v0.2.1 修复对话体验：1)按钮兜底显示(配置加载失败也默认全开) 2)重新生成先删除旧回答再流式输出 3)推荐短语点击填入输入框 4)思考模型reasoning流式输出+自动折叠 5)语音输入改长按录音松开提交 6)新建对话直接用通用助手不弹窗

backend/app/api/chat.py

@@ -34,6 +34,7 @@ def _message_out(m: ChatMessage) -> dict:
         "id": m.id, "session_id": m.session_id, "role": m.role, "content": m.content,
         "model": m.model, "file_ids": file_ids, "feedback": m.feedback or "",
         "suggestions": suggestions, "edited": bool(m.edited), "regenerated": m.regenerated or 0,
+        "reasoning": m.reasoning_content or "",
         "created_at": m.created_at.isoformat(), "updated_at": m.updated_at.isoformat(),
     }
 
@@ -142,7 +143,7 @@ async def regenerate(session_id: int,
     """REST 非流式重新生成最后一条 AI 回答（备用，前端主要走 WS 流式）。"""
     s = _get_owned_session(db, user, session_id)
     try:
-        async for delta, message_id in chat_service.regenerate_stream(db, s):
+        async for _k, _d, message_id in _regenerate_rest(chat_service.regenerate_stream(db, s)):
             pass  # 流式丢弃，最终内容已入库
     except LookupError as e:
         raise HTTPException(status_code=404, detail=str(e))
@@ -150,3 +151,9 @@ async def regenerate(session_id: int,
         raise HTTPException(status_code=502, detail=f"模型调用失败：{e}")
     m = db.get(ChatMessage, message_id)
     return ok(_message_out(m))
+
+
+async def _regenerate_rest(agen):
+    """适配 regenerate_stream 的 (kind, delta), id 产出结构。"""
+    async for (kind, delta), message_id in agen:
+        yield kind, delta, message_id

backend/app/api/ws.py

@@ -1,7 +1,7 @@
 """WebSocket 路由：流式对话 + 重新生成。协议：
 客户端 → {"type":"chat","content":"...","agent_id":null,"model":"","file_ids":[]}
         {"type":"regenerate"}
-服务端 → {"type":"delta","message_id":1,"content":"增量文本"}
+服务端 → {"type":"delta","message_id":1,"kind":"reasoning"|"content","content":"增量文本"}
         {"type":"done","message_id":1}
         {"type":"suggestions","message_id":1,"items":["..",".."]}
         {"type":"title","title":"..."}
@@ -70,20 +70,20 @@ async def chat_ws(websocket: WebSocket):
                         await websocket.send_json({"type": "error", "message": "消息不能为空"})
                         continue
                     try:
-                        async for delta, message_id in chat_service.chat_stream(
+                        async for (kind, delta), message_id in chat_service.chat_stream(
                             db, user, session, content,
                             agent_id=data.get("agent_id"), model=data.get("model", ""),
                             file_ids=file_ids,
                         ):
-                            await websocket.send_json({"type": "delta", "message_id": message_id, "content": delta})
+                            await websocket.send_json({"type": "delta", "message_id": message_id, "kind": kind, "content": delta})
                         await websocket.send_json({"type": "done", "message_id": message_id})
                     except Exception as e:
                         await websocket.send_json({"type": "error", "message": f"模型调用失败：{e}"})
 
                 elif msg_type == "regenerate":
                     try:
-                        async for delta, message_id in chat_service.regenerate_stream(db, session):
-                            await websocket.send_json({"type": "delta", "message_id": message_id, "content": delta})
+                        async for (kind, delta), message_id in chat_service.regenerate_stream(db, session):
+                            await websocket.send_json({"type": "delta", "message_id": message_id, "kind": kind, "content": delta})
                         await websocket.send_json({"type": "done", "message_id": message_id})
                     except LookupError as e:
                         await websocket.send_json({"type": "error", "message": str(e)})

