From 08fe9157cc567170c8c78c0f495bb2ae29c15610 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:01:50 -0400
Subject: [PATCH 01/12] Widen run_stt/run_tts return type to
 AsyncGenerator[Frame | None, None]

The push-based STT/TTS implementations send audio/text over a socket and
receive results via a separate receive task, so there is nothing to
yield inline. They yield `None` by design. The previous declaration of
`AsyncGenerator[Frame, None]` disagreed with that, while the consumer
(`AIService.process_generator`) already accepted `Frame | None`. Widen
the producer side (abstract base and every subclass) so the type honestly
describes the contract.

Pure annotation change; no runtime behavior difference.
---
 src/pipecat/services/assemblyai/stt.py              | 2 +-
 src/pipecat/services/asyncai/tts.py                 | 4 ++--
 src/pipecat/services/aws/stt.py                     | 2 +-
 src/pipecat/services/azure/stt.py                   | 2 +-
 src/pipecat/services/cartesia/stt.py                | 2 +-
 src/pipecat/services/cartesia/tts.py                | 4 ++--
 src/pipecat/services/deepgram/flux/sagemaker/stt.py | 2 +-
 src/pipecat/services/deepgram/flux/stt.py           | 2 +-
 src/pipecat/services/deepgram/sagemaker/stt.py      | 2 +-
 src/pipecat/services/deepgram/sagemaker/tts.py      | 2 +-
 src/pipecat/services/deepgram/stt.py                | 2 +-
 src/pipecat/services/deepgram/tts.py                | 4 ++--
 src/pipecat/services/elevenlabs/stt.py              | 4 ++--
 src/pipecat/services/elevenlabs/tts.py              | 4 ++--
 src/pipecat/services/fish/tts.py                    | 2 +-
 src/pipecat/services/gladia/stt.py                  | 2 +-
 src/pipecat/services/google/stt.py                  | 2 +-
 src/pipecat/services/gradium/stt.py                 | 2 +-
 src/pipecat/services/gradium/tts.py                 | 2 +-
 src/pipecat/services/inworld/tts.py                 | 4 ++--
 src/pipecat/services/lmnt/tts.py                    | 2 +-
 src/pipecat/services/mistral/stt.py                 | 2 +-
 src/pipecat/services/neuphonic/tts.py               | 4 ++--
 src/pipecat/services/nvidia/stt.py                  | 4 ++--
 src/pipecat/services/nvidia/tts.py                  | 2 +-
 src/pipecat/services/openai/stt.py                  | 2 +-
 src/pipecat/services/resembleai/tts.py              | 2 +-
 src/pipecat/services/rime/tts.py                    | 6 +++---
 src/pipecat/services/sarvam/stt.py                  | 2 +-
 src/pipecat/services/sarvam/tts.py                  | 4 ++--
 src/pipecat/services/smallest/stt.py                | 2 +-
 src/pipecat/services/smallest/tts.py                | 2 +-
 src/pipecat/services/soniox/stt.py                  | 2 +-
 src/pipecat/services/speechmatics/stt.py            | 2 +-
 src/pipecat/services/stt_service.py                 | 2 +-
 src/pipecat/services/tts_service.py                 | 2 +-
 src/pipecat/services/xai/stt.py                     | 2 +-
 src/pipecat/services/xai/tts.py                     | 4 ++--
 38 files changed, 50 insertions(+), 50 deletions(-)

diff --git a/src/pipecat/services/assemblyai/stt.py b/src/pipecat/services/assemblyai/stt.py
index dbaa61c86..6b2607587 100644
--- a/src/pipecat/services/assemblyai/stt.py
+++ b/src/pipecat/services/assemblyai/stt.py
@@ -418,7 +418,7 @@ class AssemblyAISTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text conversion.
 
         Args:
diff --git a/src/pipecat/services/asyncai/tts.py b/src/pipecat/services/asyncai/tts.py
index 3d8750f0c..0f72b3b9c 100644
--- a/src/pipecat/services/asyncai/tts.py
+++ b/src/pipecat/services/asyncai/tts.py
@@ -449,7 +449,7 @@ class AsyncAITTSService(WebsocketTTSService):
         await super().on_audio_context_completed(context_id)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Async API websocket endpoint.
 
         Args:
@@ -620,7 +620,7 @@ class AsyncAIHttpTTSService(TTSService):
         self._output_sample_rate = self.sample_rate
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Async's HTTP streaming API.
 
         Args:
diff --git a/src/pipecat/services/aws/stt.py b/src/pipecat/services/aws/stt.py
index 2c791bc97..7ff72c9fd 100644
--- a/src/pipecat/services/aws/stt.py
+++ b/src/pipecat/services/aws/stt.py
@@ -196,7 +196,7 @@ class AWSTranscribeSTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data and send to AWS Transcribe.
 
         Args:
diff --git a/src/pipecat/services/azure/stt.py b/src/pipecat/services/azure/stt.py
index 9b2247793..9740a1efe 100644
--- a/src/pipecat/services/azure/stt.py
+++ b/src/pipecat/services/azure/stt.py
@@ -191,7 +191,7 @@ class AzureSTTService(STTService):
 
         return changed
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text conversion.
 
         Feeds audio data to the Azure speech recognizer for processing.
diff --git a/src/pipecat/services/cartesia/stt.py b/src/pipecat/services/cartesia/stt.py
index 8e66eb965..191f778b3 100644
--- a/src/pipecat/services/cartesia/stt.py
+++ b/src/pipecat/services/cartesia/stt.py
@@ -277,7 +277,7 @@ class CartesiaSTTService(WebsocketSTTService):
             if self._websocket and self._websocket.state is State.OPEN:
                 await self._websocket.send("finalize")
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text transcription.
 
         Args:
diff --git a/src/pipecat/services/cartesia/tts.py b/src/pipecat/services/cartesia/tts.py
index 222939104..108ec33e6 100644
--- a/src/pipecat/services/cartesia/tts.py
+++ b/src/pipecat/services/cartesia/tts.py
@@ -660,7 +660,7 @@ class CartesiaTTSService(WebsocketTTSService):
             await self._connect_websocket()
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Cartesia's streaming API.
 
         Args:
@@ -873,7 +873,7 @@ class CartesiaHttpTTSService(TTSService):
         await self._close_session()
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Cartesia's HTTP API.
 
         Args:
diff --git a/src/pipecat/services/deepgram/flux/sagemaker/stt.py b/src/pipecat/services/deepgram/flux/sagemaker/stt.py
index 94738fc9c..1b110cf1e 100644
--- a/src/pipecat/services/deepgram/flux/sagemaker/stt.py
+++ b/src/pipecat/services/deepgram/flux/sagemaker/stt.py
@@ -222,7 +222,7 @@ class DeepgramFluxSageMakerSTTService(DeepgramFluxSTTBase):
     # Audio sending and response receiving
     # ------------------------------------------------------------------
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Deepgram Flux for transcription.
 
         Args:
diff --git a/src/pipecat/services/deepgram/flux/stt.py b/src/pipecat/services/deepgram/flux/stt.py
index 6874b1b69..df14442eb 100644
--- a/src/pipecat/services/deepgram/flux/stt.py
+++ b/src/pipecat/services/deepgram/flux/stt.py
@@ -354,7 +354,7 @@ class DeepgramFluxSTTService(DeepgramFluxSTTBase, WebsocketService):
     # Audio sending and receiving
     # ------------------------------------------------------------------
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Deepgram Flux for transcription.
 
         Transmits raw audio bytes to the Deepgram Flux API for real-time speech
diff --git a/src/pipecat/services/deepgram/sagemaker/stt.py b/src/pipecat/services/deepgram/sagemaker/stt.py
index c837dc30b..65788f4e3 100644
--- a/src/pipecat/services/deepgram/sagemaker/stt.py
+++ b/src/pipecat/services/deepgram/sagemaker/stt.py
@@ -256,7 +256,7 @@ class DeepgramSageMakerSTTService(STTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Deepgram for transcription.
 
         Args:
diff --git a/src/pipecat/services/deepgram/sagemaker/tts.py b/src/pipecat/services/deepgram/sagemaker/tts.py
index 5585992b1..171da06cf 100644
--- a/src/pipecat/services/deepgram/sagemaker/tts.py
+++ b/src/pipecat/services/deepgram/sagemaker/tts.py
@@ -325,7 +325,7 @@ class DeepgramSageMakerTTSService(TTSService):
                 logger.error(f"{self} error sending Flush message: {e}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Deepgram TTS on SageMaker.
 
