Fix: gluing down while scrolling and out of time issues

src/components/chat/ChatMessageList.vue

@@ -24,28 +24,22 @@ const emit = defineEmits<{
 
 const { t } = useI18n();
 const messagesContainer = ref<HTMLElement | null>(null);
-const isUserScrolledUp = ref(false);
 const BOTTOM_THRESHOLD = 80; // px from bottom before we consider the user "at bottom"
 
-const onContainerScroll = () => {
-  const el = messagesContainer.value;
-  if (!el) return;
-  const distanceFromBottom = el.scrollHeight - el.scrollTop - el.clientHeight;
-  isUserScrolledUp.value = distanceFromBottom > BOTTOM_THRESHOLD;
-};
-
 const scrollToBottom = async (force = false) => {
   await nextTick();
   const el = messagesContainer.value;
   if (!el) return;
-  if (force || !isUserScrolledUp.value) {
+  // Read position at execution time to avoid the race condition where the
+  // isUserScrolledUp flag hasn't been updated yet when a queued callback fires.
+  const distanceFromBottom = el.scrollHeight - el.scrollTop - el.clientHeight;
+  if (force || distanceFromBottom <= BOTTOM_THRESHOLD) {
     el.scrollTop = el.scrollHeight;
-    isUserScrolledUp.value = false;
   }
 };
 
-// Deep watch: fires on both new messages (length) AND text changes during rAF animation.
-// Smart scroll: only follows bottom if the user hasn't scrolled up to read history.
+// Deep watch: fires on both new messages AND text changes during rAF animation.
+// Smart scroll: reads scroll position directly at check time (no stale flag).
 watch(() => props.messages, () => {
   scrollToBottom(false);
 }, { deep: true });
@@ -63,7 +57,6 @@ defineExpose({ scrollToBottom: () => scrollToBottom(true) });
     ref="messagesContainer"
     id="tour-chat-container"
     class="flex-1 overflow-y-auto px-6 pt-8 custom-scrollbar relative"
-    @scroll="onContainerScroll"
   >
     <!-- Loading Overlay for History -->
     <div

src/stores/chat.ts

@@ -373,7 +373,10 @@ export const useChatStore = defineStore('chat', () => {
         if (!text.trim() && (!attachments || attachments.length === 0)) return;
 
         isLoading.value = true;
-        isGenerating.value = true;
+        // isGenerating stays false until first token arrives so that
+        // isBotTyping (= isLoading && !isGenerating) is true while waiting —
+        // this shows the bouncing-dots indicator during the network round-trip.
+        isGenerating.value = false;
         error.value = null;
         abortController.value = new AbortController();
 
@@ -424,6 +427,8 @@ export const useChatStore = defineStore('chat', () => {
 
                 // Token chunk — accumulate raw text (used as abort fallback); animation runs on final message
                 if (parsed.type === 'token' && typeof parsed.token === 'string') {
+                    // First token: switch from "waiting" to "generating" — hides bouncing dots
+                    if (!isGenerating.value) isGenerating.value = true;
                     _streamingRawText += parsed.token;
                     if (parsed.session_id) {
                         // Update session ID without touching messages array
@@ -440,6 +445,9 @@ export const useChatStore = defineStore('chat', () => {
                         }
                     }
                 } else if (parsed.type === 'message') {
+                    // If no tokens arrived before the final message (e.g. non-streaming
+                    // backend path), mark as generating now so state is cleaned up correctly.
+                    if (!isGenerating.value) isGenerating.value = true;
                     const streamingMsg = messages.value[streamingIndex];
                     if (streamingMsg) {
                         streamingMsg.id = parsed.id;