backend/app/core/llm.py

@@ -41,8 +41,10 @@ async def chat_completion_stream(
     model: Optional[str] = None,
     temperature: float = 0.7,
     max_tokens: int = 4096,
-) -> AsyncIterator[str]:
-    """流式对话补全：逐段产出增量文本。"""
+):
+    """流式对话补全：逐段产出增量。
+    产出 (kind, text)：kind="reasoning" 思考内容 / kind="content" 回答内容。
+    """
     client = _client_for(resolve_model(model))
     stream = await client.chat.completions.create(
         model=resolve_model(model),
@@ -52,8 +54,36 @@ async def chat_completion_stream(
         stream=True,
     )
     async for chunk in stream:
-        if chunk.choices and chunk.choices[0].delta and chunk.choices[0].delta.content:
-            yield chunk.choices[0].delta.content
+        if not chunk.choices or not chunk.choices[0].delta:
+            continue
+        delta = chunk.choices[0].delta
+        # 思考内容：deepseek-reasoner 用 reasoning_content，部分模型用 reasoning/thinking
+        reasoning = getattr(delta, "reasoning_content", None) or getattr(delta, "reasoning", None)
+        if reasoning:
+            yield ("reasoning", reasoning)
+            continue
+        if delta.content:
+            yield ("content", delta.content)
+
+
+async def chat_completion_full(
+    messages: list[dict],
+    model: Optional[str] = None,
+    temperature: float = 0.7,
+    max_tokens: int = 4096,
+) -> tuple[str, str]:
+    """非流式对话补全：返回 (回答内容, 思考内容)。"""
+    client = _client_for(resolve_model(model))
+    resp = await client.chat.completions.create(
+        model=resolve_model(model),
+        messages=messages,
+        temperature=temperature,
+        max_tokens=max_tokens,
+        stream=False,
+    )
+    msg = resp.choices[0].message
+    reasoning = getattr(msg, "reasoning_content", None) or getattr(msg, "reasoning", None) or ""
+    return (msg.content or ""), reasoning
 
 
 async def vision_analysis(prompt: str, image_urls: list[str], model: Optional[str] = None) -> str:

backend/app/core/migrate.py

@@ -12,6 +12,7 @@ def migrate(db):
         "suggestions": "TEXT DEFAULT '[]'",
         "edited": "BOOLEAN DEFAULT 0",
         "regenerated": "INTEGER DEFAULT 0",
+        "reasoning_content": "TEXT DEFAULT ''",
         "updated_at": "DATETIME",
     }
     for name, ddl in additions.items():

backend/app/models/chat.py

@@ -39,6 +39,8 @@ class ChatMessage(Base):
     feedback: Mapped[str] = mapped_column(String(8), default="")
     # 推荐短语（JSON 数组，AI 回答后生成 1-3 条）
     suggestions: Mapped[str] = mapped_column(Text, default="[]")
+    # 思考内容（思考模型的 reasoning 流式输出）
+    reasoning_content: Mapped[str] = mapped_column(Text, default="")
     # 用户消息是否被编辑过
     edited: Mapped[bool] = mapped_column(Boolean, default=False)
     # 重新生成次数

backend/app/services/chat_service.py

@@ -220,9 +220,11 @@ async def chat_once(db: Session, user: User, session: ChatSession, content: str,
     db.commit()
 
     messages = build_messages(db, session, content, file_ids)
-    reply_text = await llm.chat_completion(messages, model=session.model, temperature=agent.temperature if agent else 0.7)
+    reply_text, reply_reasoning = await llm.chat_completion_full(
+        messages, model=session.model, temperature=agent.temperature if agent else 0.7)
 
-    reply = ChatMessage(session_id=session.id, role="assistant", content=reply_text, model=session.model)
+    reply = ChatMessage(session_id=session.id, role="assistant", content=reply_text,
+                        reasoning_content=reply_reasoning, model=session.model)
     db.add(reply)
     session.updated_at = datetime.utcnow()
     db.commit()
@@ -245,25 +247,32 @@ async def chat_stream(db: Session, user: User, session: ChatSession, content: st
     db.commit()
 
     messages = build_messages(db, session, content, file_ids)
-    reply = ChatMessage(session_id=session.id, role="assistant", content="", model=session.model)
+    reply = ChatMessage(session_id=session.id, role="assistant", content="", reasoning_content="", model=session.model)
     db.add(reply)
     db.commit()
     db.refresh(reply)
 
     parts: list[str] = []
+    reasoning_parts: list[str] = []
     try:
-        async for delta in llm.chat_completion_stream(
+        async for kind, delta in llm.chat_completion_stream(
             messages, model=session.model, temperature=agent.temperature if agent else 0.7
         ):
-            parts.append(delta)
-            yield delta, reply.id
+            if kind == "reasoning":
+                reasoning_parts.append(delta)
+                yield ("reasoning", delta), reply.id
+            else:
+                parts.append(delta)
+                yield ("content", delta), reply.id
     except Exception as e:
         reply.content = "".join(parts) or f"（调用失败：{e}）"
+        reply.reasoning_content = "".join(reasoning_parts)
         session.updated_at = datetime.utcnow()
         db.commit()
         raise
     else:
         reply.content = "".join(parts)
+        reply.reasoning_content = "".join(reasoning_parts)
         session.updated_at = datetime.utcnow()
         db.commit()
         _fire_background_jobs(db, session, content, reply.id, reply.content)
@@ -276,7 +285,7 @@ def _fire_background_jobs(db: Session, session: ChatSession, first_content: str,
 
 
 async def regenerate_stream(db: Session, session: ChatSession):
-    """重新生成最后一条 AI 回答（流式）。返回 (message_id, async_iter)。"""
+    """重新生成最后一条 AI 回答（流式）：先删除旧回答内容，再流式输出。"""
     last = (
         db.query(ChatMessage)
         .filter(ChatMessage.session_id == session.id, ChatMessage.role == "assistant")
@@ -286,27 +295,36 @@ async def regenerate_stream(db: Session, session: ChatSession):
     if not last:
         raise LookupError("没有可重新生成的消息")
 
