Enhance AssistantPage with subtle scrollbar styling and improve voice preview handling

- Introduce a new utility class `.scrollbar-subtle` in `globals.css` for a refined scrollbar appearance.
- Update `AssistantPage` to apply the subtle scrollbar styling to overflow containers, enhancing the user interface.
- Reset message state and sequence in `useVoicePreview` hook upon resource release, ensuring clean state management.

frontend/src/app/globals.css

@@ -232,3 +232,43 @@
     text-transform: uppercase;
   }
 }
+
+@layer utilities {
+  .scrollbar-subtle {
+    scrollbar-width: thin;
+    scrollbar-color: color-mix(
+        in srgb,
+        var(--muted-soft) 48%,
+        transparent
+      )
+      transparent;
+  }
+
+  .scrollbar-subtle::-webkit-scrollbar {
+    width: 10px;
+    height: 10px;
+  }
+
+  .scrollbar-subtle::-webkit-scrollbar-track {
+    background: transparent;
+  }
+
+  .scrollbar-subtle::-webkit-scrollbar-thumb {
+    background-color: color-mix(
+      in srgb,
+      var(--muted-soft) 42%,
+      transparent
+    );
+    border: 3px solid transparent;
+    border-radius: 999px;
+    background-clip: content-box;
+  }
+
+  .scrollbar-subtle::-webkit-scrollbar-thumb:hover {
+    background-color: color-mix(
+      in srgb,
+      var(--muted-soft) 64%,
+      transparent
+    );
+  }
+}

frontend/src/components/pages/AssistantPage.tsx

@@ -33,7 +33,6 @@ import {
   Waves,
   Bug,
   Video,
-  SlidersHorizontal,
 } from "lucide-react";
 
 import { Button } from "@/components/ui/button";
@@ -1321,7 +1320,7 @@ export function AssistantPage(props: AssistantPageProps) {
         </div>
 
         <div className="flex min-h-0 flex-1 gap-6">
-          <div className="min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
+          <div className="scrollbar-subtle min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
             <SectionCard
               icon={<Boxes size={18} />}
               title="Dify 应用配置"
@@ -1420,7 +1419,7 @@ export function AssistantPage(props: AssistantPageProps) {
         </div>
 
         <div className="flex min-h-0 flex-1 gap-6">
-          <div className="min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
+          <div className="scrollbar-subtle min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
             <SectionCard
               icon={<Database size={18} />}
               title="FastGPT 应用配置"
@@ -1534,7 +1533,7 @@ export function AssistantPage(props: AssistantPageProps) {
         </div>
 
         <div className="flex min-h-0 flex-1 gap-6">
-          <div className="min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
+          <div className="scrollbar-subtle min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
             <SectionCard
               icon={<Terminal size={18} />}
               title="OpenCode 服务配置"
@@ -1652,7 +1651,7 @@ export function AssistantPage(props: AssistantPageProps) {
       </div>
 
       <div className="flex min-h-0 flex-1 gap-6">
-        <div className="min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
+        <div className="scrollbar-subtle min-w-0 flex-1 space-y-5 overflow-y-auto pr-1">
           <SectionCard>
             <div className="grid grid-cols-1 gap-4 md:grid-cols-2">
               <div
@@ -1727,8 +1726,9 @@ export function AssistantPage(props: AssistantPageProps) {
             </div>
             <div className="mt-4 border-t border-hairline pt-4">
               <ToggleRow
+                icon={<Eye size={16} />}
                 title="视觉理解"
-                description="开启后，右侧调试面板可切换到视频流预览，并可同时选择麦克风与摄像头"
+                hint="开启后，右侧调试面板可切换到视频流预览，并可同时选择麦克风与摄像头。"
                 checked={visionEnabled}
                 onChange={setVisionEnabled}
               />
@@ -1841,6 +1841,7 @@ type VizStyle = "aura" | "nebula" | "bars" | "wave";
 
