diff --git a/xtts/f5tts/bridge.py b/xtts/f5tts/bridge.py
index 43907e1..137e7aa 100644
--- a/xtts/f5tts/bridge.py
+++ b/xtts/f5tts/bridge.py
@@ -67,6 +67,10 @@ GPU_IDS = os.getenv("NVIDIA_VISIBLE_DEVICES", "").strip()
 WORKER_SERVICE = "f5tts"
 INSTANCE_ID = f"{WORKER_SERVICE}@{NODE_NAME}"
 WORKER_PING_INTERVAL_S = int(os.getenv("WORKER_PING_INTERVAL_S", "10"))
+# Empfangs-Watchdog: kommt in RX_STALE_S kein Broadcast rein (ein echter Raum hat
+# staendig Traffic, z.B. sat_hello alle 25s / Brain-Polling), gilt die Verbindung
+# als halb-tot (Caddy pongt die WS-Pings selbst) -> Zwangs-Reconnect.
+RX_STALE_S = int(os.getenv("RX_STALE_S", "60"))
 _tts_busy = False   # True waehrend eine Synthese laeuft (busy-Report im ping)
 
 # ── Auslastungs-Monitor (Stage E) ──────────────────────────
@@ -917,7 +921,12 @@ async def run_loop(runner: F5Runner) -> None:
                     busy_fn=lambda: _tts_busy or not _tts_queue.empty()))
 
                 try:
-                    async for raw in ws:
+                    while True:
+                        try:
+                            raw = await asyncio.wait_for(ws.recv(), timeout=RX_STALE_S)
+                        except asyncio.TimeoutError:
+                            logger.warning("Kein RVS-Traffic seit %ds — Verbindung halb-tot, reconnect", RX_STALE_S)
+                            raise ConnectionError("rvs-stale")
                         try:
                             msg = json.loads(raw)
                         except Exception:
diff --git a/xtts/llm-adapter/adapter.py b/xtts/llm-adapter/adapter.py
index aebb333..208955c 100644
--- a/xtts/llm-adapter/adapter.py
+++ b/xtts/llm-adapter/adapter.py
@@ -55,6 +55,10 @@ GPU_IDS = os.getenv("NVIDIA_VISIBLE_DEVICES", "").strip()
 WORKER_SERVICE = "llm"
 INSTANCE_ID = f"{WORKER_SERVICE}@{NODE_NAME}"
 WORKER_PING_INTERVAL_S = int(os.getenv("WORKER_PING_INTERVAL_S", "10"))
+# Empfangs-Watchdog: kommt in RX_STALE_S kein Broadcast rein (ein echter Raum hat
+# staendig Traffic, z.B. sat_hello alle 25s / Brain-Polling), gilt die Verbindung
+# als halb-tot (Caddy pongt die WS-Pings selbst) -> Zwangs-Reconnect.
+RX_STALE_S = int(os.getenv("RX_STALE_S", "60"))
 _inflight = 0   # laufende llm_requests (busy-Report im ping)
 # Qwen3 hat Thinking-Mode default AN — dann verbraet es Tokens in einem
 # <think>-Block und liefert (bei kleinem max_tokens) leeren/abgeschnittenen
@@ -424,7 +428,12 @@ async def _run() -> None:
                 retry_s = 2
                 tls_fallback_tried = False
                 ping_task = asyncio.create_task(_worker_register(ws))
-                async for raw in ws:
+                while True:
+                    try:
+                        raw = await asyncio.wait_for(ws.recv(), timeout=RX_STALE_S)
+                    except asyncio.TimeoutError:
+                        logger.warning("Kein RVS-Traffic seit %ds — Verbindung halb-tot, reconnect", RX_STALE_S)
+                        raise ConnectionError("rvs-stale")
                     try:
                         msg = json.loads(raw)
                     except Exception:
diff --git a/xtts/voxtral/bridge.py b/xtts/voxtral/bridge.py
index f3e2051..99bc92e 100644
--- a/xtts/voxtral/bridge.py
+++ b/xtts/voxtral/bridge.py
@@ -67,6 +67,10 @@ GPU_IDS = os.getenv("NVIDIA_VISIBLE_DEVICES", "").strip()
 WORKER_SERVICE = "voxtral"
 INSTANCE_ID = f"{WORKER_SERVICE}@{NODE_NAME}"
 WORKER_PING_INTERVAL_S = int(os.getenv("WORKER_PING_INTERVAL_S", "10"))
+# Empfangs-Watchdog: kommt in RX_STALE_S kein Broadcast rein (ein echter Raum hat
+# staendig Traffic, z.B. sat_hello alle 25s / Brain-Polling), gilt die Verbindung
+# als halb-tot (Caddy pongt die WS-Pings selbst) -> Zwangs-Reconnect.
+RX_STALE_S = int(os.getenv("RX_STALE_S", "60"))
 
 # ── Auslastungs-Monitor (Stage E) ──────────────────────────
 import node_stats
@@ -753,7 +757,12 @@ async def run_loop(sessions: SessionManager) -> None:
                 ping_task = asyncio.create_task(_worker_register(
                     ws, model=VOXTRAL_MODEL,
                     busy_fn=lambda: bool(sessions._sessions)))
-                async for raw in ws:
+                while True:
+                    try:
+                        raw = await asyncio.wait_for(ws.recv(), timeout=RX_STALE_S)
+                    except asyncio.TimeoutError:
+                        logger.warning("Kein RVS-Traffic seit %ds — Verbindung halb-tot, reconnect", RX_STALE_S)
+                        raise ConnectionError("rvs-stale")
                     try:
                         msg = json.loads(raw)
                     except Exception:
diff --git a/xtts/whisper/bridge.py b/xtts/whisper/bridge.py
index 4247152..42e254a 100644
--- a/xtts/whisper/bridge.py
+++ b/xtts/whisper/bridge.py
@@ -66,6 +66,10 @@ GPU_IDS = os.getenv("NVIDIA_VISIBLE_DEVICES", "").strip()
 WORKER_SERVICE = "whisper"
 INSTANCE_ID = f"{WORKER_SERVICE}@{NODE_NAME}"
 WORKER_PING_INTERVAL_S = int(os.getenv("WORKER_PING_INTERVAL_S", "10"))
+# Empfangs-Watchdog: kommt in RX_STALE_S kein Broadcast rein (ein echter Raum hat
+# staendig Traffic, z.B. sat_hello alle 25s / Brain-Polling), gilt die Verbindung
+# als halb-tot (Caddy pongt die WS-Pings selbst) -> Zwangs-Reconnect.
+RX_STALE_S = int(os.getenv("RX_STALE_S", "60"))
 
 # ── Auslastungs-Monitor (Stage E) ──────────────────────────
 import node_stats
@@ -903,7 +907,12 @@ async def run_loop(runner: WhisperRunner, sessions: SessionManager) -> None:
                     ws, model=(runner.model_size or WHISPER_MODEL),
                     busy_fn=lambda: bool(sessions._sessions)))
 
-                async for raw in ws:
+                while True:
+                    try:
+                        raw = await asyncio.wait_for(ws.recv(), timeout=RX_STALE_S)
+                    except asyncio.TimeoutError:
+                        logger.warning("Kein RVS-Traffic seit %ds — Verbindung halb-tot, reconnect", RX_STALE_S)
+                        raise ConnectionError("rvs-stale")
                     try:
                         msg = json.loads(raw)
                     except Exception: