Update python-compatibility workflow to include new user project check

Update README with local dev setup for contributors

.github/workflows/python-compatibility.yaml

@@ -9,14 +9,14 @@ on:
     paths: ['pyproject.toml']
 
 jobs:
-  test-compatibility:
+  test-dev-environment:
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
       matrix:
         python-version: ['3.10.18', '3.11.13', '3.12.11', '3.13.5']
 
-    name: Python ${{ matrix.python-version }}
+    name: Dev Environment - Python ${{ matrix.python-version }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -55,7 +55,69 @@ jobs:
             --no-extra moondream \
             --no-extra mlx-whisper
 
-      - name: Verify installation
+      - name: Verify dev installation
         run: |
           uv run python --version
-          uv run python -c "import pipecat; print('✅ Pipecat imports successfully')"
+          uv run python -c "import pipecat; print('✅ Dev environment - Pipecat imports successfully')"
+
+  test-user-experience:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ['3.10.18', '3.11.13', '3.12.11', '3.13.5']
+
+    name: User Experience - Python ${{ matrix.python-version }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y \
+            portaudio19-dev \
+            libcairo2-dev \
+            libgirepository1.0-dev \
+            pkg-config
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          version: 'latest'
+
+      - name: Set up Python ${{ matrix.python-version }}
+        run: |
+          uv python install ${{ matrix.python-version }}
+
+      - name: Build local package
+        run: |
+          uv build
+
+      - name: Create test project
+        run: |
+          mkdir test-project
+          cd test-project
+          uv init --python ${{ matrix.python-version }}
+
+      - name: Test comprehensive extras with uv add (Python 3.10-3.12)
+        if: "!startsWith(matrix.python-version, '3.13.')"
+        run: |
+          cd test-project
+          # Use uv add with built wheel to leverage dependency management
+          uv add "../dist/pipecat_ai-"*".whl[anthropic,assemblyai,asyncai,aws,aws-nova-sonic,azure,cartesia,cerebras,deepseek,daily,deepgram,elevenlabs,fal,fireworks,fish,gladia,google,grok,groq,gstreamer,heygen,inworld,koala,langchain,livekit,lmnt,local,mcp,mem0,mlx-whisper,moondream,nim,neuphonic,noisereduce,openai,openpipe,openrouter,perplexity,playht,qwen,rime,riva,runner,sambanova,sentry,local-smart-turn,remote-smart-turn,silero,simli,soniox,soundfile,speechmatics,tavus,together,tracing,ultravox,webrtc,websocket,whisper]"
+
+      - name: Test Python 3.13 compatible extras with uv add
+        if: startsWith(matrix.python-version, '3.13.')
+        run: |
+          cd test-project
+          # Use uv add with built wheel and Python 3.13 compatible extras
+          uv add "../dist/pipecat_ai-"*".whl[anthropic,assemblyai,asyncai,aws,aws-nova-sonic,azure,cartesia,cerebras,deepseek,daily,deepgram,elevenlabs,fal,fireworks,fish,gladia,google,grok,groq,gstreamer,heygen,inworld,koala,langchain,livekit,lmnt,local,mcp,mem0,nim,neuphonic,noisereduce,openai,openpipe,openrouter,perplexity,playht,qwen,rime,riva,runner,sambanova,sentry,remote-smart-turn,silero,simli,soniox,soundfile,speechmatics,tavus,together,tracing,webrtc,websocket,whisper]"
+
+      - name: Verify user installation
+        run: |
+          cd test-project
+          uv run python --version
+          uv run python -c "import pipecat; print('✅ User experience - Pipecat imports successfully')"
+          # Test that basic functionality works
+          uv run python -c "from pipecat.pipeline.pipeline import Pipeline; print('✅ Pipeline import works')"