         Args:
diff --git a/src/pipecat/services/deepgram/stt.py b/src/pipecat/services/deepgram/stt.py
index c9ebd49c8..864a0078e 100644
--- a/src/pipecat/services/deepgram/stt.py
+++ b/src/pipecat/services/deepgram/stt.py
@@ -514,7 +514,7 @@ class DeepgramSTTService(STTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Deepgram for transcription.
 
         Args:
diff --git a/src/pipecat/services/deepgram/tts.py b/src/pipecat/services/deepgram/tts.py
index cc7b88455..1a1908628 100644
--- a/src/pipecat/services/deepgram/tts.py
+++ b/src/pipecat/services/deepgram/tts.py
@@ -330,7 +330,7 @@ class DeepgramTTSService(WebsocketTTSService):
                 logger.error(f"{self} error sending Flush message: {e}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Deepgram's WebSocket TTS API.
 
         Args:
@@ -441,7 +441,7 @@ class DeepgramHttpTTSService(TTSService):
         return True
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Deepgram's TTS API.
 
         Args:
diff --git a/src/pipecat/services/elevenlabs/stt.py b/src/pipecat/services/elevenlabs/stt.py
index ac6c01876..7f65fd90b 100644
--- a/src/pipecat/services/elevenlabs/stt.py
+++ b/src/pipecat/services/elevenlabs/stt.py
@@ -370,7 +370,7 @@ class ElevenLabsSTTService(SegmentedSTTService):
         """Handle a transcription result with tracing."""
         await self.stop_processing_metrics()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Transcribe an audio segment using ElevenLabs' STT API.
 
         Args:
@@ -674,7 +674,7 @@ class ElevenLabsRealtimeSTTService(WebsocketSTTService):
                     except Exception as e:
                         logger.warning(f"Failed to send commit: {e}")
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text transcription.
 
         Args:
diff --git a/src/pipecat/services/elevenlabs/tts.py b/src/pipecat/services/elevenlabs/tts.py
index f72864ef4..d0380f778 100644
--- a/src/pipecat/services/elevenlabs/tts.py
+++ b/src/pipecat/services/elevenlabs/tts.py
@@ -889,7 +889,7 @@ class ElevenLabsTTSService(WebsocketTTSService):
             await self._websocket.send(json.dumps(msg))
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using ElevenLabs' streaming WebSocket API.
 
         Args:
@@ -1240,7 +1240,7 @@ class ElevenLabsHttpTTSService(TTSService):
         return word_times
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using ElevenLabs streaming API with timestamps.
 
         Makes a request to the ElevenLabs API to generate audio and timing data.
diff --git a/src/pipecat/services/fish/tts.py b/src/pipecat/services/fish/tts.py
index 5f4252fb2..64639150f 100644
--- a/src/pipecat/services/fish/tts.py
+++ b/src/pipecat/services/fish/tts.py
@@ -373,7 +373,7 @@ class FishAudioTTSService(InterruptibleTTSService):
                 await self.push_error(error_msg=f"Unknown error occurred: {e}", exception=e)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Fish Audio's streaming API.
 
         Args:
diff --git a/src/pipecat/services/gladia/stt.py b/src/pipecat/services/gladia/stt.py
index 8c57e47e2..e0c6d69c7 100644
--- a/src/pipecat/services/gladia/stt.py
+++ b/src/pipecat/services/gladia/stt.py
@@ -461,7 +461,7 @@ class GladiaSTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Run speech-to-text on audio data.
 
         Args:
diff --git a/src/pipecat/services/google/stt.py b/src/pipecat/services/google/stt.py
index 4665d6309..616627acc 100644
--- a/src/pipecat/services/google/stt.py
+++ b/src/pipecat/services/google/stt.py
@@ -931,7 +931,7 @@ class GoogleSTTService(STTService):
         except Exception as e:
             await self.push_error(error_msg=f"Unknown error occurred: {e}", exception=e)
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process an audio chunk for STT transcription.
 
         Args:
diff --git a/src/pipecat/services/gradium/stt.py b/src/pipecat/services/gradium/stt.py
index 223941bd0..70fde8145 100644
--- a/src/pipecat/services/gradium/stt.py
+++ b/src/pipecat/services/gradium/stt.py
@@ -333,7 +333,7 @@ class GradiumSTTService(WebsocketSTTService):
         except Exception as e:
             logger.warning(f"Failed to send flush: {e}")
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text conversion.
 
         Args:
diff --git a/src/pipecat/services/gradium/tts.py b/src/pipecat/services/gradium/tts.py
index 12cc3ddc1..06b8c2743 100644
--- a/src/pipecat/services/gradium/tts.py
+++ b/src/pipecat/services/gradium/tts.py
@@ -356,7 +356,7 @@ class GradiumTTSService(WebsocketTTSService):
                 await self.push_error(error_msg=f"Error: {msg.get('message', msg)}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Gradium's streaming API.
 
         Args:
diff --git a/src/pipecat/services/inworld/tts.py b/src/pipecat/services/inworld/tts.py
index 65922e3a1..93653e61f 100644
--- a/src/pipecat/services/inworld/tts.py
+++ b/src/pipecat/services/inworld/tts.py
@@ -283,7 +283,7 @@ class InworldHttpTTSService(TTSService):
         return (word_times, chunk_end_time)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate TTS audio for the given text.
 
         Args:
@@ -1128,7 +1128,7 @@ class InworldTTSService(WebsocketTTSService):
         await self.send_with_retry(json.dumps(msg), self._report_error)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate TTS audio for the given text using the Inworld WebSocket TTS service.
 
         Args:
diff --git a/src/pipecat/services/lmnt/tts.py b/src/pipecat/services/lmnt/tts.py
index 8daa1902a..b098b43c2 100644
--- a/src/pipecat/services/lmnt/tts.py
+++ b/src/pipecat/services/lmnt/tts.py
@@ -336,7 +336,7 @@ class LmntTTSService(InterruptibleTTSService):
                     logger.error(f"Invalid JSON message: {message}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate TTS audio from text using LMNT's streaming API.
 
         Args:
diff --git a/src/pipecat/services/mistral/stt.py b/src/pipecat/services/mistral/stt.py
index 3200bc76e..9d8b965e5 100644
--- a/src/pipecat/services/mistral/stt.py
+++ b/src/pipecat/services/mistral/stt.py
@@ -185,7 +185,7 @@ class MistralSTTService(STTService):
             if self._connection and not self._connection.is_closed:
                 await self._connection.flush_audio()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Mistral for transcription.
 
         Args:
diff --git a/src/pipecat/services/neuphonic/tts.py b/src/pipecat/services/neuphonic/tts.py
index 5ee19fb3b..a7303da6b 100644
--- a/src/pipecat/services/neuphonic/tts.py
+++ b/src/pipecat/services/neuphonic/tts.py
@@ -366,7 +366,7 @@ class NeuphonicTTSService(InterruptibleTTSService):
             await self._websocket.send(json.dumps(msg))
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Neuphonic's streaming API.
 
         Args:
@@ -565,7 +565,7 @@ class NeuphonicHttpTTSService(TTSService):
             return None
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Neuphonic streaming API.
 
         Args:
diff --git a/src/pipecat/services/nvidia/stt.py b/src/pipecat/services/nvidia/stt.py
index 3181b53ba..d81411500 100644
--- a/src/pipecat/services/nvidia/stt.py
+++ b/src/pipecat/services/nvidia/stt.py
@@ -395,7 +395,7 @@ class NvidiaSTTService(STTService):
                         )
                     )
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Process audio data for speech-to-text transcription.
 
         Args:
@@ -661,7 +661,7 @@ class NvidiaSegmentedSTTService(SegmentedSTTService):
         """Handle a transcription result with tracing."""
         pass
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Transcribe an audio segment.
 
         Args:
diff --git a/src/pipecat/services/nvidia/tts.py b/src/pipecat/services/nvidia/tts.py
index 0edc0c5f0..9d183a032 100644
--- a/src/pipecat/services/nvidia/tts.py
+++ b/src/pipecat/services/nvidia/tts.py
@@ -526,7 +526,7 @@ class NvidiaTTSService(TTSService):
         )
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using NVIDIA Nemotron Speech TTS.
 
         On the first call for a turn, starts a persistent ``synthesize_online``
diff --git a/src/pipecat/services/openai/stt.py b/src/pipecat/services/openai/stt.py
index ca0537fef..f550d4395 100644
--- a/src/pipecat/services/openai/stt.py
+++ b/src/pipecat/services/openai/stt.py
@@ -415,7 +415,7 @@ class OpenAIRealtimeSTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to the transcription session.
 
         Audio is streamed over the WebSocket. Transcription results arrive
diff --git a/src/pipecat/services/resembleai/tts.py b/src/pipecat/services/resembleai/tts.py
index 595b295e3..3053e0b70 100644
--- a/src/pipecat/services/resembleai/tts.py
+++ b/src/pipecat/services/resembleai/tts.py
@@ -431,7 +431,7 @@ class ResembleAITTSService(WebsocketTTSService):
                 await self._connect_websocket()
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Resemble AI's streaming API.
 
         Args:
diff --git a/src/pipecat/services/rime/tts.py b/src/pipecat/services/rime/tts.py
index 2745f2cf4..7ac15475b 100644
--- a/src/pipecat/services/rime/tts.py
+++ b/src/pipecat/services/rime/tts.py
@@ -603,7 +603,7 @@ class RimeTTSService(WebsocketTTSService):
                 self.reset_active_audio_context()
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Rime's streaming API.
 
         Args:
@@ -786,7 +786,7 @@ class RimeHttpTTSService(TTSService):
         return language_to_rime_language(language)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Rime's HTTP API.
 
         Args:
@@ -1142,7 +1142,7 @@ class RimeNonJsonTTSService(InterruptibleTTSService):
                 await self.push_error(error_msg=f"Error: {e}", exception=e)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Rime's streaming API.
 
         Args:
diff --git a/src/pipecat/services/sarvam/stt.py b/src/pipecat/services/sarvam/stt.py
index 7c949ac59..f41ec0faf 100644
--- a/src/pipecat/services/sarvam/stt.py
+++ b/src/pipecat/services/sarvam/stt.py
@@ -570,7 +570,7 @@ class SarvamSTTService(STTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Sarvam for transcription.
 
         Args:
diff --git a/src/pipecat/services/sarvam/tts.py b/src/pipecat/services/sarvam/tts.py
index 01b62b87d..1506b84dc 100644
--- a/src/pipecat/services/sarvam/tts.py
+++ b/src/pipecat/services/sarvam/tts.py
@@ -569,7 +569,7 @@ class SarvamHttpTTSService(TTSService):
         await super().start(frame)
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Sarvam AI's API.
 
         Args:
@@ -1192,7 +1192,7 @@ class SarvamTTSService(InterruptibleTTSService):
             logger.warning("WebSocket not ready, cannot send text")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech audio frames from input text using Sarvam TTS.
 
         Sends text over WebSocket for synthesis and yields corresponding audio or status frames.
diff --git a/src/pipecat/services/smallest/stt.py b/src/pipecat/services/smallest/stt.py
index 8673bd040..57017c76d 100644
--- a/src/pipecat/services/smallest/stt.py
+++ b/src/pipecat/services/smallest/stt.py
@@ -247,7 +247,7 @@ class SmallestSTTService(WebsocketSTTService):
                 except Exception as e:
                     logger.warning(f"{self} failed to send finalize: {e}")
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio to the Smallest Pulse WebSocket for transcription.
 
         Args:
diff --git a/src/pipecat/services/smallest/tts.py b/src/pipecat/services/smallest/tts.py
index 8bc22035c..6a2071f44 100644
--- a/src/pipecat/services/smallest/tts.py
+++ b/src/pipecat/services/smallest/tts.py
@@ -390,7 +390,7 @@ class SmallestTTSService(InterruptibleTTSService):
                 logger.warning(f"{self} unknown message status: {msg}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using Smallest's WebSocket streaming API.
 
         Args:
diff --git a/src/pipecat/services/soniox/stt.py b/src/pipecat/services/soniox/stt.py
index 66f27d40f..30c813c9d 100644
--- a/src/pipecat/services/soniox/stt.py
+++ b/src/pipecat/services/soniox/stt.py
@@ -402,7 +402,7 @@ class SonioxSTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Send audio data to Soniox STT Service.
 
         Args:
diff --git a/src/pipecat/services/speechmatics/stt.py b/src/pipecat/services/speechmatics/stt.py
index 6f049b29f..1126fef37 100644
--- a/src/pipecat/services/speechmatics/stt.py
+++ b/src/pipecat/services/speechmatics/stt.py
@@ -1059,7 +1059,7 @@ class SpeechmaticsSTTService(STTService):
         """Record transcription event for tracing."""
         pass
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Adds audio to the audio buffer and yields None."""
         try:
             if self._client:
diff --git a/src/pipecat/services/stt_service.py b/src/pipecat/services/stt_service.py
index 74d2f90ea..888c4c135 100644
--- a/src/pipecat/services/stt_service.py
+++ b/src/pipecat/services/stt_service.py
@@ -274,7 +274,7 @@ class STTService(AIService):
         return Language(language)
 
     @abstractmethod
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Run speech-to-text on the provided audio data.
 
         This method must be implemented by subclasses to provide actual speech
diff --git a/src/pipecat/services/tts_service.py b/src/pipecat/services/tts_service.py
index 65520129f..c67e47d19 100644
--- a/src/pipecat/services/tts_service.py
+++ b/src/pipecat/services/tts_service.py
@@ -445,7 +445,7 @@ class TTSService(AIService):
 
     # Converts the text to audio.
     @abstractmethod
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Run text-to-speech synthesis on the provided text.
 
         This method must be implemented by subclasses to provide actual TTS functionality.
diff --git a/src/pipecat/services/xai/stt.py b/src/pipecat/services/xai/stt.py
index df8406fc2..3cc3a23b2 100644
--- a/src/pipecat/services/xai/stt.py
+++ b/src/pipecat/services/xai/stt.py
@@ -209,7 +209,7 @@ class XAISTTService(WebsocketSTTService):
         await super().cancel(frame)
         await self._disconnect()
 
-    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
+    async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame | None, None]:
         """Forward raw audio bytes to the xAI STT WebSocket.
 
         Transcription frames are pushed from the receive task, not yielded
diff --git a/src/pipecat/services/xai/tts.py b/src/pipecat/services/xai/tts.py
index f99772baa..83e7781f2 100644
--- a/src/pipecat/services/xai/tts.py
+++ b/src/pipecat/services/xai/tts.py
@@ -188,7 +188,7 @@ class XAIHttpTTSService(TTSService):
             self._session = None
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate speech from text using xAI's TTS API."""
         logger.debug(f"{self}: Generating TTS [{text}]")
 
@@ -466,7 +466,7 @@ class XAITTSService(InterruptibleTTSService):
                 logger.debug(f"{self}: unhandled xAI message type: {msg_type}")
 
     @traced_tts
-    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame, None]:
+    async def run_tts(self, text: str, context_id: str) -> AsyncGenerator[Frame | None, None]:
         """Generate TTS audio from text using xAI's streaming WebSocket API."""
         logger.debug(f"{self}: Generating TTS [{text}]")
 

From 3b0affe5b4006a4c3869fa1c0b2e9c55a9519936 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:03:41 -0400
Subject: [PATCH 02/12] Guard run_stt WebSocket sends with try/except

AssemblyAI, Cartesia, Gradium, and Soniox STT services sent audio over
the WebSocket without catching transient send failures, so a single
network hiccup could propagate an exception up through process_frame
and end the pipeline. Other push-based STT services (Deepgram, xAI,
Azure, Smallest, etc.) already guard their sends.