 // 调试面板顶部主视图:聊天记录 / 视频流(仅视觉理解开启时可选)
 type DebugView = "chat" | "video";
+type DebugInputMode = "mic" | "text";
 
 const VIZ_OPTIONS: { style: VizStyle; label: string; icon: React.ReactNode }[] =
   [
@@ -1917,38 +1918,29 @@ function DebugDrawer({
   const camera = useCameraPreview();
   const [showTranscript, setShowTranscript] = useState(false);
   const [vizStyle, setVizStyle] = useState<VizStyle>("aura");
-  const [view, setView] = useState<DebugView>(vision ? "video" : "chat");
+  const [view, setView] = useState<DebugView>("chat");
   const effectiveView: DebugView = vision ? view : "chat";
   const stopCamera = camera.stop;
+  const startCamera = camera.start;
+  const cameraShouldRun =
+    vision &&
+    (preview.status === "connecting" || preview.status === "connected");
 
   useEffect(() => {
-    // eslint-disable-next-line react-hooks/set-state-in-effect
+    if (!vision) {
-    setView(vision ? "video" : "chat");
+      // eslint-disable-next-line react-hooks/set-state-in-effect
-    if (!vision) stopCamera();
+      setView("chat");
+      stopCamera();
+    }
   }, [vision, stopCamera]);
 
-  const { start: startCamera, active: cameraActive, starting: cameraStarting } =
-    camera;
   useEffect(() => {
-    if (
+    if (cameraShouldRun) {
-      vision &&
-      effectiveView === "video" &&
-      !cameraActive &&
-      !cameraStarting
-    ) {
       void startCamera();
+    } else {
+      stopCamera();
     }
-  }, [
+  }, [cameraShouldRun, camera.deviceId, startCamera, stopCamera]);
-    vision,
-    effectiveView,
-    cameraActive,
-    cameraStarting,
-    startCamera,
-  ]);
-
-  useEffect(() => {
-    if (effectiveView !== "video") stopCamera();
-  }, [effectiveView, stopCamera]);
 
   const containerClass = asSheet
     ? "flex h-full min-w-0 flex-1 flex-col overflow-hidden"
@@ -2001,7 +1993,6 @@ function DebugDrawer({
               </SegmentedIconGroup>
             </>
           )}
-          <DeviceMenu preview={preview} camera={camera} vision={vision} />
           <SegmentedIconGroup label="预览视图">
             <SegmentedIconButton
               selected={effectiveView === "chat"}
@@ -2022,7 +2013,13 @@ function DebugDrawer({
           </SegmentedIconGroup>
         </div>
       </div>
-
+      {vision && (
+        <div className="shrink-0 border-b border-hairline px-5 py-2.5">
+          <div className="flex h-10 min-w-0 items-center rounded-[1.4rem] border border-hairline-strong bg-background px-2">
+            <CameraDeviceField camera={camera} />
+          </div>
+        </div>
+      )}
       <DebugVoicePanel
         view={effectiveView}
         showTranscript={showTranscript}
@@ -2030,75 +2027,15 @@ function DebugDrawer({
         assistantId={assistantId}
         preview={preview}
         camera={camera}
+        vision={vision}
       />
     </aside>
   );
 }
 
