diff --git a/frontend/src/components/ChatMessage.vue b/frontend/src/components/ChatMessage.vue
index 78b7afe..9e0bb19 100644
--- a/frontend/src/components/ChatMessage.vue
+++ b/frontend/src/components/ChatMessage.vue
@@ -3,7 +3,7 @@ import { computed } from "vue";
 import { renderMarkdown } from "@/utils/markdown";
 import { useSettingsStore } from "@/stores/settings";
 import ToolCallCard from "@/components/ToolCallCard.vue";
-import type { Message } from "@/types/chat";
+import type { GenerationTiming, Message } from "@/types/chat";
 
 const settingsStore = useSettingsStore();
 
@@ -34,6 +34,26 @@ const roleLabel = computed(() => {
       return props.message.role;
   }
 });
+
+function formatMs(ms: number | null | undefined): string {
+  if (ms == null) return "";
+  if (ms < 1000) return `${ms}ms`;
+  return `${(ms / 1000).toFixed(1)}s`;
+}
+
+const timingParts = computed((): string[] => {
+  const t = props.message.timing;
+  if (!t) return [];
+  const parts: string[] = [];
+  if (t.total_ms != null) parts.push(`${formatMs(t.total_ms)} total`);
+  if (t.ttft_ms != null) parts.push(`first token ${formatMs(t.ttft_ms)}`);
+  if (t.intent_ms != null) parts.push(`analyzed ${formatMs(t.intent_ms)}`);
+  for (const tool of t.tools ?? []) {
+    parts.push(`${tool.name.replace(/_/g, " ")} ${formatMs(tool.ms)}`);
+  }
+  if (t.generation_ms != null) parts.push(`generated ${formatMs(t.generation_ms)}`);
+  return parts;
+});
 </script>
 
 <template>
@@ -66,6 +86,12 @@ const roleLabel = computed(() => {
         </div>
       </div>
       <span v-if="formattedTime" class="message-timestamp">{{ formattedTime }}</span>
+      <div v-if="timingParts.length" class="message-timing">
+        <span class="timing-icon">⏱</span>
+        <span v-for="(part, i) in timingParts" :key="i" class="timing-part">
+          <span v-if="i > 0" class="timing-sep">·</span>{{ part }}
+        </span>
+      </div>
     </div>
   </div>
 </template>
@@ -193,4 +219,24 @@ const roleLabel = computed(() => {
   margin-top: 0.15rem;
   padding: 0 0.5rem;
 }
+.message-timing {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: 0.2rem;
+  padding: 0.1rem 0.5rem 0;
+  font-size: 0.68rem;
+  color: var(--color-text-muted);
+  opacity: 0.7;
+}
+.timing-icon {
+  font-size: 0.65rem;
+}
+.timing-part {
+  white-space: nowrap;
+}
+.timing-sep {
+  margin-right: 0.2rem;
+  opacity: 0.5;
+}
 </style>
diff --git a/frontend/src/stores/chat.ts b/frontend/src/stores/chat.ts
index 66dfc11..c5529e8 100644
--- a/frontend/src/stores/chat.ts
+++ b/frontend/src/stores/chat.ts
@@ -12,6 +12,7 @@ import type {
   Conversation,
   ConversationDetail,
   ContextMeta,
+  GenerationTiming,
   Message,
   OllamaStatus,
   SendMessageResponse,
@@ -206,6 +207,7 @@ export const useChatStore = defineStore("chat", () => {
                   ? [...streamingToolCalls.value]
                   : null,
                 created_at: new Date().toISOString(),
+                timing: event.data.timing as GenerationTiming | undefined,
               };
               if (currentConversation.value?.id === convId) {
                 currentConversation.value.messages.push(assistantMsg);
diff --git a/frontend/src/types/chat.ts b/frontend/src/types/chat.ts
index beb15f9..0f560c3 100644
--- a/frontend/src/types/chat.ts
+++ b/frontend/src/types/chat.ts
@@ -1,3 +1,11 @@
+export interface GenerationTiming {
+  total_ms: number;
+  intent_ms: number | null;
+  ttft_ms: number | null;
+  generation_ms: number | null;
+  tools: Array<{ name: string; ms: number }>;
+}
+
 export interface ToolCallRecord {
   function: string;
   arguments: Record<string, unknown>;
@@ -14,6 +22,7 @@ export interface Message {
   context_note_id: number | null;
   tool_calls?: ToolCallRecord[] | null;
   created_at: string;
+  timing?: GenerationTiming;
 }
 
 export interface SendMessageResponse {
diff --git a/frontend/src/views/ChatView.vue b/frontend/src/views/ChatView.vue
index 512e3f7..268e4a6 100644
--- a/frontend/src/views/ChatView.vue
+++ b/frontend/src/views/ChatView.vue
@@ -71,7 +71,7 @@ const inputPlaceholder = computed(() => {
 onMounted(async () => {
   document.addEventListener("keydown", onGlobalKeydown);
   await store.fetchConversations();
-  if (convId.value) {
+  if (convId.value && store.currentConversation?.id !== convId.value) {
     await store.fetchConversation(convId.value);
   }
   nextTick(() => inputEl.value?.focus());
diff --git a/src/fabledassistant/services/generation_task.py b/src/fabledassistant/services/generation_task.py
index ed6710f..780300b 100644
--- a/src/fabledassistant/services/generation_task.py
+++ b/src/fabledassistant/services/generation_task.py
@@ -19,6 +19,7 @@ from fabledassistant.services.generation_buffer import GenerationBuffer, Generat
 from fabledassistant.services.llm import ChatChunk, generate_completion, stream_chat, stream_chat_with_tools
 from fabledassistant.services.chat import update_conversation_title
 from fabledassistant.services.intent import classify_intent
+from fabledassistant.services.logging import log_generation
 from fabledassistant.services.settings import get_setting
 from fabledassistant.services.tools import get_tools_for_user, execute_tool
 
