diff --git a/homelab-ai-bot/telegram_bot.py b/homelab-ai-bot/telegram_bot.py
index 88ae6df5..887ab6d8 100644
--- a/homelab-ai-bot/telegram_bot.py
+++ b/homelab-ai-bot/telegram_bot.py
@@ -117,6 +117,7 @@ log = logging.getLogger("hausmeister")
 
 ALLOWED_CHAT_IDS: set[int] = set()
 CHAT_ID: int | None = None
+ACTIVE_LLM_TASKS: dict[int, asyncio.Task] = {}
 
 
 def _load_token_and_chat():
@@ -167,6 +168,9 @@ async def cmd_status(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
         if len(text) > 4000:
             text = text[:4000] + "\n..."
         await update.message.reply_text(text)
+    except asyncio.CancelledError:
+        log.info("Freitext-Lauf abgebrochen")
+        return
     except Exception as e:
         await update.message.reply_text(f"Fehler: {e}")
 
@@ -409,7 +413,20 @@ async def handle_voice(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
         context.last_suggest_result = {"type": None}
         context.set_source_type("telegram_voice")
         handlers = context.get_tool_handlers(session_id=session_id)
-        answer = await asyncio.to_thread(llm.ask_with_tools, text, handlers, session_id=session_id)
+        llm_task = asyncio.create_task(
+            asyncio.to_thread(llm.ask_with_tools, text, handlers, session_id=session_id)
+        )
+        ACTIVE_LLM_TASKS[update.effective_chat.id] = llm_task
+
+        waited = 0
+        while not llm_task.done():
+            await asyncio.sleep(30)
+            waited += 30
+            if not llm_task.done():
+                await update.message.reply_text("⏳ Suche laeuft noch (" + str(waited) + "s)...")
+
+        answer = await llm_task
+
         if session_id:
             memory_client.log_message(session_id, "user", text)
             memory_client.log_message(session_id, "assistant", answer)
@@ -711,6 +728,16 @@ async def handle_message(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
     if not text:
         return
 
+    if text.strip().lower() in ("abbruch", "stop", "stopp", "cancel"):
+        chat_id = update.effective_chat.id
+        task = ACTIVE_LLM_TASKS.get(chat_id)
+        if task and not task.done():
+            task.cancel()
+            await update.message.reply_text("🛑 Abgebrochen.")
+        else:
+            await update.message.reply_text("Kein laufender Suchlauf.")
+        return
+
     cmd = BUTTON_MAP.get(text)
     if cmd == "status":
         return await cmd_status(update, ctx)
@@ -735,7 +762,20 @@ async def handle_message(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
         context.last_suggest_result = {"type": None}
         context.set_source_type("telegram_text")
         handlers = context.get_tool_handlers(session_id=session_id)
-        answer = await asyncio.to_thread(llm.ask_with_tools, text, handlers, session_id=session_id)
+        llm_task = asyncio.create_task(
+            asyncio.to_thread(llm.ask_with_tools, text, handlers, session_id=session_id)
+        )
+        ACTIVE_LLM_TASKS[update.effective_chat.id] = llm_task
+
+        waited = 0
+        while not llm_task.done():
+            await asyncio.sleep(30)
+            waited += 30
+            if not llm_task.done():
+                await update.message.reply_text("Suche laeuft noch (" + str(waited) + "s)...")
+
+        answer = await llm_task
+
         if session_id:
             memory_client.log_message(session_id, "user", text)
             memory_client.log_message(session_id, "assistant", answer)
@@ -744,9 +784,14 @@ async def handle_message(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
         log.info("suggest_result: type=%s", suggest.get("type"))
 
         await update.message.reply_text(answer[:4000], reply_markup=KEYBOARD)
+    except asyncio.CancelledError:
+        log.info("Freitext-Lauf abgebrochen")
+        return
     except Exception as e:
         log.exception("Fehler bei Freitext")
         await update.message.reply_text(f"Fehler: {e}")
+    finally:
+        ACTIVE_LLM_TASKS.pop(update.effective_chat.id, None)
 
 
 async def handle_callback(update: Update, ctx: ContextTypes.DEFAULT_TYPE):