Follow the deepgram/stt.py pattern: log a warning and continue. The
existing connection-state check at the top of each call handles
recovery on the next invocation.
---
 src/pipecat/services/assemblyai/stt.py | 6 +++++-
 src/pipecat/services/cartesia/stt.py   | 5 ++++-
 src/pipecat/services/gradium/stt.py    | 6 +++++-
 src/pipecat/services/soniox/stt.py     | 5 ++++-
 4 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/src/pipecat/services/assemblyai/stt.py b/src/pipecat/services/assemblyai/stt.py
index 6b2607587..f87a5a0e5 100644
--- a/src/pipecat/services/assemblyai/stt.py
+++ b/src/pipecat/services/assemblyai/stt.py
@@ -433,7 +433,11 @@ class AssemblyAISTTService(WebsocketSTTService):
             while len(self._audio_buffer) >= self._chunk_size_bytes:
                 chunk = bytes(self._audio_buffer[: self._chunk_size_bytes])
                 self._audio_buffer = self._audio_buffer[self._chunk_size_bytes :]
-                await self._websocket.send(chunk)
+                try:
+                    await self._websocket.send(chunk)
+                except Exception as e:
+                    logger.warning(f"{self}: send failed: {e}")
+                    break
 
         yield None
 
diff --git a/src/pipecat/services/cartesia/stt.py b/src/pipecat/services/cartesia/stt.py
index 191f778b3..018d95e6a 100644
--- a/src/pipecat/services/cartesia/stt.py
+++ b/src/pipecat/services/cartesia/stt.py
@@ -290,7 +290,10 @@ class CartesiaSTTService(WebsocketSTTService):
         if not self._websocket or self._websocket.state is not State.OPEN:
             await self._connect()
 
-        await self._websocket.send(audio)
+        try:
+            await self._websocket.send(audio)
+        except Exception as e:
+            logger.warning(f"{self}: send failed: {e}")
         yield None
 
     async def _connect(self):
diff --git a/src/pipecat/services/gradium/stt.py b/src/pipecat/services/gradium/stt.py
index 70fde8145..412f0a1b3 100644
--- a/src/pipecat/services/gradium/stt.py
+++ b/src/pipecat/services/gradium/stt.py
@@ -350,7 +350,11 @@ class GradiumSTTService(WebsocketSTTService):
             chunk = base64.b64encode(chunk).decode("utf-8")
             msg = {"type": "audio", "audio": chunk}
             if self._websocket and self._websocket.state is State.OPEN:
-                await self._websocket.send(json.dumps(msg))
+                try:
+                    await self._websocket.send(json.dumps(msg))
+                except Exception as e:
+                    logger.warning(f"{self}: send failed: {e}")
+                    break
 
         yield None
 
diff --git a/src/pipecat/services/soniox/stt.py b/src/pipecat/services/soniox/stt.py
index 30c813c9d..95418011c 100644
--- a/src/pipecat/services/soniox/stt.py
+++ b/src/pipecat/services/soniox/stt.py
@@ -412,7 +412,10 @@ class SonioxSTTService(WebsocketSTTService):
             Frame: None (transcription results come via WebSocket callbacks).
         """
         if self._websocket and self._websocket.state is State.OPEN:
-            await self._websocket.send(audio)
+            try:
+                await self._websocket.send(audio)
+            except Exception as e:
+                logger.warning(f"{self}: send failed: {e}")
 
         yield None
 

From 14cd476b20e5b7ad9f9d08baaa8bedc7f7b4df2d Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:09:27 -0400
Subject: [PATCH 03/12] Drop pyright ignores for services fixed by
 run_stt/run_tts widening

Deepgram STT, Gradium TTS, Smallest STT, and xAI STT/TTS had exactly
one pyright error each, all of them the AsyncGenerator return-type
mismatch resolved in 08fe9157c. Remove them from the ignore list.
---
 pyrightconfig.json | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 74682c3dc..883c7276a 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -57,7 +57,6 @@
     "src/pipecat/services/deepgram/flux/stt.py",
     "src/pipecat/services/deepgram/sagemaker/stt.py",
     "src/pipecat/services/deepgram/sagemaker/tts.py",
-    "src/pipecat/services/deepgram/stt.py",
     "src/pipecat/services/deepgram/tts.py",
     "src/pipecat/services/elevenlabs/stt.py",
     "src/pipecat/services/elevenlabs/tts.py",
@@ -71,7 +70,6 @@
     "src/pipecat/services/google/tts.py",
     "src/pipecat/services/google/vertex/llm.py",
     "src/pipecat/services/gradium/stt.py",
-    "src/pipecat/services/gradium/tts.py",
     "src/pipecat/services/groq/tts.py",
     "src/pipecat/services/heygen/api_interactive_avatar.py",
     "src/pipecat/services/heygen/base_api.py",
@@ -109,7 +107,6 @@
     "src/pipecat/services/sarvam/stt.py",
     "src/pipecat/services/sarvam/tts.py",
     "src/pipecat/services/simli/video.py",
-    "src/pipecat/services/smallest/stt.py",
     "src/pipecat/services/smallest/tts.py",
     "src/pipecat/services/soniox/stt.py",
     "src/pipecat/services/speechmatics/stt.py",
@@ -123,8 +120,6 @@
     "src/pipecat/services/whisper/stt.py",
     "src/pipecat/services/xai/realtime/events.py",
     "src/pipecat/services/xai/realtime/llm.py",
-    "src/pipecat/services/xai/stt.py",
-    "src/pipecat/services/xai/tts.py",
     "src/pipecat/services/xtts/tts.py",
     "src/pipecat/transports/base_output.py",
     "src/pipecat/transports/daily/transport.py",

From 457eb7aa9260f5d1292d59fb201a478927a25c72 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:19:23 -0400
Subject: [PATCH 04/12] Mark abstract image/vision generators as real async
 generators

`ImageGenService.run_image_gen` and `VisionService.run_vision` were
declared `async def ... -> AsyncGenerator[Frame, None]` with `pass`
bodies. Without a `yield` anywhere in the body, Python treats the
function as a coroutine returning an `AsyncGenerator`, not as an async
generator itself, so callers got a coroutine where they expected an
iterator.

Add `raise NotImplementedError; yield` so the body contains a yield
(making this a real async generator) while still raising cleanly if a
subclass ever calls `super().run_*` by mistake.
---
 pyrightconfig.json                     | 2 --
 src/pipecat/services/image_service.py  | 3 ++-
 src/pipecat/services/vision_service.py | 3 ++-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 883c7276a..0f10db1b6 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -76,7 +76,6 @@
     "src/pipecat/services/heygen/client.py",
     "src/pipecat/services/heygen/video.py",
     "src/pipecat/services/hume/tts.py",
-    "src/pipecat/services/image_service.py",
     "src/pipecat/services/inworld/realtime/llm.py",
     "src/pipecat/services/inworld/tts.py",
     "src/pipecat/services/kokoro/tts.py",
@@ -115,7 +114,6 @@
     "src/pipecat/services/tavus/video.py",
     "src/pipecat/services/tts_service.py",
     "src/pipecat/services/ultravox/llm.py",
-    "src/pipecat/services/vision_service.py",
     "src/pipecat/services/websocket_service.py",
     "src/pipecat/services/whisper/stt.py",
     "src/pipecat/services/xai/realtime/events.py",
diff --git a/src/pipecat/services/image_service.py b/src/pipecat/services/image_service.py
index df8ef66fe..28ab8d98b 100644
--- a/src/pipecat/services/image_service.py
+++ b/src/pipecat/services/image_service.py
@@ -57,7 +57,8 @@ class ImageGenService(AIService):
             Frame: Frames containing the generated image (typically ImageRawFrame
                 or URLImageRawFrame).
         """
-        pass
+        raise NotImplementedError
+        yield  # pragma: no cover
 
     async def process_frame(self, frame: Frame, direction: FrameDirection):
         """Process frames for image generation.
diff --git a/src/pipecat/services/vision_service.py b/src/pipecat/services/vision_service.py
index 74d70f1d4..22e528042 100644
--- a/src/pipecat/services/vision_service.py
+++ b/src/pipecat/services/vision_service.py
@@ -59,7 +59,8 @@ class VisionService(AIService):
             Frame: Frames containing the vision analysis results, typically TextFrame
             objects with descriptions or answers.
         """
-        pass
+        raise NotImplementedError
+        yield  # pragma: no cover
 
     async def process_frame(self, frame: Frame, direction: FrameDirection):
         """Process frames, handling vision image frames for analysis.

From 5872006d6b1251cfab72d2166d52d7534bde0c85 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:27:25 -0400
Subject: [PATCH 05/12] Encode lazy-init invariants at the right site, not at
 read sites

Three spots had the same shape: a field starts None, a later method
populates it, a read site later reads it. Pyright can't track the
cross-method invariant. Rather than spray assertions at the read
sites, fix each site at the structural level:

- `FastAPIWebsocketInputTransport._monitor_websocket` now takes the
  session timeout as an argument. The task-creation site already
  guards on truthiness, so the call can pass the non-None value
  directly and the method's signature tells the truth.
- `FrameProcessorMetrics.task_manager` raises `RuntimeError` instead
  of asserting. Asserts are stripped under `python -O`; a real raise
  keeps the runtime safety net and still narrows the type for pyright.
- `SOXRStreamAudioResampler._maybe_initialize_sox_stream` returns the
  initialized stream. Callers use the return value and never touch
  the Optional `_soxr_stream` attribute, so narrowing stays inside
  the init method where the invariant is established.
---
 pyrightconfig.json                                | 15 +++++++++------
 .../audio/resamplers/soxr_stream_resampler.py     |  9 ++++++---
 .../processors/metrics/frame_processor_metrics.py |  2 ++
 src/pipecat/transports/websocket/fastapi.py       | 10 ++++++----
 4 files changed, 23 insertions(+), 13 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 0f10db1b6..0a019fc4f 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -2,8 +2,14 @@
   "typeCheckingMode": "basic",
   "pythonVersion": "3.11",
   "pythonPlatform": "All",
-  "include": ["scripts", "src/pipecat"],
-  "exclude": ["**/*_pb2.py", "**/__pycache__"],
+  "include": [
+    "scripts",
+    "src/pipecat"
+  ],
+  "exclude": [
+    "**/*_pb2.py",
+    "**/__pycache__"
+  ],
   "ignore": [
     "tests",
     "src/pipecat/adapters/base_llm_adapter.py",
@@ -23,7 +29,6 @@
     "src/pipecat/audio/filters/aic_filter.py",
     "src/pipecat/audio/filters/krisp_viva_filter.py",
     "src/pipecat/audio/filters/rnnoise_filter.py",
-    "src/pipecat/audio/resamplers/soxr_stream_resampler.py",
     "src/pipecat/audio/turn/smart_turn/local_smart_turn_v2.py",
     "src/pipecat/audio/turn/smart_turn/local_smart_turn_v3.py",
     "src/pipecat/audio/vad/silero.py",
@@ -35,7 +40,6 @@
     "src/pipecat/processors/frameworks/rtvi/processor.py",
     "src/pipecat/processors/frameworks/strands_agents.py",
     "src/pipecat/processors/gstreamer/pipeline_source.py",
-    "src/pipecat/processors/metrics/frame_processor_metrics.py",
     "src/pipecat/services/ai_service.py",
     "src/pipecat/services/anthropic/llm.py",
     "src/pipecat/services/assemblyai/stt.py",
@@ -130,9 +134,8 @@
     "src/pipecat/transports/smallwebrtc/transport.py",
     "src/pipecat/transports/tavus/transport.py",
     "src/pipecat/transports/websocket/client.py",
-    "src/pipecat/transports/websocket/fastapi.py",
     "src/pipecat/transports/websocket/server.py",
     "src/pipecat/transports/whatsapp/client.py"
   ],
   "reportMissingImports": false
-}
+}
\ No newline at end of file
diff --git a/src/pipecat/audio/resamplers/soxr_stream_resampler.py b/src/pipecat/audio/resamplers/soxr_stream_resampler.py
index 35d42142f..33045db59 100644
--- a/src/pipecat/audio/resamplers/soxr_stream_resampler.py
+++ b/src/pipecat/audio/resamplers/soxr_stream_resampler.py
@@ -68,7 +68,7 @@ class SOXRStreamAudioResampler(BaseAudioResampler):
                 self._soxr_stream.clear()
         self._last_resample_time = current_time
 
-    def _maybe_initialize_sox_stream(self, in_rate: int, out_rate: int):
+    def _maybe_initialize_sox_stream(self, in_rate: int, out_rate: int) -> "soxr.ResampleStream":
         if self._soxr_stream is None:
             self._initialize(in_rate, out_rate)
         else:
@@ -80,6 +80,9 @@ class SOXRStreamAudioResampler(BaseAudioResampler):
                 f"expected {self._in_rate}->{self._out_rate}, got {in_rate}->{out_rate}"
             )
 
+        assert self._soxr_stream is not None
+        return self._soxr_stream
+
     async def resample(self, audio: bytes, in_rate: int, out_rate: int) -> bytes:
         """Resample audio data using soxr.ResampleStream resampler library.
 
@@ -94,8 +97,8 @@ class SOXRStreamAudioResampler(BaseAudioResampler):
         if in_rate == out_rate:
             return audio
 
-        self._maybe_initialize_sox_stream(in_rate, out_rate)
+        stream = self._maybe_initialize_sox_stream(in_rate, out_rate)
         audio_data = np.frombuffer(audio, dtype=np.int16)
-        resampled_audio = self._soxr_stream.resample_chunk(audio_data)
+        resampled_audio = stream.resample_chunk(audio_data)
         result = resampled_audio.astype(np.int16).tobytes()
         return result
diff --git a/src/pipecat/processors/metrics/frame_processor_metrics.py b/src/pipecat/processors/metrics/frame_processor_metrics.py
index 18ef7f580..97098ffed 100644
--- a/src/pipecat/processors/metrics/frame_processor_metrics.py
+++ b/src/pipecat/processors/metrics/frame_processor_metrics.py
@@ -66,6 +66,8 @@ class FrameProcessorMetrics(BaseObject):
         Returns:
             The task manager instance for async operations.
         """
+        if self._task_manager is None:
+            raise RuntimeError("task_manager not set; call setup() first")
         return self._task_manager
 
     @property
diff --git a/src/pipecat/transports/websocket/fastapi.py b/src/pipecat/transports/websocket/fastapi.py
index 1e449b01e..28bf73822 100644
--- a/src/pipecat/transports/websocket/fastapi.py
+++ b/src/pipecat/transports/websocket/fastapi.py
@@ -255,7 +255,9 @@ class FastAPIWebsocketInputTransport(BaseInputTransport):
         if self._params.serializer:
             await self._params.serializer.setup(frame)
         if not self._monitor_websocket_task and self._params.session_timeout:
-            self._monitor_websocket_task = self.create_task(self._monitor_websocket())
+            self._monitor_websocket_task = self.create_task(
+                self._monitor_websocket(self._params.session_timeout)
+            )
         await self._client.trigger_client_connected()
         await self.push_frame(ClientConnectedFrame())
         if not self._receive_task:
@@ -322,9 +324,9 @@ class FastAPIWebsocketInputTransport(BaseInputTransport):
         if not self._client.is_closing:
             await self._client.trigger_client_disconnected()
 
-    async def _monitor_websocket(self):
-        """Wait for self._params.session_timeout seconds, if the websocket is still open, trigger timeout event."""
-        await asyncio.sleep(self._params.session_timeout)
+    async def _monitor_websocket(self, timeout: int):
+        """Wait for ``timeout`` seconds, then trigger the client-timeout event if still open."""
+        await asyncio.sleep(timeout)
         await self._client.trigger_client_timeout()
 
 

From b64ed3f9e2ea499f59eeda47ca654b6c30013a63 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:50:51 -0400
Subject: [PATCH 06/12] Narrow settings.model at service boundaries, not via
 truthiness
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Two services were reading `_settings.model` (typed `str | _NotGiven |
None` because NOT_GIVEN is the default) and coercing it with `or ""`
or similar. `_NotGiven.__bool__` returns False, so the runtime
behavior happened to work, but the type was a lie — pyright saw
`str | _NotGiven` flowing into APIs that required `str` or `str | None`.

- `AIService._sync_model_name_to_metrics`: use `isinstance(model, str)`
  narrowing with an empty-string fallback. Equivalent runtime behavior,
  honest type, no truthiness dependency on a sentinel.
- `SarvamLLMService.__init__`: validate the model is a real string
  before handing it to `_validate_model(str)`. A non-string model at
  this point is a configuration bug; raise `ValueError` so the error
  is clear and survives `python -O` (unlike an assert).
---
 pyrightconfig.json                 | 14 +++-----------
 src/pipecat/services/ai_service.py |  3 ++-
 src/pipecat/services/sarvam/llm.py |  5 ++++-
 3 files changed, 9 insertions(+), 13 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 0a019fc4f..d087425bf 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -2,14 +2,8 @@
   "typeCheckingMode": "basic",
   "pythonVersion": "3.11",
   "pythonPlatform": "All",
-  "include": [
-    "scripts",
-    "src/pipecat"
-  ],
-  "exclude": [
-    "**/*_pb2.py",
-    "**/__pycache__"
-  ],
+  "include": ["scripts", "src/pipecat"],
+  "exclude": ["**/*_pb2.py", "**/__pycache__"],
   "ignore": [
     "tests",
     "src/pipecat/adapters/base_llm_adapter.py",
@@ -40,7 +34,6 @@
     "src/pipecat/processors/frameworks/rtvi/processor.py",
     "src/pipecat/processors/frameworks/strands_agents.py",
     "src/pipecat/processors/gstreamer/pipeline_source.py",
-    "src/pipecat/services/ai_service.py",
     "src/pipecat/services/anthropic/llm.py",
     "src/pipecat/services/assemblyai/stt.py",
     "src/pipecat/services/asyncai/tts.py",
@@ -106,7 +99,6 @@
     "src/pipecat/services/resembleai/tts.py",
     "src/pipecat/services/rime/tts.py",
     "src/pipecat/services/sambanova/llm.py",
-    "src/pipecat/services/sarvam/llm.py",
     "src/pipecat/services/sarvam/stt.py",
     "src/pipecat/services/sarvam/tts.py",
     "src/pipecat/services/simli/video.py",
@@ -138,4 +130,4 @@
     "src/pipecat/transports/whatsapp/client.py"
   ],
   "reportMissingImports": false
-}
\ No newline at end of file
+}
diff --git a/src/pipecat/services/ai_service.py b/src/pipecat/services/ai_service.py
index 5d914dd00..6e4064976 100644
--- a/src/pipecat/services/ai_service.py
+++ b/src/pipecat/services/ai_service.py
@@ -66,8 +66,9 @@ class AIService(FrameProcessor):
         Args:
             model: The name of the AI model to use.
         """
+        model = self._settings.model
         self.set_core_metrics_data(
-            MetricsData(processor=self.name, model=self._settings.model or "")
+            MetricsData(processor=self.name, model=model if isinstance(model, str) else "")
         )
 