-    agent = db.get(Agent, session.agent_id) if session.agent_id else None
-    messages = build_messages(db, session, "", exclude_last=True)
-
-    # 重新生成后旧建议清空，等新建议
+    # 先删除旧回答（内容清空，前端同步清空显示）
+    last.content = ""
+    last.reasoning_content = ""
     last.suggestions = "[]"
     db.commit()
 
+    agent = db.get(Agent, session.agent_id) if session.agent_id else None
+    messages = build_messages(db, session, "", exclude_last=True)
+
     parts: list[str] = []
+    reasoning_parts: list[str] = []
     try:
-        async for delta in llm.chat_completion_stream(
+        async for kind, delta in llm.chat_completion_stream(
             messages, model=session.model, temperature=agent.temperature if agent else 0.7
         ):
-            parts.append(delta)
-            yield delta, last.id
+            if kind == "reasoning":
+                reasoning_parts.append(delta)
+                yield ("reasoning", delta), last.id
+            else:
+                parts.append(delta)
+                yield ("content", delta), last.id
     except Exception as e:
         last.content = "".join(parts) or f"（调用失败：{e}）"
+        last.reasoning_content = "".join(reasoning_parts)
         last.regenerated += 1
         db.commit()
         raise
     else:
         last.content = "".join(parts)
+        last.reasoning_content = "".join(reasoning_parts)
         last.regenerated += 1
         session.updated_at = datetime.utcnow()
         db.commit()

frontend/src/views/ChatListView.vue

@@ -2,7 +2,7 @@
   <div>
     <van-nav-bar title="对话" fixed placeholder>
       <template #right>
-        <van-icon name="plus" size="22" @click="showNew = true" />
+        <van-icon name="plus" size="22" @click="createSession" />
       </template>
     </van-nav-bar>
 
@@ -23,26 +23,6 @@
         </van-cell>
       </van-cell-group>
     </van-pull-refresh>
-
-    <van-popup v-model:show="showNew" position="bottom" round safe-area-inset-bottom>
-      <div style="padding: 20px 16px 28px">
-        <h3 style="margin-bottom: 12px">新建会话</h3>
-        <van-field v-model="newTitle" label="标题" placeholder="新对话" />
-        <van-field label="智能体" readonly :model-value="selectedAgent?.name || '默认（通用助手）'" is-link @click="showAgentPicker = true" />
-        <div style="margin-top: 20px; display: flex; gap: 12px">
-          <van-button block plain type="primary" @click="showNew = false">取消</van-button>
-          <van-button block type="primary" :loading="creating" @click="createSession">创建</van-button>
-        </div>
-      </div>
-    </van-popup>
-
-    <van-popup v-model:show="showAgentPicker" position="bottom" round safe-area-inset-bottom>
-      <van-picker
-        :columns="agentColumns"
-        @confirm="onPickAgent"
-        @cancel="showAgentPicker = false"
-      />
-    </van-popup>
   </div>
 </template>
 
@@ -54,15 +34,10 @@ import client from '../api/client'
 
 const router = useRouter()
 const sessions = ref<any[]>([])
-const agents = ref<any[]>([])
 const loading = ref(false)
 const refreshing = ref(false)
-const showNew = ref(false)
-const showAgentPicker = ref(false)
 const creating = ref(false)
-const newTitle = ref('')
-const selectedAgent = ref<any>(null)
-const agentColumns = ref<any[]>([])
+let generalAgentId: number | null = null
 
 function timeText(iso: string) {
   return new Date(iso).toLocaleString('zh-CN', { month: '2-digit', day: '2-digit', hour: '2-digit', minute: '2-digit' })
@@ -72,6 +47,11 @@ async function load() {
   loading.value = true
   try {
     sessions.value = await client.get('/chat/sessions')
+    // 预取通用助手 id（新建会话直接用，不弹窗）
+    try {
+      const agents = await client.get('/agents')
+      generalAgentId = agents.find((a: any) => a.is_builtin && a.name === '通用助手')?.id ?? null
+    } catch {}
   } catch (e: any) {
     showToast(e.message)
   } finally {
@@ -80,31 +60,12 @@ async function load() {
   }
 }
 
