diff --git a/ai_knowledge_base_v2/apps/backend/app/api/chat.py b/ai_knowledge_base_v2/apps/backend/app/api/chat.py
index 7a4f718..15f157c 100644
--- a/ai_knowledge_base_v2/apps/backend/app/api/chat.py
+++ b/ai_knowledge_base_v2/apps/backend/app/api/chat.py
@@ -268,7 +268,13 @@ async def _chat_stream(payload: ChatCompletionRequest, db: Session, current_user
             waitingCount=queue_request.waiting_count,
             reasoningVisible=reasoning_visible,
         )
-        chunks = ChatStreamService.stream_answer_async(db, current_user, payload.sessionId, payload.message)
+        chunks = ChatStreamService.stream_answer_async(
+            db,
+            current_user,
+            payload.sessionId,
+            payload.message,
+            retry_failed_question=payload.retry,
+        )
         async for segment in ReasoningPolicyService.iter_segments(chunks):
             if segment.kind == "content":
                 yield _sse_event("content", content=segment.content)
diff --git a/ai_knowledge_base_v2/apps/backend/app/schemas/chat.py b/ai_knowledge_base_v2/apps/backend/app/schemas/chat.py
index 931220c..0c19a86 100644
--- a/ai_knowledge_base_v2/apps/backend/app/schemas/chat.py
+++ b/ai_knowledge_base_v2/apps/backend/app/schemas/chat.py
@@ -37,6 +37,7 @@ class UpdateSessionTitleRequest(BaseModel):
 class ChatCompletionRequest(BaseModel):
     sessionId: int
     message: str = Field(min_length=1, max_length=4000)
+    retry: bool = False
 
 
 class StopChatRequest(BaseModel):
diff --git a/ai_knowledge_base_v2/apps/backend/app/services/chat_stream_service.py b/ai_knowledge_base_v2/apps/backend/app/services/chat_stream_service.py
index d9c8a3a..ca26f3a 100644
--- a/ai_knowledge_base_v2/apps/backend/app/services/chat_stream_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/chat_stream_service.py
@@ -10,7 +10,7 @@ from fastapi import HTTPException, status
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
-from app.models.chat import ChatMessage
+from app.models.chat import ChatMessage, TopicSession
 from app.models.knowledge import KnowledgeRetrievalLog
 from app.models.user import User
 from app.services.ai_request_log_service import AiRequestLogService
@@ -212,7 +212,14 @@ class ChatStreamService:
         db.commit()
 
     @staticmethod
-    async def stream_answer_async(db: Session, user: User, session_id: int, question: str) -> AsyncIterator[str]:
+    async def stream_answer_async(
+        db: Session,
+        user: User,
+        session_id: int,
+        question: str,
+        *,
+        retry_failed_question: bool = False,
+    ) -> AsyncIterator[str]:
         user = ChatService.prepare_daily_quota(db, user)
         session = ChatService._get_user_session(db, user, session_id)
         ChatService._ensure_quota(user)
@@ -225,25 +232,37 @@ class ChatStreamService:
 
         now = _now()
         normalized_question = question.strip()
-        topic = TopicSessionService.get_or_create_active(
-            db,
-            user=user,
-            session=session,
-            question=normalized_question,
-            deduct_quota=entitlement.deduct_quota,
+        user_message = (
+            _retryable_user_message(
+                db,
+                session_id=session.id,
+                user_id=user.id,
+                question=normalized_question,
+            )
+            if retry_failed_question
+            else None
         )
-        user_message = ChatMessage(
-            session_id=session.id,
-            topic_session_id=topic.id,
-            user_id=user.id,
-            role="user",
-            content=normalized_question,
-            message_status="FINISHED",
-            created_at=now,
-        )
-        db.add(user_message)
-        TopicSessionService.attach_user_message(user_message, topic)
-        db.flush()
+        topic = db.get(TopicSession, user_message.topic_session_id) if user_message and user_message.topic_session_id else None
+        if user_message is None or topic is None:
+            topic = TopicSessionService.get_or_create_active(
+                db,
+                user=user,
+                session=session,
+                question=normalized_question,
+                deduct_quota=entitlement.deduct_quota,
+            )
+            user_message = ChatMessage(
+                session_id=session.id,
+                topic_session_id=topic.id,
+                user_id=user.id,
+                role="user",
+                content=normalized_question,
+                message_status="FINISHED",
+                created_at=now,
+            )
+            db.add(user_message)
+            TopicSessionService.attach_user_message(user_message, topic)
+            db.flush()
 
         history = list(
             db.scalars(
@@ -362,6 +381,27 @@ def _now() -> datetime:
     return datetime.now(UTC).replace(tzinfo=None)
 