-function DeviceMenu({
+function DeviceSelectField({
-  preview,
-  camera,
-  vision,
-}: {
-  preview: VoicePreview;
-  camera: CameraPreview;
-  vision: boolean;
-}) {
-  return (
-    <Popover>
-      <PopoverTrigger asChild>
-        <button
-          type="button"
-          aria-label="设备设置"
-          title="设备设置"
-          className="flex h-8 items-center gap-1.5 rounded-full border border-hairline bg-canvas-soft px-2.5 text-xs text-muted-soft transition-colors hover:text-foreground"
-        >
-          <SlidersHorizontal size={14} />
-          <span className="hidden sm:inline">设备</span>
-        </button>
-      </PopoverTrigger>
-      <PopoverContent align="end" className="w-72 p-0">
-        <div className="border-b border-hairline px-3 py-2.5">
-          <div className="text-xs font-medium text-foreground">输入设备</div>
-          <p className="mt-0.5 text-[11px] leading-4 text-muted-foreground">
-            {vision
-              ? "选择麦克风与摄像头,分别用于语音与视觉预览。"
-              : "选择麦克风用于语音对话。"}
-          </p>
-        </div>
-        <div className="space-y-3 p-3">
-          <DeviceSelectRow
-            icon={<Mic size={13} className="shrink-0 text-muted-soft" />}
-            label="麦克风"
-            placeholder="默认麦克风"
-            fallbackLabel="麦克风"
-            value={preview.selectedDeviceId}
-            devices={preview.audioInputs}
-            onSelect={(id) => preview.selectDevice(id)}
-          />
-          {vision && (
-            <>
-              <div className="border-t border-hairline" />
-              <DeviceSelectRow
-                icon={<Video size={13} className="shrink-0 text-muted-soft" />}
-                label="摄像头"
-                placeholder="默认摄像头"
-                fallbackLabel="摄像头"
-                value={camera.deviceId}
-                devices={camera.devices}
-                onSelect={(id) => void camera.selectCamera(id)}
-              />
-            </>
-          )}
-        </div>
-      </PopoverContent>
-    </Popover>
-  );
-}
-
-function DeviceSelectRow({
   icon,
-  label,
+  ariaLabel,
   placeholder,
   fallbackLabel,
   value,
@@ -2106,7 +2043,7 @@ function DeviceSelectRow({
   onSelect,
 }: {
   icon: React.ReactNode;
-  label: string;
+  ariaLabel: string;
   placeholder: string;
   fallbackLabel: string;
   value: string;
@@ -2114,30 +2051,96 @@ function DeviceSelectRow({
   onSelect: (deviceId: string) => void;
 }) {
   return (
-    <div className="space-y-1.5">
+    <Select
-      <div className="text-xs font-medium text-muted-foreground">{label}</div>
+      value={value || "default"}
-      <Select
+      onValueChange={(nextValue) =>
-        value={value || "default"}
+        onSelect(nextValue === "default" ? "" : nextValue)
-        onValueChange={(v) => onSelect(v === "default" ? "" : v)}
+      }
+    >
+      <SelectTrigger
+        size="sm"
+        className="h-10 min-w-0 flex-1 justify-between rounded-full border-transparent bg-transparent px-3 text-sm text-foreground shadow-none hover:bg-surface-strong focus-visible:ring-0 data-[size=sm]:h-10"
+        aria-label={ariaLabel}
       >
-        <SelectTrigger
+        <span className="flex min-w-0 items-center gap-2">
-          size="sm"
-          className="w-full gap-2 rounded-full border-hairline bg-canvas-soft text-xs text-muted-foreground"
-          aria-label={`选择${label}`}
-        >
           {icon}
-          <SelectValue placeholder={placeholder} />
+          <span className="min-w-0 truncate text-left">
-        </SelectTrigger>
+            <SelectValue placeholder={placeholder} />
-        <SelectContent>
+          </span>
-          <SelectItem value="default">{placeholder}</SelectItem>
+        </span>
-          {devices.map((device, index) => (
+      </SelectTrigger>
-            <SelectItem key={device.deviceId} value={device.deviceId}>
+      <SelectContent
-              {device.label || `${fallbackLabel} ${index + 1}`}
+        position="popper"
-            </SelectItem>
+        side="top"
-          ))}
+        align="start"
-        </SelectContent>
+        className="w-[280px]"
-      </Select>
+      >
-    </div>
+        <SelectItem value="default">{placeholder}</SelectItem>
+        {devices.map((device, index) => (
+          <SelectItem key={device.deviceId} value={device.deviceId}>
+            {device.label || `${fallbackLabel} ${index + 1}`}
+          </SelectItem>
+        ))}
+      </SelectContent>
+    </Select>
+  );
+}
+
+function MicrophoneDeviceField({ preview }: { preview: VoicePreview }) {
+  return (
+    <DeviceSelectField
+      icon={<Mic size={15} className="shrink-0 text-muted-soft" />}
+      ariaLabel="选择麦克风"
+      placeholder="默认麦克风"
+      fallbackLabel="麦克风"
+      value={preview.selectedDeviceId}
+      devices={preview.audioInputs}
+      onSelect={(deviceId) => preview.selectDevice(deviceId)}
+    />
+  );
+}
+
+function CameraDeviceField({ camera }: { camera: CameraPreview }) {
+  return (
+    <DeviceSelectField
+      icon={<Video size={15} className="shrink-0 text-muted-soft" />}
+      ariaLabel="选择摄像头"
+      placeholder="默认摄像头"
+      fallbackLabel="摄像头"
+      value={camera.deviceId}
+      devices={camera.devices}
+      onSelect={(deviceId) => void camera.selectCamera(deviceId)}
+    />
+  );
+}
+
+function DebugInputModeButton({
+  selected,
+  label,
+  onClick,
+  children,
+}: {
+  selected: boolean;
+  label: string;
+  onClick: () => void;
+  children: React.ReactNode;
+}) {
+  return (
+    <button
+      type="button"
+      aria-label={label}
+      aria-pressed={selected}
+      title={label}
+      onClick={onClick}
+      className={[
+        "flex h-8 w-8 shrink-0 items-center justify-center rounded-full transition-colors",
+        selected
+          ? "bg-surface-strong text-foreground"
+          : "text-muted-soft hover:bg-surface-strong hover:text-foreground",
+      ].join(" ")}
+    >
+      {children}
+    </button>
   );
 }
 
