diff --git a/pipeline/conversation/agentic.py b/pipeline/conversation/agentic.py
index dc57660..57123f2 100644
--- a/pipeline/conversation/agentic.py
+++ b/pipeline/conversation/agentic.py
@@ -68,6 +68,9 @@ class AgenticBackend:
             if delta:
                 yield delta
 
+    def commit(self, text: str) -> None:
+        pass  # remote service owns history
+
     def reset(self) -> None:
         pass  # session reset would require an API call if supported
 
diff --git a/pipeline/conversation/base.py b/pipeline/conversation/base.py
index 8014fad..19c70de 100644
--- a/pipeline/conversation/base.py
+++ b/pipeline/conversation/base.py
@@ -11,6 +11,11 @@ class ConversationBackend(Protocol):
         """Send a user message, yield assistant text deltas as they arrive."""
         ...
 
+    def commit(self, text: str) -> None:
+        """Persist the assistant response after successful delivery. Call
+        only after the turn actually played out. No-op for managed backends."""
+        ...
+
     def reset(self) -> None:
         """Clear conversation history / start a new session."""
         ...
diff --git a/pipeline/conversation/conversational.py b/pipeline/conversation/conversational.py
index b5755ff..c41ba8c 100644
--- a/pipeline/conversation/conversational.py
+++ b/pipeline/conversation/conversational.py
@@ -75,17 +75,17 @@ class ConversationalBackend:
         kwargs["stream"] = True
         stream = await self.client.chat.completions.create(**kwargs)
 
-        parts: list[str] = []
-        try:
-            async for chunk in stream:
-                if not chunk.choices:
-                    continue
-                delta = chunk.choices[0].delta.content or ""
-                if delta:
-                    parts.append(delta)
-                    yield delta
-        finally:
-            self._finalize("".join(parts))
+        async for chunk in stream:
+            if not chunk.choices:
+                continue
+            delta = chunk.choices[0].delta.content or ""
+            if delta:
+                yield delta
+
+    def commit(self, text: str) -> None:
+        """Persist the assistant response to history after successful
+        delivery. The caller joins whatever was actually sent to TTS."""
+        self._finalize(text)
 
     def reset(self) -> None:
         self.messages = [{"role": "system", "content": self.cfg["system_prompt"]}]
diff --git a/pipeline/main.py b/pipeline/main.py
index 47925d7..4de75f0 100644
--- a/pipeline/main.py
+++ b/pipeline/main.py
@@ -389,6 +389,8 @@ async def process_utterances(config: dict, manager: DeviceManager, utterance_que
                                  volume=0, fade=0)
 
             response_text = " ".join(full_response)
+            if response_text:
+                device.conversation.commit(response_text)
             device.last_response = response_text
             elapsed = time.monotonic() - turn_t0
             ttfs = f"{first_sentence_at - turn_t0:.2f}s" if first_sentence_at else "—"