diff --git a/js/components/input-area.js b/js/components/input-area.js
index 18f016b..be530e3 100644
--- a/js/components/input-area.js
+++ b/js/components/input-area.js
@@ -370,6 +370,9 @@ export async function sendMessage() {
     filePreviewEl.innerHTML = '';
     autoResizeTextarea();
 
+    // 助手占位
+    appendAssistantPlaceholder();
+
     // 流式回复
     state.set(KEYS.IS_STREAMING, true);
     updateSendButton(true);
@@ -465,8 +468,8 @@ ${ragContext}
         console.log('[RAG] numCtx:', numCtx, 'system长度:', chatParams.system?.length || 0);
         console.log('[RAG] 开始调用 chatStream...');
 
-        // 助手占位（放在RAG检索之后，确保消息顺序：用户 → 检索状态 → AI）
-        appendAssistantPlaceholder();
+        // 开始回复时清除 RAG 临时状态消息
+        document.querySelectorAll('.rag-status').forEach(el => el.remove());
 
         await api.chatStream(chatParams, (chunk) => {
             if (chunk.message) {
@@ -510,6 +513,9 @@ ${ragContext}
     } catch (err) {
         console.error('[InputArea] 流式聊天错误:', err);
 
+        // 清除 RAG 临时状态消息
+        document.querySelectorAll('.rag-status').forEach(el => el.remove());
+
         if (err.name === 'AbortError') {
             const placeholder = document.querySelector('#messagesContainer .message.assistant.loading');
             const loadingDots = placeholder?.querySelector('.loading-dots');