@@ -2148,6 +2151,7 @@ function DebugVoicePanel({
   assistantId,
   preview,
   camera,
+  vision,
 }: {
   view: DebugView;
   showTranscript: boolean;
@@ -2155,6 +2159,7 @@ function DebugVoicePanel({
   assistantId: string | null;
   preview: VoicePreview;
   camera: CameraPreview;
+  vision: boolean;
 }) {
   const {
     status,
@@ -2170,11 +2175,13 @@ function DebugVoicePanel({
   } = preview;
   const recording = status === "connecting" || status === "connected";
   const [textDraft, setTextDraft] = useState("");
+  const [inputMode, setInputMode] = useState<DebugInputMode>("mic");
   const inChatView = view === "chat" && (!SHOW_VOICE_VIZ || showTranscript);
+  const idleOrFailed = status === "idle" || status === "failed";
   const showIdleHub =
-    inChatView &&
     messages.length === 0 &&
-    (status === "idle" || status === "failed");
+    idleOrFailed &&
+    (inChatView || (vision && view === "video"));
 
   function handleSendText() {
     if (sendText(textDraft)) {
@@ -2186,7 +2193,14 @@ function DebugVoicePanel({
     <div className="flex min-h-0 flex-1 flex-col">
       {/* 后端 TTS 音频经 WebRTC 媒体流过来,挂这里播放 */}
       <audio ref={audioRef} autoPlay playsInline className="hidden" />
-      {view === "video" ? (
+      {view === "video" && showIdleHub ? (
+        <DebugIdleHub
+          status={status}
+          error={error}
+          assistantId={assistantId}
+          connect={connect}
+        />
+      ) : view === "video" ? (
         <DebugVideoPanel camera={camera} />
       ) : !SHOW_VOICE_VIZ || showTranscript ? (
         showIdleHub ? (
@@ -2200,7 +2214,7 @@ function DebugVoicePanel({
           <DebugTranscriptPanel messages={messages} recording={recording} />
         )
       ) : (
-        <div className="relative flex min-h-0 flex-1 flex-col items-center justify-center gap-3 overflow-y-auto px-6 py-3 text-center">
+        <div className="scrollbar-subtle relative flex min-h-0 flex-1 flex-col items-center justify-center gap-3 overflow-y-auto px-6 py-3 text-center">
           <div
             className="pointer-events-none absolute left-1/2 top-2 h-72 w-72 -translate-x-1/2 rounded-full opacity-50 blur-3xl"
             style={{
@@ -2293,34 +2307,63 @@ function DebugVoicePanel({
       )}
 