-async function loadAgents() {
-  try {
-    agents.value = await client.get('/agents')
-    agentColumns.value = [
-      { text: '默认（通用助手）', value: null },
-      ...agents.value.map((a: any) => ({ text: `${a.avatar || '🤖'} ${a.name}`, value: a.id })),
-    ]
-  } catch {}
-}
-
-function onPickAgent({ selectedOptions }: any) {
-  const opt = selectedOptions[0]
-  selectedAgent.value = agents.value.find((a: any) => a.id === opt.value) || null
-  showAgentPicker.value = false
-}
-
 async function createSession() {
+  if (creating.value) return
   creating.value = true
   try {
-    const s = await client.post('/chat/sessions', {
-      title: newTitle.value || '新对话',
-      agent_id: selectedAgent.value?.id ?? null,
-    })
-    showNew.value = false
-    newTitle.value = ''
+    // 直接用通用助手，不弹窗
+    const s = await client.post('/chat/sessions', { title: '新对话', agent_id: generalAgentId })
     router.push(`/chat/${s.id}`)
   } catch (e: any) {
     showToast(e.message)
@@ -113,8 +74,5 @@ async function createSession() {
   }
 }
 
-onMounted(() => {
-  load()
-  loadAgents()
-})
+onMounted(load)
 </script>

frontend/src/views/ChatRoomView.vue

@@ -16,37 +16,47 @@
           <template v-if="m.role === 'user'">
             <div class="bubble">
               <div class="plain-text">{{ m.content }}</div>
-              <!-- 附件图片 -->
               <div v-if="m.file_ids?.length" class="attach-preview">
                 <img v-for="id in m.file_ids" :key="id" v-if="imgUrlFor(id)" :src="imgUrlFor(id)" class="attach-img" @click="previewImg(imgUrlFor(id))" />
               </div>
               <div v-if="m.edited" class="edited-tag">已编辑</div>
             </div>
-            <div v-if="ui.chat_ui?.user_edit || ui.chat_ui?.user_copy || ui.chat_ui?.user_time" class="msg-tools user-tools">
-              <van-icon v-if="ui.chat_ui?.user_edit" name="edit" @click="startEdit(m)" />
-              <van-icon v-if="ui.chat_ui?.user_copy" name="copy" @click="copyText(m.content)" />
-              <span v-if="ui.chat_ui?.user_time" class="msg-time">{{ fmtTime(m.created_at) }}</span>
+            <div v-if="uiOn('user_edit') || uiOn('user_copy') || uiOn('user_time')" class="msg-tools user-tools">
+              <van-icon v-if="uiOn('user_edit')" name="edit" @click="startEdit(m)" />
+              <van-icon v-if="uiOn('user_copy')" name="copy" @click="copyText(m.content)" />
+              <span v-if="uiOn('user_time')" class="msg-time">{{ fmtTime(m.created_at) }}</span>
             </div>
           </template>
 
