
 
     # ---- 首选后端（健康检查/日志/embed 使用） -----------------------------
     @property
@@ -185,10 +200,14 @@ class LLMClient:
                 "max_tokens": request_max_tokens,
                 "stream": False,
             }
+            # SQL/分类等短输出任务不需要深度思考；DeepSeek-V4 若开启思考，
+            # 可能耗尽 token 预算而返回空的 content。
+            if model.lower().startswith("deepseek-v4"):
+                payload["enable_thinking"] = False
             try:
-                async with backend.client(self.cfg.timeout) as client:
-                    r = await client.post(url, headers=backend.headers, json=payload)
-                    r.raise_for_status()
+                client = self._client_for(backend)
+                r = await client.post(url, headers=backend.headers, json=payload)
+                r.raise_for_status()
                 choice = r.json()["choices"][0]
                 message = choice["message"]
                 content = message.get("content")
@@ -250,20 +269,16 @@ class LLMClient:
         response = None
         max_retries = max(1, int(getattr(self.cfg, "max_retries", 1)))
         retry_backoff_sec = float(getattr(self.cfg, "retry_backoff_sec", 0))
-        async with backend.client(self.cfg.timeout) as client:
-            for attempt in range(max_retries):
-                try:
-                    response = await client.post(
-                        url,
-                        headers=backend.headers,
-                        json=payload,
-                    )
-                    response.raise_for_status()
-                    break
-                except Exception:
-                    if attempt == max_retries - 1:
-                        raise
-                    await asyncio.sleep(retry_backoff_sec * (2**attempt))
+        client = self._client_for(backend)
+        for attempt in range(max_retries):
+            try:
+                response = await client.post(url, headers=backend.headers, json=payload)
+                response.raise_for_status()
+                break
+            except Exception:
+                if attempt == max_retries - 1:
+                    raise
+                await asyncio.sleep(retry_backoff_sec * (2**attempt))
         if response is None:  # pragma: no cover - defensive guard
             raise LLMFailError("Embedding 请求未返回响应")
         data = response.json()
@@ -286,10 +301,10 @@ class LLMClient:
             url = backend.base_url.removesuffix("/v1") + "/api/tags"
         else:
             url = f"{backend.base_url}/models"
-        async with backend.client(min(self.cfg.timeout, 10)) as client:
-            r = await client.get(url, headers=backend.headers)
-            if r.status_code >= 500:
-                r.raise_for_status()
+        client = self._client_for(backend)
+        r = await client.get(url, headers=backend.headers)
+        if r.status_code >= 500:
+            r.raise_for_status()
 
 
 # 全局单例：Agent 统一引入