CHANGELOG.md

@@ -5,6 +5,42 @@ All notable changes to **Pipecat** will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## Unreleased
+
+### Changed
+
+- Updated `pyproject.toml` to once again pin `numba` to `>=0.61.2` in order to
+  resolve package versioning issues.
+
+### Other
+
+- Updated `15-switch-voices.py` and `15a-switch-languages.py` examples to show
+  how to enclose complex logic (e.g. `ParallelPipeline`) into a single processor
+  so the main pipeline becomes simpler.
+
+## [0.0.79] - 2025-08-07
+
+### Changed
+
+- Changed `pipecat-ai`'s `openai` dependency to `>=1.74.0,<=1.99.1` due to a
+  breaking change in `openai` 1.99.2 ([commit](https://github.com/openai/openai-python/commit/657f551dbe583ffb259d987dafae12c6211fba06))
+
+### Deprecated
+
+- `TTSService.say()` is deprecated, push a `TTSSpeakFrame` instead. Calling
+  functions directly is a discouraged pattern in Pipecat because, for example,
+  it might cause issues with frame ordering.
+
+- `LLMMessagesFrame` is deprecated, in favor of either:
+
+  - `LLMMessagesUpdateFrame` with `run_llm=True`
+  - `OpenAILLMContextFrame` with desired messages in a new context
+
+- `LLMUserResponseAggregator` and `LLMAssistantResponseAggregator` are
+  deprecated, as they depended on the now-deprecated `LLMMessagesFrame`. Use
+  `LLMUserContextAggregator` and `LLMAssistantResponseAggregator` (or
+  LLM-specific subclasses thereof) instead.
+
 ## [0.0.78] - 2025-08-07
 
 ### Added
@@ -53,7 +89,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added Chinese, Japanese, Korean word timestamp support to
   `CartesiaTTSService`.
 
-- Added `region` parameter to `GladiaSTTService`. Accepted values: eu-west (default), us-west.
+- Added `region` parameter to `GladiaSTTService`. Accepted values: eu-west
+  (default), us-west.
 
 ### Changed
 

README.md

@@ -128,7 +128,7 @@ You can get started with Pipecat running on your local machine, then move your a
 2. Install development and testing dependencies:
 
    ```bash
-   uv sync --group dev --all-extras --no-extra krisp
+   uv sync --group dev --all-extras --no-extra gstreamer --no-extra krisp --no-extra local
    ```
 
 3. Install the git pre-commit hooks:
@@ -137,18 +137,25 @@ You can get started with Pipecat running on your local machine, then move your a
    uv run pre-commit install
    ```
 
-### Python 3.13+ Note
+### Python 3.13+ Compatibility
 
-Some features require PyTorch (not yet available on Python 3.13+):
-
-- `ultravox`, `local-smart-turn`, `moondream`, `mlx-whisper`
-
-**For full compatibility:** Use Python 3.12
+Some features require PyTorch, which doesn't yet support Python 3.13+. Install using:
 
 ```bash
-uv python pin 3.12 && uv sync --group dev --all-extras --no-extra krisp
+uv sync --group dev --all-extras \
+  --no-extra gstreamer \
+  --no-extra krisp \
+  --no-extra local \
+  --no-extra local-smart-turn \
+  --no-extra mlx-whisper \
+  --no-extra moondream \
+  --no-extra ultravox
 ```
 
+> **Tip:** For full compatibility, use Python 3.12: `uv python pin 3.12`
+
+> **Note**: Some extras (local, gstreamer) require system dependencies. See documentation if you encounter build errors.
+
 ### Running tests
 
 To run all tests, from the root directory:

env.example

@@ -29,6 +29,9 @@ CARTESIA_API_KEY=...
 DAILY_API_KEY=...
 DAILY_SAMPLE_ROOM_URL=https://...
 
+# Deepgram
+DEEPGRAM_API_KEY=...
+
 # ElevenLabs
 ELEVENLABS_API_KEY=...
 ELEVENLABS_VOICE_ID=...

examples/foundational/02-llm-say-one-thing.py

@@ -9,10 +9,14 @@ import os
 from dotenv import load_dotenv
 from loguru import logger
 
-from pipecat.frames.frames import EndFrame, LLMMessagesFrame
+from pipecat.frames.frames import EndFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineTask
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
+)
 from pipecat.runner.types import RunnerArguments
 from pipecat.runner.utils import create_transport
 from pipecat.services.cartesia.tts import CartesiaTTSService
@@ -59,7 +63,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     # Register an event handler so we can play the audio when the client joins
     @transport.event_handler("on_client_connected")
     async def on_client_connected(transport, client):
-        await task.queue_frames([LLMMessagesFrame(messages), EndFrame()])
+        await task.queue_frames([OpenAILLMContextFrame(OpenAILLMContext(messages)), EndFrame()])
 