           <!-- AI 回答 -->
           <template v-else>
             <div class="bubble">
-              <div class="markdown-body" v-html="renderMd(m.content)"></div>
-              <div v-if="m.id === streamingId" class="typing">
-                <van-loading size="14" /> 正在思考…
+              <!-- 思考过程：流式展开，输出完毕后折叠 -->
+              <div v-if="m.reasoning" class="reasoning-box" :class="{ collapsed: m.reasoningCollapsed }">
+                <div class="reasoning-header" @click="toggleReasoning(m)">
+                  <van-icon :name="m.reasoningCollapsed ? 'arrow-down' : 'arrow-up'" />
+                  <span>💭 思考过程</span>
+                  <span class="reasoning-hint">{{ m.reasoningCollapsed ? '点击展开' : '点击收起' }}</span>
+                </div>
+                <div v-if="!m.reasoningCollapsed" class="reasoning-content">{{ m.reasoning }}</div>
               </div>
+
+              <div class="markdown-body" v-html="renderMd(m.content)"></div>
+              <div v-if="m.id === streamingId && !m.content" class="typing">
+                <van-loading size="14" /> {{ m.reasoning ? '思考中…' : '正在思考…' }}
+              </div>
+
               <!-- 推荐短语 -->
-              <div v-if="ui.chat_ui?.ai_suggestions && m.suggestions?.length && m.id !== streamingId" class="sugg-chips">
-                <div v-for="s in m.suggestions" :key="s" class="sugg-chip" @click="quickSend(s)">{{ s }}</div>
+              <div v-if="uiOn('ai_suggestions') && m.suggestions?.length && m.id !== streamingId" class="sugg-chips">
+                <div v-for="s in m.suggestions" :key="s" class="sugg-chip" @click="useSuggestion(s)">{{ s }}</div>
               </div>
             </div>
             <div v-if="hasAiTools" class="msg-tools ai-tools">
-              <van-icon v-if="ui.chat_ui?.ai_copy" name="copy" @click="copyText(m.content)" />
-              <van-icon v-if="ui.chat_ui?.ai_regenerate" name="replay" @click="regenerate()" />
-              <van-icon v-if="ui.chat_ui?.ai_feedback" name="good-job" :class="{ active: m.feedback === 'like' }" @click="feedback(m, 'like')" />
-              <van-icon v-if="ui.chat_ui?.ai_feedback" name="bad-job" :class="{ active: m.feedback === 'dislike' }" @click="feedback(m, 'dislike')" />
-              <van-icon v-if="ui.chat_ui?.ai_voice" :name="speakingId === m.id ? 'volume-o' : 'volume'" @click="speak(m)" />
+              <van-icon v-if="uiOn('ai_copy')" name="copy" @click="copyText(m.content)" />
+              <van-icon v-if="uiOn('ai_regenerate')" name="replay" @click="regenerate()" />
+              <van-icon v-if="uiOn('ai_feedback')" name="good-job" :class="{ active: m.feedback === 'like' }" @click="feedback(m, 'like')" />
+              <van-icon v-if="uiOn('ai_feedback')" name="bad-job" :class="{ active: m.feedback === 'dislike' }" @click="feedback(m, 'dislike')" />
+              <van-icon v-if="uiOn('ai_voice')" :name="speakingId === m.id ? 'volume-o' : 'volume'" @click="speak(m)" />
             </div>
           </template>
         </div>
@@ -56,9 +66,9 @@
     <!-- 输入区 -->
     <div class="input-bar">
       <div class="input-row">
-        <van-icon v-if="ui.chat_ui?.voice_input" name="audio" size="22" class="input-icon" @click="startVoice" />
-        <van-icon v-if="ui.chat_ui?.attachment" name="plus" size="22" class="input-icon" @click="showAttachSheet = true" />
+        <van-icon v-if="uiOn('attachment')" name="plus" size="22" class="input-icon" @click="showAttachSheet = true" />
         <van-field
+          ref="inputRef"
           v-model="input"
           type="textarea"
           autosize
@@ -67,10 +77,26 @@
           :placeholder="editingMsg ? '编辑消息…（保存后自动重新生成回答）' : '输入消息…'"
           @keydown.enter.exact.prevent="send"
         />
+        <!-- 语音：长按说话，松开结束并自动提交 -->
+        <div
+          v-if="uiOn('voice_input')"
+          class="voice-btn"
+          :class="{ holding: voiceHolding }"
+          @touchstart.prevent="startPress"
+          @touchend="endPress"
+          @touchcancel="endPress"
+          @mousedown.prevent="startPress"
+          @mouseup="endPress"
+          @mouseleave="endPress"
+        >
+          <van-icon name="audio" size="22" />
+        </div>
         <van-button type="primary" :disabled="sending || (!input.trim() && attachments.length === 0)" @click="send">
           {{ editingMsg ? '更新' : '发送' }}
         </van-button>
       </div>
+      <!-- 录音提示 -->
+      <div v-if="voiceHolding" class="voice-tip">🎙️ 正在录音，松开结束并发送</div>
       <!-- 已选附件 -->
       <div v-if="attachments.length" class="attach-list">
         <div v-for="(a, i) in attachments" :key="i" class="attach-item">
@@ -84,9 +110,6 @@
 
     <!-- 附件类型选择 -->
     <van-action-sheet v-model:show="showAttachSheet" :actions="attachActions" cancel-text="取消" @select="onAttachSelect" />
-
-    <!-- 图片预览 -->
-    <van-image-preview v-model:show="showPreview" :images="previewList" :start-position="previewIndex" />
   </div>
 </template>
 
@@ -116,14 +139,33 @@ const input = ref('')
 const sending = ref(false)
 const streamingId = ref<number | null>(null)
 const scrollRef = ref<HTMLElement>()
+const inputRef = ref<any>()
 const ui = ref<any>({ chat_ui: {} })
 const attachments = ref<any[]>([])
 const showAttachSheet = ref(false)
 const editingMsg = ref<any>(null)
 const speakingId = ref<number | null>(null)
-const showPreview = ref(false)
-const previewList = ref<string[]>([])
-const previewIndex = ref(0)
+const voiceHolding = ref(false)
+
+let ws: WebSocket | null = null
+let titleTimer: number | null = null
+let suggTimer: number | null = null
+let recorder: any = null
+
+// UI 开关：配置缺失时默认开启（兜底，保证按钮始终可用）
+function uiOn(key: string): boolean {
+  return (ui.value.chat_ui?.[key] ?? true) !== false
+}
+
+const hasAiTools = computed(() => uiOn('ai_copy') || uiOn('ai_regenerate') || uiOn('ai_feedback') || uiOn('ai_voice'))
+
+const attachActions = computed(() => {
+  const types = ui.value.chat_ui?.attachment_types || ['image', 'text']
+  const actions: any[] = []
+  if (types.includes('image')) actions.push({ name: '上传图片', subname: 'JPG/PNG/GIF/WebP', value: 'image' })
+  if (types.includes('text')) actions.push({ name: '上传文本文件', subname: 'TXT/MD/CSV/JSON/代码文件', value: 'text' })
+  return actions
+})
 