 
+def _retryable_user_message(
+    db: Session,
+    *,
+    session_id: int,
+    user_id: int,
+    question: str,
+) -> ChatMessage | None:
+    """Reuse only the latest unanswered user turn to avoid duplicate retry messages."""
+    latest = db.scalar(
+        select(ChatMessage)
+        .where(ChatMessage.session_id == session_id, ChatMessage.user_id == user_id)
+        .order_by(ChatMessage.id.desc())
+        .limit(1)
+    )
+    if latest is None or latest.role != "user":
+        return None
+    if latest.content.strip() != question.strip():
+        return None
+    return latest
+
+
 def _rough_token_count(text: str) -> int:
     return max(1, len(text.strip()) // 2)
 
diff --git a/ai_knowledge_base_v2/apps/backend/tests/test_chat_retry.py b/ai_knowledge_base_v2/apps/backend/tests/test_chat_retry.py
new file mode 100644
index 0000000..7a71894
--- /dev/null
+++ b/ai_knowledge_base_v2/apps/backend/tests/test_chat_retry.py
@@ -0,0 +1,59 @@
+from __future__ import annotations
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session
+from sqlalchemy.pool import StaticPool
+
+from app.models import Base
+from app.models.chat import ChatMessage, ChatSession
+from app.services.chat_stream_service import _retryable_user_message
+
+
+def _db() -> Session:
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(engine)
+    return Session(engine)
+
+
+def test_retry_reuses_only_latest_unanswered_matching_user_message():
+    with _db() as db:
+        session = ChatSession(id=1, user_id=7, title="重试测试", message_count=0, is_deleted=0)
+        user_message = ChatMessage(
+            id=1,
+            session_id=1,
+            user_id=7,
+            role="user",
+            content="原来的问题",
+            message_status="FINISHED",
+        )
+        db.add_all([session, user_message])
+        db.commit()
+
+        retried = _retryable_user_message(
+            db,
+            session_id=1,
+            user_id=7,
+            question=" 原来的问题 ",
+        )
+        assert retried is not None
+        assert retried.id == user_message.id
+
+        assert _retryable_user_message(db, session_id=1, user_id=7, question="另一个问题") is None
+
+        db.add(
+            ChatMessage(
+                id=2,
+                session_id=1,
+                user_id=7,
+                role="assistant",
+                content="已经回答",
+                message_status="FINISHED",
+            )
+        )
+        db.commit()
+
+        assert _retryable_user_message(db, session_id=1, user_id=7, question="原来的问题") is None
diff --git a/ai_knowledge_base_v2/apps/backend/tests/test_production_readiness.py b/ai_knowledge_base_v2/apps/backend/tests/test_production_readiness.py
index b8fd344..a2aca07 100644
--- a/ai_knowledge_base_v2/apps/backend/tests/test_production_readiness.py
+++ b/ai_knowledge_base_v2/apps/backend/tests/test_production_readiness.py
@@ -127,7 +127,7 @@ def test_queued_chat_reports_position_and_completes(monkeypatch):
     monkeypatch.setattr(chat.ChatStreamService, "stream_answer_async", stream_answer)
     monkeypatch.setattr(chat.ReasoningPolicyService, "is_visible", lambda _db: False)
 
-    payload = type("Payload", (), {"sessionId": 1, "message": "问题"})()
+    payload = type("Payload", (), {"sessionId": 1, "message": "问题", "retry": False})()
     async def collect_events():
         return [item async for item in chat._chat_stream(payload, object(), object())]
 
diff --git a/ai_knowledge_base_v2/apps/user-client/src/App.vue b/ai_knowledge_base_v2/apps/user-client/src/App.vue
index dd3e0be..d8fb630 100644
--- a/ai_knowledge_base_v2/apps/user-client/src/App.vue
+++ b/ai_knowledge_base_v2/apps/user-client/src/App.vue
@@ -137,7 +137,8 @@ async function selectSession(sessionId: number, force = false) {
 }
 