       <div className="shrink-0 border-t border-hairline bg-card p-3">
-        <div className="flex items-end gap-2">
+        <div className="flex items-center gap-2">
-          <Textarea
+          <div
-            rows={1}
+            className={[
-            value={textDraft}
+              "flex h-10 min-w-0 items-center gap-1 rounded-[1.4rem] border border-hairline-strong bg-background px-2",
-            disabled={status !== "connected"}
+              "flex-1",
-            onChange={(event) => setTextDraft(event.target.value)}
+            ].join(" ")}
-            onKeyDown={(event) => {
-              if (event.key === "Enter" && !event.shiftKey && !event.nativeEvent.isComposing) {
-                event.preventDefault();
-                handleSendText();
-              }
-            }}
-            placeholder={
-              status === "connected"
-                ? "输入文字发送给助手，将打断当前播报…"
-                : "开始对话后可输入文字…"
-            }
-            className="max-h-24 min-h-10 flex-1 resize-none border-hairline-strong bg-background text-sm text-foreground placeholder:text-muted-soft"
-          />
-          <Button
-            size="icon"
-            className="shrink-0"
-            aria-label="发送调试消息"
-            disabled={status !== "connected" || !textDraft.trim()}
-            onClick={handleSendText}
           >
-            <Send size={16} />
+            <div className="flex shrink-0 items-center gap-0.5 rounded-full bg-canvas-soft p-0.5">
-          </Button>
+              <DebugInputModeButton
+                selected={inputMode === "mic"}
+                label="选择麦克风设备"
+                onClick={() => setInputMode("mic")}
+              >
+                <Mic size={15} />
+              </DebugInputModeButton>
+              <DebugInputModeButton
+                selected={inputMode === "text"}
+                label="文字输入"
+                onClick={() => setInputMode("text")}
+              >
+                <MessageSquareText size={15} />
+              </DebugInputModeButton>
+            </div>
+            {inputMode === "mic" ? (
+              <MicrophoneDeviceField preview={preview} />
+            ) : (
+              <Textarea
+                rows={1}
+                value={textDraft}
+                disabled={status !== "connected"}
+                onChange={(event) => setTextDraft(event.target.value)}
+                onKeyDown={(event) => {
+                  if (event.key === "Enter" && !event.shiftKey && !event.nativeEvent.isComposing) {
+                    event.preventDefault();
+                    handleSendText();
+                  }
+                }}
+                placeholder={
+                  status === "connected"
+                    ? "输入文字发送给助手，将打断当前播报…"
+                    : "开始对话后可输入文字…"
+                }
+                className="h-8 min-h-8 flex-1 resize-none overflow-hidden border-transparent bg-transparent px-2 py-1 text-sm leading-6 text-foreground shadow-none outline-none placeholder:text-muted-soft focus-visible:ring-0 disabled:opacity-100"
+              />
+            )}
+          </div>
+          {inputMode === "text" && (
+            <Button
+              size="icon"
+              className="h-10 w-10 shrink-0 rounded-full"
+              aria-label="发送调试消息"
+              disabled={status !== "connected" || !textDraft.trim()}
+              onClick={handleSendText}
+            >
+              <Send size={16} />
+            </Button>
+          )}
           {!showIdleHub && (
             <Button
               size="sm"
@@ -2333,7 +2376,7 @@ function DebugVoicePanel({
                 }
               }}
               className={[
-                "shrink-0 gap-1.5 rounded-full",
+                "h-10 shrink-0 gap-1.5 rounded-full px-4",
                 recording
                   ? "bg-destructive text-white hover:bg-destructive/90"
                   : "",
@@ -2356,7 +2399,7 @@ function DebugVoicePanel({
       {/* 底部双轨波形:会话连通后默认展开,空闲时收起 */}
       <WaveformTimelinePanel
         key={status === "connected" ? "active" : "idle"}
-        defaultOpen={status === "connected"}
+        defaultOpen={false}
         userStream={localStream}
         agentStream={remoteStream}
         active={status === "connected"}
@@ -2365,10 +2408,9 @@ function DebugVoicePanel({
   );
 }
 
-// 空闲态:图标 + 说明 + 开始按钮合一,替代「撑满高度的空 transcript + 底部细控制条」
+// 空闲态只保留一个明确入口,避免中间区域显得拥挤。
 function DebugIdleHub({
   status,
-  error,
   assistantId,
   connect,
 }: {
@@ -2377,42 +2419,23 @@ function DebugIdleHub({
   assistantId: string | null;
   connect: () => Promise<void>;
 }) {
-  const hint =
-    status === "failed"
-      ? error || "连接失败，请确认后端已启动且助手已保存后重试。"
-      : !assistantId
-        ? "请先保存助手，再开始语音预览。"
-        : "测试语音识别、响应速度与助手的播报效果。";
-
   return (
     <div className="flex min-h-0 flex-1 items-center justify-center px-6 py-8">
-      <div className="flex max-w-xs flex-col items-center gap-3 text-center">
+      <Button
-        <div className="flex h-12 w-12 items-center justify-center rounded-full bg-canvas-soft">
+        disabled={!assistantId}
-          <Mic size={22} className="text-muted-soft" />
+        onClick={() => void connect()}
-        </div>
+        className="h-11 gap-2 rounded-full px-6 text-sm font-medium shadow-sm"
-        <div className="font-display display-sm text-foreground">
+        aria-label={status === "failed" ? "重新连接" : "开始语音测试"}
-          {status === "failed" ? "连接失败" : "开始一次语音对话"}
+      >
-        </div>
+        <Mic size={18} />
-        <p className="text-xs leading-5 text-muted-foreground">{hint}</p>
+        {status === "failed" ? "重新连接" : "开始对话"}
-        <Button
+      </Button>
-          disabled={!assistantId}
-          onClick={() => void connect()}
-          className="h-11 gap-2 rounded-full px-6 text-sm font-medium shadow-sm"
-          aria-label={status === "failed" ? "重新连接" : "开始语音测试"}
-        >
-          <Mic size={18} />
-          {status === "failed" ? "重新连接" : "开始对话"}
-        </Button>
-        <p className="text-[11px] leading-4 text-muted-soft">
-          在右上角「设备」中选择麦克风
-        </p>
-      </div>
     </div>
   );
 }
 