 // 附件图片缓存：file_id -> objectURL
 const imgCache = new Map<number, string>()
@@ -150,23 +192,6 @@ async function scanImages() {
   }
 }
 
-let ws: WebSocket | null = null
-let titleTimer: number | null = null
-let suggTimer: number | null = null
-
-const hasAiTools = computed(() => {
-  const c = ui.value.chat_ui || {}
-  return c.ai_copy || c.ai_regenerate || c.ai_feedback || c.ai_voice
-})
-
-const attachActions = computed(() => {
-  const types = ui.value.chat_ui?.attachment_types || ['image', 'text']
-  const actions: any[] = []
-  if (types.includes('image')) actions.push({ name: '上传图片', subname: 'JPG/PNG/GIF/WebP', value: 'image' })
-  if (types.includes('text')) actions.push({ name: '上传文本文件', subname: 'TXT/MD/CSV/JSON/代码文件', value: 'text' })
-  return actions
-})
-
 function renderMd(text: string) {
   return marked.parse(text || '')
 }
@@ -195,6 +220,7 @@ async function load() {
     const sessions = await client.get('/chat/sessions')
     session.value = sessions.find((s: any) => s.id === sessionId.value) || null
     messages.value = await client.get(`/chat/sessions/${sessionId.value}/messages`)
+    scanImages()
     scrollBottom()
   } catch (e: any) {
     showToast(e.message)
@@ -208,7 +234,6 @@ function send() {
   if (!content && attachments.value.length === 0) return
 
   if (editingMsg.value) {
-    // 编辑：更新消息内容，然后重新生成
     saveEdit(content)
     return
   }
@@ -221,24 +246,13 @@ function send() {
 
   const tempId = -Date.now()
   messages.value.push({ id: tempId, role: 'user', content, file_ids: fileIds })
-  const aiMsg = { id: -Date.now() - 1, role: 'assistant', content: '', suggestions: [] }
+  const aiMsg = { id: -Date.now() - 1, role: 'assistant', content: '', reasoning: '', suggestions: [], reasoningCollapsed: false }
   messages.value.push(aiMsg)
   streamingId.value = aiMsg.id
   scrollBottom()
 
-  if (!ws || ws.readyState !== WebSocket.OPEN) {
-    ws = new WebSocket(wsUrl(sessionId.value))
-    ws.onmessage = (ev) => handleWs(ev, aiMsg)
-    ws.onopen = () => ws!.send(JSON.stringify({ type: 'chat', content, file_ids: fileIds }))
-    ws.onerror = () => {
-      showToast('连接失败，请重试')
-      sending.value = false
-      streamingId.value = null
-    }
-    ws.onclose = () => { ws = null }
-  } else {
-    ws.send(JSON.stringify({ type: 'chat', content, file_ids: fileIds }))
-  }
+  const doSend = () => ws!.send(JSON.stringify({ type: 'chat', content, file_ids: fileIds }))
+  ensureWs(doSend, aiMsg)
 }
 
 async function saveEdit(content: string) {
@@ -246,10 +260,9 @@ async function saveEdit(content: string) {
   sending.value = true
   try {
     await client.put(`/chat/messages/${target.id}`, { content })
-    // 本地更新
     const msg = messages.value.find((m) => m.id === target.id)
     if (msg) { msg.content = content; msg.edited = true }
-    // 触发重新生成
+    // 保存后自动重新生成（先删旧答再流式）
     regenerate()
   } catch (e: any) {
     showToast(e.message)
@@ -257,30 +270,54 @@ async function saveEdit(content: string) {
   }
 }
 
+// ---------- 重新生成：先删除原回答，再流式输出 ----------
 function regenerate() {
-  const aiMsg = { id: -Date.now() - 2, role: 'assistant', content: '', suggestions: [] }
-  messages.value.push(aiMsg)
-  streamingId.value = aiMsg.id
+  if (sending.value) return
+  // 找到最后一条真实 AI 回答
+  const last = [...messages.value].reverse().find((m) => m.role === 'assistant' && m.id > 0)
+  if (!last) return showToast('没有可重新生成的消息')
+  // 删除原回答：清空内容，从空白开始流式
+  last.content = ''
+  last.reasoning = ''
+  last.reasoningCollapsed = false
+  last.suggestions = []
+  streamingId.value = last.id
   sending.value = true
   editingMsg.value = null
   scrollBottom()
 