     async def start(self, frame: StartFrame):
diff --git a/src/pipecat/services/sarvam/llm.py b/src/pipecat/services/sarvam/llm.py
index d86ba1874..224b2da43 100644
--- a/src/pipecat/services/sarvam/llm.py
+++ b/src/pipecat/services/sarvam/llm.py
@@ -83,7 +83,10 @@ class SarvamLLMService(OpenAILLMService):
         if settings is not None:
             default_settings.apply_update(settings)
 
-        self._validate_model(default_settings.model)
+        model = default_settings.model
+        if not isinstance(model, str):
+            raise ValueError("Sarvam LLM requires a non-empty model string.")
+        self._validate_model(model)
 
         super().__init__(
             api_key=api_key,

From 0c3c5e5c7d3a3dc499eaea81c082d19b8ec1e57d Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:54:20 -0400
Subject: [PATCH 07/12] Widen ToolsSchema.standard_tools to Sequence for
 covariance
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

`ToolsSchema.__init__` declared `standard_tools: list[FunctionSchema |
DirectFunction]`. Callers (`BaseLLMAdapter`, `MCPService`) pass in
`list[FunctionSchema]`, which is not assignable to the union list
because `list` is invariant in its element type.

Widen the parameter to `Sequence[...]` (covariant) so `list[X]` and
`list[X | Y]` both fit. A narrower `list[FunctionSchema]` is still
accepted, and nothing in this class mutates the argument — the
constructor immediately copies it via `_map_standard_tools`.

Also correct the `custom_tools` property return type to include
`None`, matching the stored `_custom_tools` field.

This single edit clears the pyright errors for three ignore-list
entries: `tools_schema.py`, `base_llm_adapter.py`, and `mcp_service.py`.
---
 pyrightconfig.json                           | 3 ---
 src/pipecat/adapters/schemas/tools_schema.py | 5 +++--
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index d087425bf..75d7ca777 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -6,9 +6,7 @@
   "exclude": ["**/*_pb2.py", "**/__pycache__"],
   "ignore": [
     "tests",
-    "src/pipecat/adapters/base_llm_adapter.py",
     "src/pipecat/adapters/schemas/direct_function.py",
-    "src/pipecat/adapters/schemas/tools_schema.py",
     "src/pipecat/adapters/services/anthropic_adapter.py",
     "src/pipecat/adapters/services/aws_nova_sonic_adapter.py",
     "src/pipecat/adapters/services/bedrock_adapter.py",
@@ -78,7 +76,6 @@
     "src/pipecat/services/kokoro/tts.py",
     "src/pipecat/services/llm_service.py",
     "src/pipecat/services/lmnt/tts.py",
-    "src/pipecat/services/mcp_service.py",
     "src/pipecat/services/mem0/memory.py",
     "src/pipecat/services/mistral/llm.py",
     "src/pipecat/services/mistral/stt.py",
diff --git a/src/pipecat/adapters/schemas/tools_schema.py b/src/pipecat/adapters/schemas/tools_schema.py
index 28c2b9b88..147014f16 100644
--- a/src/pipecat/adapters/schemas/tools_schema.py
+++ b/src/pipecat/adapters/schemas/tools_schema.py
@@ -10,6 +10,7 @@ This module provides schemas for managing both standardized function tools
 and custom adapter-specific tools in the Pipecat framework.
 """
 
+from collections.abc import Sequence
 from enum import Enum
 from typing import Any
 
@@ -39,7 +40,7 @@ class ToolsSchema:
 
     def __init__(
         self,
-        standard_tools: list[FunctionSchema | DirectFunction],
+        standard_tools: Sequence[FunctionSchema | DirectFunction],
         custom_tools: dict[AdapterType, list[dict[str, Any]]] | None = None,
     ) -> None:
         """Initialize the tools schema.
@@ -75,7 +76,7 @@ class ToolsSchema:
         return self._standard_tools
 
     @property
-    def custom_tools(self) -> dict[AdapterType, list[dict[str, Any]]]:
+    def custom_tools(self) -> dict[AdapterType, list[dict[str, Any]]] | None:
         """Get the custom tools dictionary.
 
         Returns:

From 8ec56092c0e6073fdf933e3c991b5ec91d85c5ac Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 11:58:15 -0400
Subject: [PATCH 08/12] Remove duplicate ResponseCreated type

---
 pyrightconfig.json                          |  1 -
 src/pipecat/services/xai/realtime/events.py | 12 ------------
 2 files changed, 13 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 75d7ca777..23ed1f3f0 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -109,7 +109,6 @@
     "src/pipecat/services/ultravox/llm.py",
     "src/pipecat/services/websocket_service.py",
     "src/pipecat/services/whisper/stt.py",
-    "src/pipecat/services/xai/realtime/events.py",
     "src/pipecat/services/xai/realtime/llm.py",
     "src/pipecat/services/xtts/tts.py",
     "src/pipecat/transports/base_output.py",
diff --git a/src/pipecat/services/xai/realtime/events.py b/src/pipecat/services/xai/realtime/events.py
index c5e4ab755..7e1701db9 100644
--- a/src/pipecat/services/xai/realtime/events.py
+++ b/src/pipecat/services/xai/realtime/events.py
@@ -541,18 +541,6 @@ class InputAudioBufferCleared(ServerEvent):
     type: Literal["input_audio_buffer.cleared"]
 
 
-class ResponseCreated(ServerEvent):
-    """Event indicating an assistant response has been created.
-
-    Parameters:
-        type: Event type, always "response.created".
-        response: The created response object.
-    """
-
-    type: Literal["response.created"]
-    response: "Response"
-
-
 class ResponseOutputItemAdded(ServerEvent):
     """Event indicating an output item has been added to a response.
 

From 10b86b4bbebf1e58d67d8099f9bae28c2a2c89f6 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 12:01:00 -0400
Subject: [PATCH 09/12] Coerce inspect.getdoc() None to empty string before
 parsing

`inspect.getdoc()` returns `str | None`, but `docstring_parser.parse()`
requires `str`. Functions without a docstring produced `None`, which
the type checker correctly flagged.

Coerce to `""` at the call site. `docstring_parser.parse("")` returns
an empty docstring whose `.description` and `.params` are already
handled by the surrounding `or ""` fallbacks, so runtime behavior is
unchanged.
---
 pyrightconfig.json                              | 11 ++++++++---
 src/pipecat/adapters/schemas/direct_function.py |  2 +-
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 23ed1f3f0..0f42d8550 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -2,11 +2,16 @@
   "typeCheckingMode": "basic",
   "pythonVersion": "3.11",
   "pythonPlatform": "All",
-  "include": ["scripts", "src/pipecat"],
-  "exclude": ["**/*_pb2.py", "**/__pycache__"],
+  "include": [
+    "scripts",
+    "src/pipecat"
+  ],
+  "exclude": [
+    "**/*_pb2.py",
+    "**/__pycache__"
+  ],
   "ignore": [
     "tests",
-    "src/pipecat/adapters/schemas/direct_function.py",
     "src/pipecat/adapters/services/anthropic_adapter.py",
     "src/pipecat/adapters/services/aws_nova_sonic_adapter.py",
     "src/pipecat/adapters/services/bedrock_adapter.py",
diff --git a/src/pipecat/adapters/schemas/direct_function.py b/src/pipecat/adapters/schemas/direct_function.py
index 8ef93d1a7..b42fef604 100644
--- a/src/pipecat/adapters/schemas/direct_function.py
+++ b/src/pipecat/adapters/schemas/direct_function.py
@@ -127,7 +127,7 @@ class BaseDirectFunctionWrapper:
         self.name = self.function.__name__
 
         # Parse docstring for description and parameters
-        docstring = docstring_parser.parse(inspect.getdoc(self.function))
+        docstring = docstring_parser.parse(inspect.getdoc(self.function) or "")
 
         # Get function description
         self.description = (docstring.description or "").strip()

From ec7c35fe9864711dd44b81b463a322805fd2d23b Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 12:15:47 -0400
Subject: [PATCH 10/12] Move Mistral message fixups into MistralLLMAdapter
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Mistral imposes three conversation-history quirks on top of the
OpenAI-compatible wire format: tool messages must be followed by an
assistant message; non-initial system messages are rejected; trailing
assistant messages require `prefix=True`. These rules were applied
inline in `MistralLLMService.build_chat_completion_params`, which is the
wrong layer — every other provider with OpenAI-compatible-but-quirky
shape (Perplexity, etc.) owns its transformations in a
`BaseLLMAdapter` subclass that runs during `get_llm_invocation_params`.