 function DebugVideoPanel({ camera }: { camera: CameraPreview }) {
-  const { stream, error, starting, active, start } = camera;
+  const { stream, error, starting, active } = camera;
   const videoRef = useRef<HTMLVideoElement>(null);
 
   useEffect(() => {
@@ -2445,28 +2468,14 @@ function DebugVideoPanel({ camera }: { camera: CameraPreview }) {
           预览中
         </Badge>
       ) : (
-        <div className="flex max-w-xs flex-col items-center gap-3 px-6 text-center">
+        <div className="flex max-w-xs flex-col items-center gap-2 px-6 text-center">
           <Video size={28} className="text-muted-soft" />
           <div className="text-sm font-medium text-foreground">
-            {starting ? "正在开启摄像头…" : "摄像头未开启"}
+            {starting ? "正在开启摄像头…" : "摄像头不可用"}
           </div>
-          <p className="text-xs leading-5 text-muted-foreground">
+          {error && (
-            {error ?? "开启摄像头预览画面。后续可将视频流接入后端视觉理解管线。"}
+            <p className="text-xs leading-5 text-muted-foreground">{error}</p>
-          </p>
+          )}
-          <Button
-            size="sm"
-            disabled={starting}
-            onClick={() => void start()}
-            className="gap-1.5 rounded-full"
-            aria-label="开启摄像头"
-          >
-            {starting ? (
-              <Loader2 size={14} className="animate-spin" />
-            ) : (
-              <Video size={14} />
-            )}
-            开启摄像头
-          </Button>
         </div>
       )}
     </div>
@@ -2536,7 +2545,7 @@ function DebugTranscriptPanel({
     return (
       <div
         className={[
-          "flex min-h-0 flex-1 flex-col overflow-y-auto px-5 py-3",
+          "scrollbar-subtle flex min-h-0 flex-1 flex-col overflow-y-auto px-5 py-3",
           recording ? "items-center justify-center" : "pt-5",
         ].join(" ")}
       >
@@ -2558,7 +2567,7 @@ function DebugTranscriptPanel({
   return (
     <div
       ref={scrollRef}
-      className="flex min-h-0 flex-1 flex-col overflow-y-auto px-5 py-4"
+      className="scrollbar-subtle flex min-h-0 flex-1 flex-col overflow-y-auto px-5 py-4"
     >
       <div className="flex flex-col gap-4">
         {messages.map((message) => {
@@ -2949,21 +2958,51 @@ function ResourceSelectField({
 }
 
 function ToggleRow({
+  icon,
   title,
   description,
+  hint,
   checked,
   onChange,
 }: {
+  icon?: React.ReactNode;
   title: string;
-  description: string;
+  description?: string;
+  hint?: string;
   checked: boolean;
   onChange: (checked: boolean) => void;
 }) {
+  const hasIcon = Boolean(icon);
+
   return (
-    <div className="flex items-center justify-between rounded-xl border border-hairline bg-canvas-soft p-4">
+    <div
+      className={[
+        "flex items-center justify-between border border-hairline bg-canvas-soft",
+        hasIcon ? "rounded-2xl p-5" : "rounded-xl p-4",
+      ].join(" ")}
+    >
       <div>
-        <div className="font-medium text-foreground">{title}</div>
+        <div
-        <div className="mt-1 text-sm text-muted-foreground">{description}</div>
+          className={[
+            "flex items-center font-medium text-foreground",
+            hasIcon ? "gap-3" : "gap-1.5",
+          ].join(" ")}
+        >
+          {icon && (
+            <span className="flex h-10 w-10 shrink-0 items-center justify-center rounded-full bg-surface-strong text-foreground">
+              {icon}
+            </span>
+          )}
+          <span className="flex items-center gap-1.5">
+            <span>{title}</span>
+            {hint && <HelpHint text={hint} />}
+          </span>
+        </div>
+        {description && (
+          <div className="mt-1 text-sm text-muted-foreground">
+            {description}
+          </div>
+        )}
       </div>
       <Switch checked={checked} onCheckedChange={onChange} />
     </div>

frontend/src/hooks/use-voice-preview.ts

@@ -187,6 +187,8 @@ export function useVoicePreview(
     releaseResources();
     setLocalStream(null);
     setRemoteStream(null);
+    setMessages([]);
+    messageSeqRef.current = 0;
     setError(null);
     setMicWarning(null);
     setStatus("idle");