   const doSend = () => ws!.send(JSON.stringify({ type: 'regenerate' }))
+  ensureWs(doSend, last)
+}
+
+function ensureWs(onOpen: () => void, streamMsg: any) {
   if (!ws || ws.readyState !== WebSocket.OPEN) {
     ws = new WebSocket(wsUrl(sessionId.value))
-    ws.onmessage = (ev) => handleWs(ev, aiMsg)
-    ws.onopen = doSend
-    ws.onerror = () => { showToast('连接失败'); sending.value = false; streamingId.value = null }
+    ws.onmessage = (ev) => handleWs(ev)
+    ws.onopen = onOpen
+    ws.onerror = () => {
+      showToast('连接失败，请重试')
+      sending.value = false
+      streamingId.value = null
+    }
     ws.onclose = () => { ws = null }
   } else {
-    doSend()
+    onOpen()
   }
 }
 
-function handleWs(ev: MessageEvent, aiMsg: any) {
+function handleWs(ev: MessageEvent) {
   const data = JSON.parse(ev.data)
   if (data.type === 'delta') {
-    aiMsg.content += data.content
+    const msg = messages.value.find((m) => m.id === data.message_id)
+    if (!msg) return
+    if (data.kind === 'reasoning') {
+      msg.reasoning = (msg.reasoning || '') + data.content
+    } else {
+      // 思考内容输出完毕后自动折叠，再展示正式回答
+      if (!msg.content && msg.reasoning && !msg.reasoningCollapsed) msg.reasoningCollapsed = true
+      msg.content += data.content
+    }
     scrollBottom()
   } else if (data.type === 'done') {
     streamingId.value = null
@@ -293,7 +330,8 @@ function handleWs(ev: MessageEvent, aiMsg: any) {
   } else if (data.type === 'title') {
     if (session.value) session.value.title = data.title
   } else if (data.type === 'error') {
-    aiMsg.content = `⚠️ ${data.message}`
+    const msg = messages.value.find((m) => m.id === data.message_id) || messages.value[messages.value.length - 1]
+    if (msg && msg.role === 'assistant') msg.content = `⚠️ ${data.message}`
     streamingId.value = null
     sending.value = false
     scrollBottom()
@@ -303,6 +341,9 @@ function handleWs(ev: MessageEvent, aiMsg: any) {
 async function refreshMessages() {
   try {
     messages.value = await client.get(`/chat/sessions/${sessionId.value}/messages`)
+    messages.value.forEach((m: any) => {
+      if (m.reasoning && !m.content && !m.reasoningCollapsed) m.reasoningCollapsed = false
+    })
     const sessions = await client.get('/chat/sessions')
     session.value = sessions.find((s: any) => s.id === sessionId.value) || session.value
     scanImages()
@@ -311,7 +352,6 @@ async function refreshMessages() {
 }
 
 function scheduleSuggestRefresh() {
-  // 建议异步生成，稍后刷新一次
   if (suggTimer) window.clearTimeout(suggTimer)
   suggTimer = window.setTimeout(refreshMessages, 4000)
   titleTimer = window.setTimeout(async () => {
@@ -326,6 +366,7 @@ function scheduleSuggestRefresh() {
 function startEdit(m: any) {
   editingMsg.value = m
   input.value = m.content
+  inputRef.value?.focus?.()
   scrollBottom()
 }
 
@@ -346,6 +387,17 @@ async function feedback(m: any, kind: string) {
   }
 }
 
+function toggleReasoning(m: any) {
+  m.reasoningCollapsed = !m.reasoningCollapsed
+}
+
+// 推荐短语：点击填入输入框（不直接发送）
+function useSuggestion(s: string) {
+  input.value = s
+  inputRef.value?.focus?.()
+  scrollBottom()
+}
+
 // 语音播放（浏览器 TTS）
 function speak(m: any) {
   if (!('speechSynthesis' in window)) return showToast('当前浏览器不支持语音播放')
@@ -365,23 +417,54 @@ function speak(m: any) {
   window.speechSynthesis.speak(u)
 }
 
-// 语音输入
-function startVoice() {
+// ---------- 语音输入：长按说话，松开结束并自动提交 ----------
+function startPress() {
+  if (voiceHolding.value) return
   const SR: any = (window as any).SpeechRecognition || (window as any).webkitSpeechRecognition
   if (!SR) return showToast('当前浏览器不支持语音输入（建议用 Chrome/Edge）')
+  voiceHolding.value = true
   const rec = new SR()
   rec.lang = 'zh-CN'
   rec.interimResults = true
-  rec.continuous = true
+  rec.continuous = false
+  let finalText = ''
   rec.onresult = (e: any) => {
-    let text = ''
-    for (let i = 0; i < e.results.length; i++) text += e.results[i][0].transcript
-    input.value = text
+    let interim = ''
+    for (let i = 0; i < e.results.length; i++) {
+      if (e.results[i].isFinal) finalText += e.results[i][0].transcript
+      else interim += e.results[i][0].transcript
+    }
+    input.value = (finalText + interim).trim()
+  }
+  rec.onend = () => {
+    voiceHolding.value = false
+    const text = input.value.trim()
+    if (text) {
+      showToast('识别完成，正在发送')
+      send()
+    } else {
+      showToast('未识别到语音')
+    }
+  }
+  rec.onerror = (e: any) => {
+    voiceHolding.value = false
+    if (e.error !== 'aborted') showToast(`语音识别失败：${e.error || ''}`)
+  }
+  recorder = rec
+  try {
+    rec.start()
+  } catch {
+    voiceHolding.value = false
+  }
+}
+
+function endPress() {
+  if (recorder) {
+    try { recorder.stop() } catch {}
+    recorder = null
+  } else {
+    voiceHolding.value = false
   }
-  rec.onend = () => showToast('语音输入结束')
-  rec.onerror = (e: any) => showToast(`语音识别失败：${e.error || ''}`)
-  showToast('请说话…')
-  rec.start()
 }
 
 // ---------- 附件 ----------
@@ -396,7 +479,6 @@ function onAttachSelect(action: any) {
     if (!file) return
     if (file.size > 20 * 1024 * 1024) return showToast('文件不能超过 20MB')
     const preview = isImage ? URL.createObjectURL(file) : ''
-    // 先展示，再上传
     const item = { name: file.name, type: isImage ? 'image' : 'text', preview, fileId: 0, uploading: true }
     attachments.value.push(item)
     try {
@@ -413,10 +495,6 @@ function onAttachSelect(action: any) {
   fileInput.click()
 }
 
-function imageUrls(m: any) {
-  return (m.file_ids || []).map((id: number) => ({ id, url: '' })).filter(() => false)
-}
-
 async function previewImg(url: string) {
   showImagePreview([url])
 }
@@ -458,6 +536,13 @@ async function delSession() {
 .msg-tools .van-icon.active { color: #1989fa; }
 .msg-time { font-size: 11px; color: #c8c9cc; }
 .typing { display: flex; align-items: center; gap: 6px; color: #969799; font-size: 12px; margin-top: 4px; }
+/* 思考过程 */
+.reasoning-box { background: #f7f8fa; border-radius: 8px; margin-bottom: 8px; font-size: 12.5px; }
+.reasoning-box.collapsed .reasoning-content { display: none; }
+.reasoning-header { display: flex; align-items: center; gap: 6px; padding: 6px 10px; color: #969799; cursor: pointer; user-select: none; }
+.reasoning-hint { margin-left: auto; font-size: 11px; color: #c8c9cc; }
+.reasoning-content { padding: 0 10px 8px; color: #646566; white-space: pre-wrap; max-height: 200px; overflow-y: auto; }
+/* 推荐短语 */
 .sugg-chips { display: flex; flex-wrap: wrap; gap: 6px; margin-top: 8px; }
 .sugg-chip { background: #e8f3ff; color: #1989fa; font-size: 12px; padding: 4px 10px; border-radius: 14px; cursor: pointer; }
 .attach-preview { display: flex; flex-wrap: wrap; gap: 6px; margin-top: 6px; }
@@ -466,6 +551,10 @@ async function delSession() {
 .input-row { display: flex; align-items: flex-end; gap: 8px; max-width: 860px; margin: 0 auto; }
 .input-icon { color: #646566; flex-shrink: 0; margin-bottom: 8px; }
 .input-row .van-field { flex: 1; background: #f7f8fa; border-radius: 8px; padding: 4px 10px; }
+/* 语音按钮：长按说话 */
+.voice-btn { flex-shrink: 0; margin-bottom: 8px; color: #646566; border-radius: 50%; padding: 4px; display: flex; align-items: center; justify-content: center; transition: all .15s; }
+.voice-btn.holding { color: #fff; background: #ee0a24; transform: scale(1.15); box-shadow: 0 2px 12px rgba(238, 10, 36, .4); }
+.voice-tip { text-align: center; font-size: 12px; color: #ee0a24; padding-top: 4px; }
 .attach-list { display: flex; flex-wrap: wrap; gap: 8px; padding: 8px 4px 0; max-width: 860px; margin: 0 auto; }
 .attach-item { display: flex; align-items: center; gap: 6px; background: #f7f8fa; border-radius: 8px; padding: 4px 8px; font-size: 12px; position: relative; }
 .attach-thumb { width: 32px; height: 32px; object-fit: cover; border-radius: 6px; }