 async function send(message: string, complete: (success: boolean) => void) {
-  if (!activeSessionId.value || sending.value) {
+  const sessionId = activeSessionId.value;
+  if (!sessionId || sending.value) {
     complete(false);
     return;
   }
@@ -152,15 +153,35 @@ async function send(message: string, complete: (success: boolean) => void) {
   };
   messages.value.push(userMessage, assistantMessage);
   const assistantIndex = messages.value.length - 1;
+  complete(true);
+  await messageList.value?.scrollToMessage(userMessage.id);
+  await runGeneration(sessionId, message, assistantIndex, false);
+}
+
+async function retryMessage(messageId: string) {
+  const sessionId = activeSessionId.value;
+  if (!sessionId || sending.value) return;
+  const assistantIndex = messages.value.findIndex((item) => item.id === messageId);
+  const assistantMessage = messages.value[assistantIndex];
+  if (!assistantMessage?.retryQuestion) return;
+  await messageList.value?.scrollToMessage(messageId);
+  await runGeneration(sessionId, assistantMessage.retryQuestion, assistantIndex, true);
+}
+
+async function runGeneration(sessionId: number, message: string, assistantIndex: number, retry: boolean) {
   const currentAssistant = () => messages.value[assistantIndex];
+  currentAssistant().content = "";
+  currentAssistant().reasoning = "";
+  currentAssistant().showReasoning = false;
+  currentAssistant().errorMessage = undefined;
+  currentAssistant().streaming = true;
   sending.value = true;
   followingOutput.value = true;
   activeAbortController.value = new AbortController();
   let hasContent = false;
-  await messageList.value?.scrollToMessage(userMessage.id);
   try {
     await streamChat(
-      activeSessionId.value,
+      sessionId,
       message,
       async (chunk) => {
         if (!hasContent) {
@@ -183,21 +204,22 @@ async function send(message: string, complete: (success: boolean) => void) {
         if (followingOutput.value) await messageList.value?.scrollToBottom();
       },
       activeAbortController.value.signal,
+      retry,
     );
     currentAssistant().streaming = false;
+    currentAssistant().retryQuestion = undefined;
     currentAssistant().createdAt = new Date().toISOString();
-    complete(true);
     await refreshSessionList();
     await refreshProfile();
   } catch (error) {
     currentAssistant().streaming = false;
     if (error instanceof DOMException && error.name === "AbortError") {
       currentAssistant().content = currentAssistant().content || "已停止生成";
-      complete(true);
+      currentAssistant().retryQuestion = undefined;
     } else {
-      currentAssistant().content = currentAssistant().content || "回答生成失败，请稍后重试。";
-      complete(false);
-      handleError(error, "AI 回复失败");
+      currentAssistant().errorMessage = chatErrorMessage(error);
+      currentAssistant().retryQuestion = message;
+      showToast("本次回答未完成，可以点击重新生成");
     }
   } finally {
     sending.value = false;
@@ -205,6 +227,14 @@ async function send(message: string, complete: (success: boolean) => void) {
   }
 }
 
+function chatErrorMessage(error: unknown) {
+  const message = error instanceof Error ? error.message : "";
+  if (/429|too many|请求过多|排队/i.test(message)) return "当前请求较多，AI 服务暂时繁忙。";
+  if (/timeout|超时/i.test(message)) return "等待 AI 回复超时，请稍后重新生成。";
+  if (/network|网络|连接/i.test(message)) return "网络连接不稳定，本次回答没有完成。";
+  return "AI 回答生成失败，本次不会扣除成功回答额度。";
+}
+
 async function stop() {
   if (!activeSessionId.value || !sending.value) return;
   activeAbortController.value?.abort();
@@ -551,6 +581,7 @@ async function copyText(text: string) {
           :messages="messages"
           :loading-session="loadingSession"
           @follow-change="followingOutput = $event"
+          @retry="retryMessage"
         />
         <ChatComposer :loading="sending" :disabled="!activeSessionId || loadingSession" @send="send" @stop="stop" />
         <SessionDrawer
diff --git a/ai_knowledge_base_v2/apps/user-client/src/components/ChatMessage.vue b/ai_knowledge_base_v2/apps/user-client/src/components/ChatMessage.vue
index 7e8dd84..1de0f40 100644
--- a/ai_knowledge_base_v2/apps/user-client/src/components/ChatMessage.vue
+++ b/ai_knowledge_base_v2/apps/user-client/src/components/ChatMessage.vue
@@ -1,5 +1,5 @@
 <script setup lang="ts">
-import { Bot } from "@lucide/vue";
+import { Bot, RotateCcw } from "@lucide/vue";
 import MarkdownIt from "markdown-it";
 import { computed } from "vue";
 