Create `MistralLLMAdapter(OpenAILLMAdapter)` on the Perplexity template
and wire it in via the existing `adapter_class` dispatch. The service
now only handles Mistral-specific request-level mapping (`random_seed`
in place of `seed`), and the message shape concerns live with other
provider format logic.

No behavior change. The transform function casts to `list[dict[str,
Any]]` internally because mutating `role` and attaching Mistral's
non-standard `prefix` field both step outside OpenAI's TypedDict
contract; the cast at the return boundary encodes that we're emitting
Mistral's extended schema, not OpenAI's.
---
 pyrightconfig.json                            |  11 +-
 .../adapters/services/mistral_adapter.py      | 135 ++++++++++++++++++
 src/pipecat/services/mistral/llm.py           |  70 +--------
 3 files changed, 144 insertions(+), 72 deletions(-)
 create mode 100644 src/pipecat/adapters/services/mistral_adapter.py

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 0f42d8550..6dafdcd85 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -2,14 +2,8 @@
   "typeCheckingMode": "basic",
   "pythonVersion": "3.11",
   "pythonPlatform": "All",
-  "include": [
-    "scripts",
-    "src/pipecat"
-  ],
-  "exclude": [
-    "**/*_pb2.py",
-    "**/__pycache__"
-  ],
+  "include": ["scripts", "src/pipecat"],
+  "exclude": ["**/*_pb2.py", "**/__pycache__"],
   "ignore": [
     "tests",
     "src/pipecat/adapters/services/anthropic_adapter.py",
@@ -82,7 +76,6 @@
     "src/pipecat/services/llm_service.py",
     "src/pipecat/services/lmnt/tts.py",
     "src/pipecat/services/mem0/memory.py",
-    "src/pipecat/services/mistral/llm.py",
     "src/pipecat/services/mistral/stt.py",
     "src/pipecat/services/mistral/tts.py",
     "src/pipecat/services/moondream/vision.py",
diff --git a/src/pipecat/adapters/services/mistral_adapter.py b/src/pipecat/adapters/services/mistral_adapter.py
new file mode 100644
index 000000000..272ea129a
--- /dev/null
+++ b/src/pipecat/adapters/services/mistral_adapter.py
@@ -0,0 +1,135 @@
+#
+# Copyright (c) 2024-2026, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+"""Mistral LLM adapter for Pipecat.
+
+Mistral's API uses an OpenAI-compatible interface but imposes three
+conversation-history constraints that OpenAI does not:
+
+1. **Tool messages must be followed by an assistant message.** A ``"tool"``
+   role message that isn't followed by an ``"assistant"`` message is
+   rejected.
+
+2. **Only the initial contiguous system block is permitted.** A
+   ``"system"`` message appearing after any non-system message must be
+   converted to ``"user"``.
+
+3. **A trailing assistant message requires ``prefix=True``.** When the
+   conversation ends on an assistant message, Mistral expects the
+   ``prefix`` flag set so it can continue from that partial reply.
+
+This adapter extends ``OpenAILLMAdapter`` and applies those three fixups
+before the messages reach ``build_chat_completion_params``.
+"""
+
+import copy
+from typing import Any, cast
+
+from openai.types.chat import ChatCompletionMessageParam
+
+from pipecat.adapters.services.open_ai_adapter import OpenAILLMAdapter, OpenAILLMInvocationParams
+from pipecat.processors.aggregators.llm_context import LLMContext
+
+
+class MistralLLMAdapter(OpenAILLMAdapter):
+    """Adapter that transforms messages to satisfy Mistral's API constraints.
+
+    Mistral accepts the OpenAI chat-completions schema but enforces extra
+    rules on conversation history. This adapter extends ``OpenAILLMAdapter``
+    and rewrites the messages produced by the parent to comply with those
+    rules before the request is built.
+    """
+
+    def get_llm_invocation_params(
+        self,
+        context: LLMContext,
+        *,
+        system_instruction: str | None = None,
+        convert_developer_to_user: bool,
+    ) -> OpenAILLMInvocationParams:
+        """Get OpenAI-compatible invocation parameters with Mistral message fixes applied.
+
+        Args:
+            context: The LLM context containing messages, tools, etc.
+            system_instruction: Optional system instruction from service settings
+                or ``run_inference``. Forwarded to the parent adapter.
+            convert_developer_to_user: If True, convert "developer"-role messages
+                to "user"-role messages. Forwarded to the parent adapter.
+
+        Returns:
+            Dictionary of parameters for Mistral's ChatCompletion API, with
+            messages transformed to satisfy Mistral's constraints.
+        """
+        params = super().get_llm_invocation_params(
+            context,
+            system_instruction=system_instruction,
+            convert_developer_to_user=convert_developer_to_user,
+        )
+        params["messages"] = self._transform_messages(list(params["messages"]))
+        return params
+
+    def _transform_messages(
+        self, messages: list[ChatCompletionMessageParam]
+    ) -> list[ChatCompletionMessageParam]:
+        """Transform messages to satisfy Mistral's API constraints.
+
+        Applies three transformation steps in order:
+
+        1. **Insert assistant messages after tool messages** — Any ``"tool"``
+           message not followed by an ``"assistant"`` message gets a minimal
+           ``{"role": "assistant", "content": " "}`` inserted after it.
+
+        2. **Convert non-initial system messages to user** — System messages
+           after the initial contiguous system block are converted to
+           ``"user"``, since Mistral only accepts system messages at the
+           start of a conversation.
+
+        3. **Set prefix on trailing assistant message** — If the final message
+           is an assistant message without a ``prefix`` field, set
+           ``prefix=True`` so Mistral will continue the partial reply.
+
+        Args:
+            messages: List of OpenAI-shaped message dicts.
+
+        Returns:
+            Transformed list of messages satisfying Mistral's constraints.
+        """
+        if not messages:
+            return messages
+
+        # Work on plain dicts: we need to mutate "role" (which OpenAI TypedDict
+        # variants tag with fixed Literals) and to attach Mistral's non-standard
+        # "prefix" field. Cast back on return — the outgoing list is valid for
+        # Mistral's extended schema even though it doesn't fit OpenAI's.
+        msgs: list[dict[str, Any]] = copy.deepcopy([dict(m) for m in messages])
+
+        # Step 1: ensure every "tool" message is followed by an "assistant".
+        insert_at: list[int] = []
+        for i, msg in enumerate(msgs):
+            if msg.get("role") == "tool":
+                is_last = i == len(msgs) - 1
+                if is_last or msgs[i + 1].get("role") != "assistant":
+                    insert_at.append(i + 1)
+        for idx in reversed(insert_at):
+            msgs.insert(idx, {"role": "assistant", "content": " "})
+
+        # Step 2: convert non-initial system messages to "user".
+        # Mistral rejects system messages after any non-system message.
+        first_non_system = next(
+            (i for i, m in enumerate(msgs) if m.get("role") != "system"),
+            len(msgs),
+        )
+        for i in range(first_non_system, len(msgs)):
+            if msgs[i].get("role") == "system":
+                msgs[i]["role"] = "user"
+
+        # Step 3: set prefix on a trailing assistant message so Mistral will
+        # continue it rather than rejecting the turn.
+        last = msgs[-1]
+        if last.get("role") == "assistant" and "prefix" not in last:
+            last["prefix"] = True
+
+        return cast(list[ChatCompletionMessageParam], msgs)
diff --git a/src/pipecat/services/mistral/llm.py b/src/pipecat/services/mistral/llm.py
index c1ac8b652..a85d80ff5 100644
--- a/src/pipecat/services/mistral/llm.py
+++ b/src/pipecat/services/mistral/llm.py
@@ -10,8 +10,8 @@ from collections.abc import Sequence
 from dataclasses import dataclass
 
 from loguru import logger
-from openai.types.chat import ChatCompletionMessageParam
 
