fix(room): clear AI stream state on room switch and hide cursor for saved chunks

- Clear activeAiStream, streamingChunks, and timers when room changes
- Add showCursor prop to OrderedStreamChunks — only show cursor
  during active streaming, not for saved content

src/components/room/message/MessageBubble.tsx

@@ -52,9 +52,12 @@ function parseSavedChunks(raw: string | null | undefined): Array<{ type: string;
 function OrderedStreamChunks({
   chunks,
   onMentionClick,
+  showCursor = true,
 }: {
   chunks: Array<{ type: string; content: string }>;
   onMentionClick?: (type: string, id: string, label: string) => void;
+  /** Show blinking cursor — only during active streaming */
+  showCursor?: boolean;
 }) {
   // Group consecutive same-type chunks (tool_call hidden)
   const groups: Array<{ type: 'thinking' | 'answer'; content: string }> = [];
@@ -78,8 +81,8 @@ function OrderedStreamChunks({
           <MessageContent key={i} content={group.content} onMentionClick={onMentionClick} />
         ),
       )}
-      {/* Streaming cursor */}
-      <span className="discord-streaming-cursor" />
+      {/* Streaming cursor — only shown during active streaming */}
+      {showCursor && <span className="discord-streaming-cursor" />}
     </>
   );
 }
@@ -415,10 +418,11 @@ export const MessageBubble = memo(function MessageBubble({
                           onMentionClick={handleMentionClick}
                         />
                       ) : parseSavedChunks(message.thinking_content) ? (
-                        /* Saved ordered chunks — render in original order */
+                        /* Saved ordered chunks — render in original order (no cursor) */
                         <OrderedStreamChunks
                           chunks={parseSavedChunks(message.thinking_content)!}
                           onMentionClick={handleMentionClick}
+                          showCursor={false}
                         />
                       ) : (
                         /* Legacy: aggregated thinking at top, content at bottom */

src/contexts/room-context.tsx

@@ -278,6 +278,13 @@ export function RoomProvider({
       setMessages([]);
       setIsHistoryLoaded(false);
       setNextCursor(null);
+      // Clear AI stream state — prevents ghost "thinking..." indicator on room switch
+      setActiveAiStream(null);
+      setStreamingChunks(new Map());
+      streamingChunksRef.current.clear();
+      // Clear all streaming timers
+      streamingTimersRef.current.forEach((timer) => clearTimeout(timer));
+      streamingTimersRef.current.clear();
 
       // Merge any buffered messages for the new room (Bug 3 fix)
       if (activeRoomId) {
@@ -1310,6 +1317,7 @@ export function RoomProvider({
   }, [projectName]);
 
   // Fetch room AI configs for @ai: mention suggestions
+  // Retry up to 3 times if any config has missing modelName
   const fetchRoomAiConfigs = useCallback(async () => {
     const client = wsClientRef.current;
     if (!activeRoomId || !client) {
@@ -1318,13 +1326,28 @@ export function RoomProvider({
     }
     setAiConfigsLoading(true);
     try {
-      const configs = await client.aiList(activeRoomId);
-      setRoomAiConfigs(
-        configs.map((cfg) => ({
+      const fetchOnce = async (): Promise<RoomAiConfig[]> => {
+        const configs = await client.aiList(activeRoomId);
+        return configs.map((cfg) => ({
           model: cfg.model,
           modelName: cfg.model_name,
-        })),
-      );
+        }));
+      };
+
+      let configs = await fetchOnce();
+      let retries = 0;
+      const maxRetries = 3;
+
+      while (
+        configs.some((c) => !c.modelName) &&
+        retries < maxRetries
+      ) {
+        retries++;
+        await new Promise((resolve) => setTimeout(resolve, 500));
+        configs = await fetchOnce();
+      }
+
+      setRoomAiConfigs(configs);
     } catch {
       setRoomAiConfigs([]);
     } finally {