From 1ffc71657504d43effd36bc1a4e9685792ad99a7 Mon Sep 17 00:00:00 2001
From: Xin Wang <wangx@XindeMac-mini-2.local>
Date: Fri, 22 May 2026 21:30:54 +0800
Subject: [PATCH] fix text input no context update

---
 engine/pipeline.py    | 18 ++++++++----------
 engine/text_stream.py | 34 +++++++++++++++++++++++++++++++++-
 2 files changed, 41 insertions(+), 11 deletions(-)

diff --git a/engine/pipeline.py b/engine/pipeline.py
index 1dcb6b7..7eb2ae5 100644
--- a/engine/pipeline.py
+++ b/engine/pipeline.py
@@ -14,7 +14,6 @@ from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
 from pipecat.processors.aggregators.llm_context import LLMContext
 from pipecat.processors.aggregators.llm_response_universal import (
-    AssistantTurnStoppedMessage,
     LLMContextAggregatorPair,
     LLMUserAggregatorParams,
     UserTurnStoppedMessage,
@@ -34,7 +33,7 @@ from .config import EngineConfig
 from .product_protocol import ProductWebsocketSerializer
 from .services import create_llm_service, create_stt_service, create_tts_service
 from .text_input import ProductTextInputProcessor
-from .text_stream import ProductTextStreamProcessor
+from .text_stream import ProductAssistantTurnStoppedMessage, ProductTextStreamProcessor
 from .transcript_stream import ProductTranscriptStreamProcessor
 from .turn_start import InterruptionGateUserTurnStartStrategy
 
@@ -118,13 +117,14 @@ async def run_pipeline_with_serializer(
             ),
         ],
     )
-    user_aggregator, assistant_aggregator = LLMContextAggregatorPair(
+    user_aggregator, _ = LLMContextAggregatorPair(
         context,
         user_params=LLMUserAggregatorParams(
             vad_analyzer=SileroVADAnalyzer(params=vad_params),
             user_turn_strategies=user_turn_strategies,
         ),
     )
+    text_stream = ProductTextStreamProcessor(context)
 
     pipeline = Pipeline(
         [
@@ -134,10 +134,9 @@ async def run_pipeline_with_serializer(
             ProductTranscriptStreamProcessor(),
             user_aggregator,
             llm,
-            ProductTextStreamProcessor(),
+            text_stream,
             tts,
             transport.output(),
-            assistant_aggregator,
         ]
     )
 
@@ -188,11 +187,10 @@ async def run_pipeline_with_serializer(
             )
         )
 
-    # NOTE: assistant turn started/final events are emitted by
-    # ProductTextStreamProcessor, upstream of TTS, so text streams to the
-    # client ahead of audio. This logger is kept for server-side visibility.
-    @assistant_aggregator.event_handler("on_assistant_turn_stopped")
-    async def on_assistant_turn_stopped(_aggregator, message: AssistantTurnStoppedMessage):
+    @text_stream.event_handler("on_assistant_turn_stopped")
+    async def on_assistant_turn_stopped(
+        _aggregator, message: ProductAssistantTurnStoppedMessage
+    ):
         logger.info(f"Assistant: {message.content}")
 
     runner = PipelineRunner(handle_sigint=False)
diff --git a/engine/text_stream.py b/engine/text_stream.py
index e2e997e..d3f8882 100644
--- a/engine/text_stream.py
+++ b/engine/text_stream.py
@@ -1,5 +1,9 @@
 from __future__ import annotations
 
+from dataclasses import dataclass
+
+from loguru import logger
+
 from pipecat.frames.frames import (
     Frame,
     InterruptionFrame,
@@ -9,7 +13,16 @@ from pipecat.frames.frames import (
     OutputTransportMessageUrgentFrame,
     TTSSpeakFrame,
 )
+from pipecat.processors.aggregators.llm_context import LLMContext
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
+from pipecat.utils.time import time_now_iso8601
+
+
+@dataclass
+class ProductAssistantTurnStoppedMessage:
+    content: str
+    interrupted: bool
+    timestamp: str
 
 
 class ProductTextStreamProcessor(FrameProcessor):
@@ -29,10 +42,13 @@ class ProductTextStreamProcessor(FrameProcessor):
     started/delta/final sequence for its fixed text.
     """
 
-    def __init__(self) -> None:
+    def __init__(self, context: LLMContext | None = None) -> None:
         super().__init__()
+        self._context = context
         self._aggregation: list[str] = []
         self._turn_active = False
+        self._turn_start_timestamp = ""
+        self._register_event_handler("on_assistant_turn_stopped")
 
     async def process_frame(self, frame: Frame, direction: FrameDirection) -> None:
         await super().process_frame(frame, direction)
@@ -62,6 +78,7 @@ class ProductTextStreamProcessor(FrameProcessor):
             return
         self._turn_active = True
         self._aggregation = []
+        self._turn_start_timestamp = time_now_iso8601()
         await self._emit("response.text.started")
 
     async def _delta(self, text: str) -> None:
@@ -78,11 +95,26 @@ class ProductTextStreamProcessor(FrameProcessor):
         full_text = "".join(self._aggregation)
         self._turn_active = False
         self._aggregation = []
+        if self._context and full_text:
+            self._context.add_message({"role": "assistant", "content": full_text})
+            logger.info(
+                "Assistant committed to LLM context before TTS: "
+                f"{full_text[:120]}"
+            )
         await self._emit(
             "response.text.final",
             text=full_text,
             interrupted=interrupted,
         )
+        await self._call_event_handler(
+            "on_assistant_turn_stopped",
+            ProductAssistantTurnStoppedMessage(
+                content=full_text,
+                interrupted=interrupted,
+                timestamp=self._turn_start_timestamp or time_now_iso8601(),
+            ),
+        )
+        self._turn_start_timestamp = ""
 
     async def _emit(self, event_type: str, **payload: object) -> None:
         await self.push_frame(