+from pipecat.adapters.services.mistral_adapter import MistralLLMAdapter
 from pipecat.adapters.services.open_ai_adapter import OpenAILLMInvocationParams
 from pipecat.frames.frames import FunctionCallFromLLM
 from pipecat.services.openai.base_llm import BaseOpenAILLMService
@@ -36,6 +36,8 @@ class MistralLLMService(OpenAILLMService):
     # This value is used by BaseOpenAILLMService when calling the adapter.
     supports_developer_role = False
 
+    adapter_class = MistralLLMAdapter
+
     Settings = MistralLLMSettings
     _settings: Settings
 
@@ -92,60 +94,6 @@ class MistralLLMService(OpenAILLMService):
         logger.debug(f"Creating Mistral client with api {base_url}")
         return super().create_client(api_key, base_url, **kwargs)
 
-    def _apply_mistral_fixups(
-        self, messages: list[ChatCompletionMessageParam]
-    ) -> list[ChatCompletionMessageParam]:
-        """Apply fixups to messages to meet Mistral-specific requirements.
-
-        1. A "tool"-role message must be followed by an assistant message.
-
-        2. "system"-role messages must only appear at the start of a
-           conversation.
-
-        3. Assistant messages must have prefix=True when they are the final
-           message in a conversation (but at no other point).
-
-        Args:
-            messages: The original list of messages.
-
-        Returns:
-            Messages with Mistral prefix requirement applied to final assistant message.
-        """
-        if not messages:
-            return messages
-
-        # Create a copy to avoid modifying the original
-        fixed_messages = [dict(msg) for msg in messages]
-
-        # Ensure all tool responses are followed by an assistant message
-        assistant_insert_indices = []
-        for i, msg in enumerate(fixed_messages):
-            if msg.get("role") == "tool":
-                # If this is the last message or the next message is not assistant
-                if i == len(fixed_messages) - 1 or fixed_messages[i + 1].get("role") != "assistant":
-                    assistant_insert_indices.append(i + 1)
-        for idx in reversed(assistant_insert_indices):
-            fixed_messages.insert(idx, {"role": "assistant", "content": " "})
-
-        # Convert any "system" messages that aren't at the start (i.e., after the initial contiguous block) to "user"
-        first_non_system_idx = next(
-            (i for i, msg in enumerate(fixed_messages) if msg.get("role") != "system"),
-            len(fixed_messages),
-        )
-        for i, msg in enumerate(fixed_messages):
-            if msg.get("role") == "system" and i >= first_non_system_idx:
-                msg["role"] = "user"
-
-        # Get the last message
-        last_message = fixed_messages[-1]
-
-        # Only add prefix=True to the last message if it's an assistant message
-        # and Mistral would otherwise reject it
-        if last_message.get("role") == "assistant" and "prefix" not in last_message:
-            last_message["prefix"] = True
-
-        return fixed_messages
-
     async def run_function_calls(self, function_calls: Sequence[FunctionCallFromLLM]):
         """Execute function calls, filtering out already-completed ones.
 
@@ -208,18 +156,14 @@ class MistralLLMService(OpenAILLMService):
     def build_chat_completion_params(self, params_from_context: OpenAILLMInvocationParams) -> dict:
         """Build parameters for Mistral chat completion request.
 
-        Handles Mistral-specific requirements including:
-        - Assistant message prefix requirement for API compatibility
-        - Parameter mapping (random_seed instead of seed)
-        - Core completion settings
+        Handles Mistral-specific parameter mapping (``random_seed`` in place
+        of ``seed``). Message-shape fixups required by Mistral are applied
+        by :class:`MistralLLMAdapter` upstream.
         """
-        # Apply Mistral's assistant prefix requirement for API compatibility
-        fixed_messages = self._apply_mistral_fixups(params_from_context["messages"])
-
         params = {
             "model": self._settings.model,
             "stream": True,
-            "messages": fixed_messages,
+            "messages": params_from_context["messages"],
             "tools": params_from_context["tools"],
             "tool_choice": params_from_context["tool_choice"],
             "frequency_penalty": self._settings.frequency_penalty,

From b0962861c8f4b544ed39d4b73c841fa74a1ede65 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 12:19:16 -0400
Subject: [PATCH 11/12] Acknowledge Tkinter's GC-reference idiom with a scoped
 type ignore
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Tkinter's `Label` only stores `PhotoImage` references at the C level, so
Python GC eats them unless something on the Python side keeps a
reference. The canonical fix is to stash the reference on the widget
itself: `label.image = photo`. Tkinter widgets are plain Python objects,
so the assignment works at runtime, but the stub declares no `image`
attribute (correctly — there isn't one; we're adding it).

Narrow the suppression to `# type: ignore[attr-defined]` on the one
line. The existing comment above the assignment already documents why.
---
 pyrightconfig.json                 | 1 -
 src/pipecat/transports/local/tk.py | 2 +-
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/pyrightconfig.json b/pyrightconfig.json
index 6dafdcd85..1130ba338 100644
--- a/pyrightconfig.json
+++ b/pyrightconfig.json
@@ -114,7 +114,6 @@
     "src/pipecat/transports/heygen/transport.py",
     "src/pipecat/transports/lemonslice/transport.py",
     "src/pipecat/transports/livekit/transport.py",
-    "src/pipecat/transports/local/tk.py",
     "src/pipecat/transports/smallwebrtc/connection.py",
     "src/pipecat/transports/smallwebrtc/request_handler.py",
     "src/pipecat/transports/smallwebrtc/transport.py",
diff --git a/src/pipecat/transports/local/tk.py b/src/pipecat/transports/local/tk.py
index a11f4f277..be9d6db1c 100644
--- a/src/pipecat/transports/local/tk.py
+++ b/src/pipecat/transports/local/tk.py
@@ -228,7 +228,7 @@ class TkOutputTransport(BaseOutputTransport):
 
         # This holds a reference to the photo, preventing it from being garbage
         # collected.
-        self._image_label.image = photo
+        self._image_label.image = photo  # type: ignore[attr-defined]
 
 
 class TkLocalTransport(BaseTransport):

From 4f6e76e6fdbf61c300606ce84307c5e1a84b6e92 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Wed, 22 Apr 2026 12:23:33 -0400
Subject: [PATCH 12/12] Add changelog entries for #4352

---
 changelog/4352.changed.md | 1 +
 changelog/4352.fixed.2.md | 1 +
 changelog/4352.fixed.md   | 1 +
 3 files changed, 3 insertions(+)
 create mode 100644 changelog/4352.changed.md
 create mode 100644 changelog/4352.fixed.2.md
 create mode 100644 changelog/4352.fixed.md

diff --git a/changelog/4352.changed.md b/changelog/4352.changed.md
new file mode 100644
index 000000000..ed77eec98
--- /dev/null
+++ b/changelog/4352.changed.md
@@ -0,0 +1 @@
+- `ToolsSchema(standard_tools=...)` now accepts any `Sequence[FunctionSchema | DirectFunction]` rather than requiring an exact `list` of the union. Callers can pass a narrower `list[FunctionSchema]` (or any other `Sequence`) without the type checker complaining about list invariance.
diff --git a/changelog/4352.fixed.2.md b/changelog/4352.fixed.2.md
new file mode 100644
index 000000000..5ea5de888
--- /dev/null
+++ b/changelog/4352.fixed.2.md
@@ -0,0 +1 @@
+- Fixed direct-function registration crashing for functions without a docstring. `DirectFunctionWrapper` passed `inspect.getdoc()`'s result to `docstring_parser.parse()`, which raises when the docstring is `None`. Functions now register cleanly whether or not they have a docstring; an empty docstring produces empty description and parameter metadata as expected.
diff --git a/changelog/4352.fixed.md b/changelog/4352.fixed.md
new file mode 100644
index 000000000..6e66fb122
--- /dev/null
+++ b/changelog/4352.fixed.md
@@ -0,0 +1 @@
+- Fixed `AssemblyAISTTService`, `CartesiaSTTService`, `GradiumSTTService`, and `SonioxSTTService` crashing the pipeline on transient WebSocket send failures. Each `run_stt` sent audio directly without catching errors, so a single network hiccup mid-stream raised an uncaught exception through `process_frame`. The guards now log a warning and let the connection-state check on the next call handle recovery, matching the pattern used by Deepgram, xAI, Azure, and other push-based STTs.