@@ -112,6 +113,15 @@ async def run_generation(
     # Emit context event
     buf.append_event("context", {"context": context_meta})
 
+    t_start = time.monotonic()
+    timing: dict = {
+        "intent_ms": None,
+        "tools": [],
+        "ttft_ms": None,
+        "generation_ms": None,
+        "total_ms": None,
+    }
+
     last_flush = time.monotonic()
     all_tool_calls: list[dict] = []
 
@@ -142,7 +152,9 @@ async def run_generation(
                     if m.get("role") in ("user", "assistant") and m.get("content")
                 ][-6:]
                 buf.append_event("status", {"status": "Analyzing your request..."})
+                t_intent = time.monotonic()
                 intent = await classify_intent(user_content, tools, intent_model, history=intent_history)
+                timing["intent_ms"] = int((time.monotonic() - t_intent) * 1000)
                 if intent.should_execute:
                     logger.info(
                         "Intent router detected tool (confidence=%s): %s(%s)",
@@ -155,7 +167,9 @@ async def run_generation(
                     )
                 if intent.should_execute:
                     buf.append_event("status", {"status": f"{_TOOL_LABELS.get(intent.tool_name, 'Working')}..."})
+                    t_tool = time.monotonic()
                     result = await execute_tool(user_id, intent.tool_name, intent.arguments)
+                    timing["tools"].append({"name": intent.tool_name, "ms": int((time.monotonic() - t_tool) * 1000)})
                     logger.info("Intent-routed tool %s result: success=%s", intent.tool_name, result.get("success"))
 
                     tool_record = {
@@ -182,12 +196,15 @@ async def run_generation(
                     continue  # Next round: LLM streams response incorporating result
 
             buf.append_event("status", {"status": "Generating response..." if _round == 0 else "Composing response..."})
+            t_stream = time.monotonic()
             async for chunk in stream_chat_with_tools(messages, model, tools=tools):
                 if buf.cancel_event.is_set():
                     cancelled = True
                     break
 
                 if chunk.type == "content":
+                    if timing["ttft_ms"] is None:
+                        timing["ttft_ms"] = int((time.monotonic() - t_start) * 1000)
                     buf.content_so_far += chunk.content
                     # Filter out "[TOOL_CALLS]" marker from streaming output
                     clean = _TOOL_CALL_MARKER.sub("", chunk.content)
@@ -213,7 +230,9 @@ async def run_generation(
                         logger.info("Executing tool: %s(%s)", tool_name, json.dumps(arguments)[:200])
                         buf.append_event("status", {"status": f"{_TOOL_LABELS.get(tool_name, 'Working')}..."})
 
+                        t_tool = time.monotonic()
                         result = await execute_tool(user_id, tool_name, arguments)
+                        timing["tools"].append({"name": tool_name, "ms": int((time.monotonic() - t_tool) * 1000)})
                         logger.info("Tool %s result: success=%s", tool_name, result.get("success"))
 
                         tool_record = {
@@ -228,6 +247,8 @@ async def run_generation(
                         # Emit tool_call SSE event
                         buf.append_event("tool_call", {"tool_call": tool_record})
 
+            timing["generation_ms"] = int((time.monotonic() - t_stream) * 1000)
+
             if cancelled:
                 logger.info("Generation cancelled for conv %d", conv_id)
                 break
@@ -298,9 +319,20 @@ async def run_generation(
                         fallback += "..."
                     await update_conversation_title(user_id, conv_id, fallback)
 
+        timing["total_ms"] = int((time.monotonic() - t_start) * 1000)
+        logger.info(
+            "Generation timing for conv %d: total=%dms ttft=%s intent=%s tools=%s generation=%s",
+            conv_id, timing["total_ms"], timing["ttft_ms"], timing["intent_ms"],
+            [(t["name"], t["ms"]) for t in timing["tools"]], timing["generation_ms"],
+        )
+        try:
+            await log_generation(user_id, conv_id, model, timing)
+        except Exception:
+            logger.warning("Failed to persist generation timing for conv %d", conv_id, exc_info=True)
+
         buf.state = GenerationState.COMPLETED
         buf.finished_at = time.monotonic()
-        buf.append_event("done", {"done": True, "message_id": msg_id})
+        buf.append_event("done", {"done": True, "message_id": msg_id, "timing": timing})
 
     except Exception as e:
         logger.exception("Error in generation task for conversation %d", conv_id)
diff --git a/src/fabledassistant/services/logging.py b/src/fabledassistant/services/logging.py
index df8dac3..e7a07c0 100644
--- a/src/fabledassistant/services/logging.py
+++ b/src/fabledassistant/services/logging.py
@@ -59,6 +59,26 @@ async def log_usage(
         await session.commit()
 
 
+async def log_generation(
+    user_id: int,
+    conv_id: int,
+    model: str,
+    timing: dict,
+) -> None:
+    """Persist per-generation timing breakdown to app_logs for benchmarking."""
+    async with async_session() as session:
+        log = AppLog(
+            category="usage",
+            user_id=user_id,
+            action="generation",
+            endpoint=f"/chat/conversations/{conv_id}",
+            duration_ms=timing.get("total_ms"),
+            details=json.dumps({"model": model, **timing}),
+        )
+        session.add(log)
+        await session.commit()
+
+
 async def log_error(
     user_id: int | None = None,
     username: str | None = None,