@@ -11,6 +11,12 @@ const props = defineProps<{
   showReasoning?: boolean;
   createdAt: string;
   streaming?: boolean;
+  errorMessage?: string;
+  canRetry?: boolean;
+}>();
+
+const emit = defineEmits<{
+  retry: [];
 }>();
 
 const markdown = new MarkdownIt({ breaks: true, html: false, linkify: true });
@@ -95,6 +101,13 @@ const displayTime = computed(() => {
           <span></span><span></span><span></span>
           思考中
         </div>
+        <div v-if="errorMessage" class="message-error-state" role="alert">
+          <span>{{ errorMessage }}</span>
+          <button v-if="canRetry" type="button" :disabled="streaming" @click="emit('retry')">
+            <RotateCcw :size="15" aria-hidden="true" />
+            重新生成
+          </button>
+        </div>
       </template>
       <div v-else class="message-content">{{ renderedContent }}</div>
       <time v-if="displayTime" :datetime="createdAt">{{ displayTime }}</time>
diff --git a/ai_knowledge_base_v2/apps/user-client/src/components/MessageList.vue b/ai_knowledge_base_v2/apps/user-client/src/components/MessageList.vue
index 70d08b0..fef0a25 100644
--- a/ai_knowledge_base_v2/apps/user-client/src/components/MessageList.vue
+++ b/ai_knowledge_base_v2/apps/user-client/src/components/MessageList.vue
@@ -11,6 +11,8 @@ export interface DisplayMessage {
   showReasoning?: boolean;
   createdAt: string;
   streaming?: boolean;
+  errorMessage?: string;
+  retryQuestion?: string;
 }
 
 defineProps<{
@@ -20,6 +22,7 @@ defineProps<{
 
 const emit = defineEmits<{
   followChange: [following: boolean];
+  retry: [messageId: string];
 }>();
 
 const scroller = ref<HTMLElement | null>(null);
@@ -63,6 +66,9 @@ defineExpose({ scrollToBottom, scrollToMessage });
         :show-reasoning="message.showReasoning"
         :created-at="message.createdAt"
         :streaming="message.streaming"
+        :error-message="message.errorMessage"
+        :can-retry="Boolean(message.retryQuestion)"
+        @retry="emit('retry', message.id)"
       />
     </template>
   </section>
diff --git a/ai_knowledge_base_v2/apps/user-client/src/services/api.ts b/ai_knowledge_base_v2/apps/user-client/src/services/api.ts
index d0bc41b..1e75fbb 100644
--- a/ai_knowledge_base_v2/apps/user-client/src/services/api.ts
+++ b/ai_knowledge_base_v2/apps/user-client/src/services/api.ts
@@ -123,6 +123,7 @@ export async function streamChat(
   onReasoning?: (chunk: string) => void,
   onStatus?: (message: string, type: string, reasoningVisible: boolean) => void,
   signal?: AbortSignal,
+  retry = false,
 ) {
   const token = getToken();
   const response = await fetch(`${API_BASE}/chat/completions`, {
@@ -131,7 +132,7 @@ export async function streamChat(
       "Content-Type": "application/json",
       ...(token ? { Authorization: `Bearer ${token}` } : {}),
     },
-    body: JSON.stringify({ sessionId, message }),
+    body: JSON.stringify({ sessionId, message, retry }),
     signal,
   });
   if (!response.ok || !response.body) {
diff --git a/ai_knowledge_base_v2/apps/user-client/src/styles.css b/ai_knowledge_base_v2/apps/user-client/src/styles.css
index 76639a8..0641c85 100644
--- a/ai_knowledge_base_v2/apps/user-client/src/styles.css
+++ b/ai_knowledge_base_v2/apps/user-client/src/styles.css
@@ -1400,6 +1400,51 @@ textarea:focus-visible {
 .generation-state span:nth-child(2) { animation-delay: 0.14s; }
 .generation-state span:nth-child(3) { animation-delay: 0.28s; margin-right: 4px; }
 
+.message-error-state {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: 9px 12px;
+  margin-top: 10px;
+  padding: 10px 11px;
+  border: 1px solid #ead5cf;
+  border-radius: 11px;
+  background: #fff8f5;
+  color: #8a5043;
+  font-size: 13px;
+  line-height: 1.5;
+}
+
+.message-error-state span {
+  flex: 1 1 190px;
+}
+
+.message-error-state button {
+  min-height: 32px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
+  padding: 0 11px;
+  border: 1px solid #bf7765;
+  border-radius: 9px;
+  background: #ffffff;
+  color: #8a4434;
+  font: inherit;
+  font-weight: 700;
+  white-space: nowrap;
+  cursor: pointer;
+}
+
+.message-error-state button:hover {
+  background: #fff0eb;
+}
+
+.message-error-state button:disabled {
+  cursor: not-allowed;
+  opacity: 0.55;
+}
+
 @keyframes generation-pulse {
   0%, 70%, 100% { opacity: 0.25; transform: translateY(0); }
   35% { opacity: 1; transform: translateY(-2px); }