diff --git a/modules/features/teamsbot/service.py b/modules/features/teamsbot/service.py
index 4eaee284..a8469661 100644
--- a/modules/features/teamsbot/service.py
+++ b/modules/features/teamsbot/service.py
@@ -524,11 +524,47 @@ class TeamsbotService:
         if not text:
             return
 
-        # Speaker hints are lightweight caption-derived signals used only to
-        # attribute audio-stream STT to likely speakers. They are not persisted.
+        # Speaker hints are lightweight caption-derived signals used for
+        # speaker attribution. We do NOT persist caption text as transcript.
+        # However, for address detection we still allow transient analysis from
+        # speaker hints (without DB write), otherwise direct calls like
+        # "Nyla, hörst du mich?" can be missed when audio capture is silent.
         if source in ("caption", "speakerHint"):
             self._registerSpeakerHint(speaker, text)
-        if source in ("caption", "speakerHint"):
+
+            if (
+                source == "speakerHint"
+                and isFinal
+                and self.config.responseMode != TeamsbotResponseMode.TRANSCRIBE_ONLY
+            ):
+                # Keep hint text only in volatile context (not persisted).
+                self._contextBuffer.append({
+                    "speaker": speaker or "Unknown",
+                    "text": text,
+                    "timestamp": getUtcTimestamp(),
+                    "source": "speakerHint",
+                })
+                maxSegments = self.config.contextWindowSegments
+                if len(self._contextBuffer) > maxSegments:
+                    self._contextBuffer = self._contextBuffer[-maxSegments:]
+
+                shouldTriggerFromHint = self._shouldTriggerAnalysis(text, allowPeriodic=False)
+                logger.debug(
+                    f"Session {sessionId}: speakerHint shouldTriggerAnalysis={shouldTriggerFromHint}, "
+                    f"bufferSize={len(self._contextBuffer)}"
+                )
+                if shouldTriggerFromHint:
+                    logger.info(
+                        f"Session {sessionId}: Triggering AI analysis from speakerHint "
+                        f"(buffer: {len(self._contextBuffer)} segments)"
+                    )
+                    await self._analyzeAndRespond(
+                        sessionId,
+                        interface,
+                        voiceInterface,
+                        websocket,
+                        {"id": None, "speaker": speaker, "text": text, "source": source},
+                    )
             return
 
         # Filter out the bot's own speech entirely — captions of the bot's
@@ -667,7 +703,7 @@ class TeamsbotService:
         
         return False
 
-    def _shouldTriggerAnalysis(self, transcriptText: str) -> bool:
+    def _shouldTriggerAnalysis(self, transcriptText: str, allowPeriodic: bool = True) -> bool:
         """
         Decide whether to trigger AI analysis based on the latest transcript.
         Triggers:
@@ -711,7 +747,7 @@ class TeamsbotService:
             return False
 
         # Periodic trigger
-        if timeSinceLastCall >= self.config.triggerIntervalSeconds:
+        if allowPeriodic and timeSinceLastCall >= self.config.triggerIntervalSeconds:
             logger.info(f"Trigger: Periodic interval ({self.config.triggerIntervalSeconds}s) elapsed ({timeSinceLastCall:.1f}s since last call)")
             return True