     runner = PipelineRunner(handle_sigint=runner_args.handle_sigint)
 

examples/foundational/05-sync-speech-and-image.py

@@ -15,13 +15,16 @@ from pipecat.frames.frames import (
     DataFrame,
     Frame,
     LLMFullResponseStartFrame,
-    LLMMessagesFrame,
     TextFrame,
 )
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.sync_parallel_pipeline import SyncParallelPipeline
 from pipecat.pipeline.task import PipelineTask
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
+)
 from pipecat.processors.aggregators.sentence import SentenceAggregator
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 from pipecat.runner.types import RunnerArguments
@@ -153,7 +156,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
                 }
             ]
             frames.append(MonthFrame(month=month))
-            frames.append(LLMMessagesFrame(messages))
+            frames.append(OpenAILLMContextFrame(OpenAILLMContext(messages)))
 
         task = PipelineTask(
             pipeline,

examples/foundational/05a-local-sync-speech-and-image.py

@@ -15,7 +15,6 @@ from loguru import logger
 
 from pipecat.frames.frames import (
     Frame,
-    LLMMessagesFrame,
     OutputAudioRawFrame,
     TextFrame,
     TTSAudioRawFrame,
@@ -25,6 +24,10 @@ from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.sync_parallel_pipeline import SyncParallelPipeline
 from pipecat.pipeline.task import PipelineTask
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
+)
 from pipecat.processors.aggregators.sentence import SentenceAggregator
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 from pipecat.services.cartesia.tts import CartesiaHttpTTSService
@@ -137,7 +140,7 @@ async def main():
             )
 
             task = PipelineTask(pipeline)
-            await task.queue_frame(LLMMessagesFrame(messages))
+            await task.queue_frame(OpenAILLMContextFrame(OpenAILLMContext(messages)))
             await task.stop_when_done()
 
             await runner.run(task)

examples/foundational/07b-interruptible-langchain.py

@@ -16,13 +16,16 @@ from langchain_openai import ChatOpenAI
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import LLMMessagesFrame
+from pipecat.frames.frames import LLMMessagesUpdateFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
 from pipecat.processors.aggregators.llm_response import (
-    LLMAssistantResponseAggregator,
-    LLMUserResponseAggregator,
+    LLMAssistantContextAggregator,
+    LLMUserContextAggregator,
+)
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
 )
 from pipecat.processors.frameworks.langchain import LangchainProcessor
 from pipecat.runner.types import RunnerArguments
@@ -97,8 +100,9 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     )
     lc = LangchainProcessor(history_chain)
 
-    tma_in = LLMUserResponseAggregator()
-    tma_out = LLMAssistantResponseAggregator()
+    context = OpenAILLMContext()
+    tma_in = LLMUserContextAggregator(context=context)
+    tma_out = LLMAssistantContextAggregator(context=context)
 
     pipeline = Pipeline(
         [
@@ -125,11 +129,11 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     async def on_client_connected(transport, client):
         logger.info(f"Client connected")
         # Kick off the conversation.
-        # the `LLMMessagesFrame` will be picked up by the LangchainProcessor using
+        # An `OpenAILLMContextFrame` will be picked up by the LangchainProcessor using
         # only the content of the last message to inject it in the prompt defined
         # above. So no role is required here.
         messages = [({"content": "Please briefly introduce yourself to the user."})]
-        await task.queue_frames([LLMMessagesFrame(messages)])
+        await task.queue_frames([LLMMessagesUpdateFrame(messages, run_llm=True)])
 
     @transport.event_handler("on_client_disconnected")
     async def on_client_disconnected(transport, client):

examples/foundational/08-bots-arguing.py

@@ -6,9 +6,13 @@ from typing import Tuple
 import aiohttp
 from dotenv import load_dotenv
 
-from pipecat.frames.frames import AudioFrame, EndFrame, ImageFrame, LLMMessagesFrame, TextFrame
+from pipecat.frames.frames import AudioFrame, EndFrame, ImageFrame, TextFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.processors.aggregators import SentenceAggregator
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
+)
 from pipecat.runner.daily import configure
 from pipecat.services.azure import AzureLLMService, AzureTTSService
 from pipecat.services.elevenlabs import ElevenLabsTTSService
@@ -79,7 +83,7 @@ async def main():
             sentence_aggregator = SentenceAggregator()
             pipeline = Pipeline([llm, sentence_aggregator, tts1], source_queue, sink_queue)
 
-            await source_queue.put(LLMMessagesFrame(messages))
+            await source_queue.put(OpenAILLMContextFrame(OpenAILLMContext(messages)))
             await source_queue.put(EndFrame())
             await pipeline.run_pipeline()
 

examples/foundational/12-describe-video.py

@@ -11,7 +11,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import Frame, TextFrame, UserImageRequestFrame
+from pipecat.frames.frames import Frame, TextFrame, TTSSpeakFrame, UserImageRequestFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineTask
@@ -119,7 +119,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         image_requester.set_participant_id(client_id)
 
         # Welcome message
-        await tts.say("Hi there! Feel free to ask me what I see.")
+        await task.queue_frame(TTSSpeakFrame("Hi there! Feel free to ask me what I see."))
 
     @transport.event_handler("on_client_disconnected")
     async def on_client_disconnected(transport, client):

examples/foundational/12a-describe-video-gemini-flash.py

@@ -11,7 +11,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import Frame, TextFrame, UserImageRequestFrame
+from pipecat.frames.frames import Frame, TextFrame, TTSSpeakFrame, UserImageRequestFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -123,7 +123,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         image_requester.set_participant_id(client_id)
 
         # Welcome message
-        await tts.say("Hi there! Feel free to ask me what I see.")
+        await task.queue_frame(TTSSpeakFrame("Hi there! Feel free to ask me what I see."))
 
     @transport.event_handler("on_client_disconnected")
     async def on_client_disconnected(transport, client):

examples/foundational/12b-describe-video-gpt-4o.py

@@ -11,7 +11,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import Frame, TextFrame, UserImageRequestFrame
+from pipecat.frames.frames import Frame, TextFrame, TTSSpeakFrame, UserImageRequestFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -123,7 +123,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         image_requester.set_participant_id(client_id)
 
         # Welcome message
-        await tts.say("Hi there! Feel free to ask me what I see.")
+        await task.queue_frame(TTSSpeakFrame("Hi there! Feel free to ask me what I see."))
 
     @transport.event_handler("on_client_disconnected")
     async def on_client_disconnected(transport, client):

examples/foundational/12c-describe-video-anthropic.py

@@ -11,7 +11,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import Frame, TextFrame, UserImageRequestFrame
+from pipecat.frames.frames import Frame, TextFrame, TTSSpeakFrame, UserImageRequestFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -123,7 +123,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         image_requester.set_participant_id(client_id)
 
         # Welcome message
-        await tts.say("Hi there! Feel free to ask me what I see.")
+        await task.queue_frame(TTSSpeakFrame("Hi there! Feel free to ask me what I see."))
 
     @transport.event_handler("on_client_disconnected")
     async def on_client_disconnected(transport, client):

examples/foundational/15-switch-voices.py

@@ -12,6 +12,7 @@ from loguru import logger
 from openai.types.chat import ChatCompletionToolParam
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
+from pipecat.frames.frames import Frame
 from pipecat.pipeline.parallel_pipeline import ParallelPipeline
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
@@ -31,29 +32,54 @@ from pipecat.transports.services.daily import DailyParams
 load_dotenv(override=True)
 
 
-current_voice = "News Lady"
+class SwitchVoices(ParallelPipeline):
+    def __init__(self):
+        self._current_voice = "News Lady"
 
+        news_lady = CartesiaTTSService(
+            api_key=os.getenv("CARTESIA_API_KEY"),
+            voice_id="bf991597-6c13-47e4-8411-91ec2de5c466",  # Newslady
+        )
 
-async def switch_voice(params: FunctionCallParams):
-    global current_voice
-    current_voice = params.arguments["voice"]
-    await params.result_callback(
-        {
-            "voice": f"You are now using your {current_voice} voice. Your responses should now be as if you were a {current_voice}."
-        }
-    )
+        british_lady = CartesiaTTSService(
+            api_key=os.getenv("CARTESIA_API_KEY"),
+            voice_id="71a7ad14-091c-4e8e-a314-022ece01c121",  # British Reading Lady
+        )
 
+        barbershop_man = CartesiaTTSService(
+            api_key=os.getenv("CARTESIA_API_KEY"),
+            voice_id="a0e99841-438c-4a64-b679-ae501e7d6091",  # Barbershop Man
+        )
 
-async def news_lady_filter(frame) -> bool:
-    return current_voice == "News Lady"
+        super().__init__(
+            # News Lady voice
+            [FunctionFilter(self.news_lady_filter), news_lady],
+            # British Reading Lady voice
+            [FunctionFilter(self.british_lady_filter), british_lady],
+            # Barbershop Man voice
+            [FunctionFilter(self.barbershop_man_filter), barbershop_man],
+        )
 
+    @property
+    def current_voice(self):
+        return self._current_voice
 
-async def british_lady_filter(frame) -> bool:
-    return current_voice == "British Lady"
+    async def switch_voice(self, params: FunctionCallParams):
+        self._current_voice = params.arguments["voice"]
+        await params.result_callback(
+            {
+                "voice": f"You are now using your {self.current_voice} voice. Your responses should now be as if you were a {self.current_voice}."
+            }
+        )
 
+    async def news_lady_filter(self, _: Frame) -> bool:
+        return self.current_voice == "News Lady"
 
-async def barbershop_man_filter(frame) -> bool:
-    return current_voice == "Barbershop Man"
+    async def british_lady_filter(self, _: Frame) -> bool:
+        return self.current_voice == "British Lady"
+
+    async def barbershop_man_filter(self, _: Frame) -> bool:
+        return self.current_voice == "Barbershop Man"
 
 
 # We store functions so objects (e.g. SileroVADAnalyzer) don't get
@@ -83,23 +109,10 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
     stt = DeepgramSTTService(api_key=os.getenv("DEEPGRAM_API_KEY"))
 
-    news_lady = CartesiaTTSService(
-        api_key=os.getenv("CARTESIA_API_KEY"),
-        voice_id="bf991597-6c13-47e4-8411-91ec2de5c466",  # Newslady
-    )
-
-    british_lady = CartesiaTTSService(
-        api_key=os.getenv("CARTESIA_API_KEY"),
-        voice_id="71a7ad14-091c-4e8e-a314-022ece01c121",  # British Reading Lady
-    )
-
-    barbershop_man = CartesiaTTSService(
-        api_key=os.getenv("CARTESIA_API_KEY"),
-        voice_id="a0e99841-438c-4a64-b679-ae501e7d6091",  # Barbershop Man
-    )
+    tts = SwitchVoices()
 
     llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"))
-    llm.register_function("switch_voice", switch_voice)
+    llm.register_function("switch_voice", tts.switch_voice)
 
     tools = [
         ChatCompletionToolParam(
@@ -136,14 +149,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
             stt,
             context_aggregator.user(),  # User responses
             llm,  # LLM
-            ParallelPipeline(  # TTS (one of the following vocies)
-                [FunctionFilter(news_lady_filter), news_lady],  # News Lady voice
-                [
-                    FunctionFilter(british_lady_filter),
-                    british_lady,
-                ],  # British Reading Lady voice
-                [FunctionFilter(barbershop_man_filter), barbershop_man],  # Barbershop Man voice
-            ),
+            tts,  # TTS with switch voice functionality
             transport.output(),  # Transport bot output
             context_aggregator.assistant(),  # Assistant spoken responses
         ]
@@ -165,7 +171,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         messages.append(
             {
                 "role": "system",
-                "content": f"Please introduce yourself to the user and let them know the voices you can do. Your initial responses should be as if you were a {current_voice}.",
+                "content": f"Please introduce yourself to the user and let them know the voices you can do. Your initial responses should be as if you were a {tts.current_voice}.",
             }
         )
         await task.queue_frames([context_aggregator.user().get_context_frame()])

examples/foundational/15a-switch-languages.py

@@ -13,6 +13,7 @@ from loguru import logger
 from openai.types.chat import ChatCompletionToolParam
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
+from pipecat.frames.frames import Frame
 from pipecat.pipeline.parallel_pipeline import ParallelPipeline
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
@@ -32,23 +33,42 @@ from pipecat.transports.services.daily import DailyParams
 load_dotenv(override=True)
 
 
-current_language = "English"
+class SwitchLanguage(ParallelPipeline):
+    def __init__(self):
+        self._current_language = "English"
 
+        english_tts = CartesiaTTSService(
+            api_key=os.getenv("CARTESIA_API_KEY"),
+            voice_id="71a7ad14-091c-4e8e-a314-022ece01c121",  # British Reading Lady
+        )
 
-async def switch_language(params: FunctionCallParams):
-    global current_language
-    current_language = params.arguments["language"]
-    await params.result_callback(
-        {"voice": f"Your answers from now on should be in {current_language}."}
-    )
+        spanish_tts = CartesiaTTSService(
+            api_key=os.getenv("CARTESIA_API_KEY"),
+            voice_id="d4db5fb9-f44b-4bd1-85fa-192e0f0d75f9",  # Spanish-speaking Lady
+        )
 
+        super().__init__(
+            # English
+            [FunctionFilter(self.english_filter), english_tts],
+            # Spanish
+            [FunctionFilter(self.spanish_filter), spanish_tts],
+        )
 
-async def english_filter(frame) -> bool:
-    return current_language == "English"
+    @property
+    def current_language(self):
+        return self._current_language
 
+    async def switch_language(self, params: FunctionCallParams):
+        self._current_language = params.arguments["language"]
+        await params.result_callback(
+            {"voice": f"Your answers from now on should be in {self.current_language}."}
+        )
 
-async def spanish_filter(frame) -> bool:
-    return current_language == "Spanish"
+    async def english_filter(self, _: Frame) -> bool:
+        return self.current_language == "English"
+
+    async def spanish_filter(self, _: Frame) -> bool:
+        return self.current_language == "Spanish"
 
 
 # We store functions so objects (e.g. SileroVADAnalyzer) don't get
@@ -80,18 +100,10 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         api_key=os.getenv("DEEPGRAM_API_KEY"), live_options=LiveOptions(language="multi")
     )
 
-    english_tts = CartesiaTTSService(
-        api_key=os.getenv("CARTESIA_API_KEY"),
-        voice_id="71a7ad14-091c-4e8e-a314-022ece01c121",  # British Reading Lady
-    )
-
-    spanish_tts = CartesiaTTSService(
-        api_key=os.getenv("CARTESIA_API_KEY"),
-        voice_id="d4db5fb9-f44b-4bd1-85fa-192e0f0d75f9",  # Spanish-speaking Lady
-    )
+    tts = SwitchLanguage()
 
     llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"))
-    llm.register_function("switch_language", switch_language)
+    llm.register_function("switch_language", tts.switch_language)
 
     tools = [
         ChatCompletionToolParam(
@@ -128,10 +140,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
             stt,  # STT
             context_aggregator.user(),  # User responses
             llm,  # LLM
-            ParallelPipeline(  # TTS (bot will speak the chosen language)
-                [FunctionFilter(english_filter), english_tts],  # English
-                [FunctionFilter(spanish_filter), spanish_tts],  # Spanish
-            ),
+            tts,  # TTS (bot will speak the chosen language)
             transport.output(),  # Transport bot output
             context_aggregator.assistant(),  # Assistant spoken responses
         ]
@@ -153,7 +162,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         messages.append(
             {
                 "role": "system",
-                "content": f"Please introduce yourself to the user and let them know the languages you speak. Your initial responses should be in {current_language}.",
+                "content": f"Please introduce yourself to the user and let them know the languages you speak. Your initial responses should be in {tts.current_language}.",
             }
         )
         await task.queue_frames([context_aggregator.user().get_context_frame()])

examples/foundational/17-detect-user-idle.py

@@ -11,7 +11,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
-from pipecat.frames.frames import EndFrame, LLMMessagesFrame, TTSSpeakFrame
+from pipecat.frames.frames import EndFrame, LLMMessagesAppendFrame, TTSSpeakFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -75,23 +75,19 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     async def handle_user_idle(user_idle: UserIdleProcessor, retry_count: int) -> bool:
         if retry_count == 1:
             # First attempt: Add a gentle prompt to the conversation
-            messages.append(
-                {
-                    "role": "system",
-                    "content": "The user has been quiet. Politely and briefly ask if they're still there.",
-                }
-            )
-            await user_idle.push_frame(LLMMessagesFrame(messages))
+            message = {
+                "role": "system",
+                "content": "The user has been quiet. Politely and briefly ask if they're still there.",
+            }
+            await user_idle.push_frame(LLMMessagesAppendFrame([message], run_llm=True))
             return True
         elif retry_count == 2:
             # Second attempt: More direct prompt
-            messages.append(
-                {
-                    "role": "system",
-                    "content": "The user is still inactive. Ask if they'd like to continue our conversation.",
-                }
-            )
-            await user_idle.push_frame(LLMMessagesFrame(messages))
+            message = {
+                "role": "system",
+                "content": "The user is still inactive. Ask if they'd like to continue our conversation.",
+            }
+            await user_idle.push_frame(LLMMessagesAppendFrame([message], run_llm=True))
             return True
         else:
             # Third attempt: End the conversation

examples/foundational/20a-persistent-context-openai.py

@@ -14,6 +14,7 @@ from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
 from pipecat.audio.vad.vad_analyzer import VADParams
+from pipecat.frames.frames import TTSSpeakFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -34,7 +35,6 @@ load_dotenv(override=True)
 
 
 BASE_FILENAME = "/tmp/pipecat_conversation_"
-tts = None
 
 
 async def fetch_weather_from_api(params: FunctionCallParams):
@@ -87,7 +87,7 @@ async def load_conversation(params: FunctionCallParams):
             logger.debug(
                 f"loaded conversation from {filename}\n{json.dumps(params.context.messages, indent=4)}"
             )
-        await tts.say("Ok, I've loaded that conversation.")
+        await params.llm.queue_frame(TTSSpeakFrame("Ok, I've loaded that conversation."))
     except Exception as e:
         await params.result_callback({"success": False, "error": str(e)})
 
@@ -190,7 +190,7 @@ transport_params = {
 async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     logger.info(f"Starting bot")
 
-    global tts
+    stt = DeepgramSTTService(api_key=os.getenv("DEEPGRAM_API_KEY"))
 
     tts = CartesiaTTSService(
         api_key=os.getenv("CARTESIA_API_KEY"),

examples/foundational/20c-persistent-context-anthropic.py

@@ -14,6 +14,7 @@ from loguru import logger
 
 from pipecat.audio.vad.silero import SileroVADAnalyzer
 from pipecat.audio.vad.vad_analyzer import VADParams
+from pipecat.frames.frames import TTSSpeakFrame
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineParams, PipelineTask
@@ -88,7 +89,7 @@ async def load_conversation(params: FunctionCallParams):
             logger.debug(
                 f"loaded conversation from {filename}\n{json.dumps(params.context.messages, indent=4)}"
             )
-        await tts.say("Ok, I've loaded that conversation.")
+        await params.llm.queue_frame(TTSSpeakFrame("Ok, I've loaded that conversation."))
     except Exception as e:
         await params.result_callback({"success": False, "error": str(e)})
 

examples/foundational/22b-natural-conversation-proposal.py

@@ -19,7 +19,6 @@ from pipecat.frames.frames import (
     Frame,
     FunctionCallInProgressFrame,
     FunctionCallResultFrame,
-    LLMMessagesFrame,
     StartFrame,
     StartInterruptionFrame,
     StopInterruptionFrame,
@@ -60,10 +59,6 @@ classifier_statement = "Determine if the user's statement ends with a complete t
 
 
 class StatementJudgeContextFilter(FrameProcessor):
-    def __init__(self, notifier: BaseNotifier, **kwargs):
-        super().__init__(**kwargs)
-        self._notifier = notifier
-
     async def process_frame(self, frame: Frame, direction: FrameDirection):
         await super().process_frame(frame, direction)
         # We must not block system frames.
@@ -71,13 +66,8 @@ class StatementJudgeContextFilter(FrameProcessor):
             await self.push_frame(frame, direction)
             return
 
-        # Just treat an LLMMessagesFrame as complete, no matter what.
-        if isinstance(frame, LLMMessagesFrame):
-            await self._notifier.notify()
-            return
-
-        # Otherwise, we only want to handle OpenAILLMContextFrames, and only want to push a simple
-        # messages frame that contains a system prompt and the most recent user messages,
+        # We only want to handle OpenAILLMContextFrames, and only want to push through a simplified
+        # context frame that contains a system prompt and the most recent user messages,
         # concatenated.
         if isinstance(frame, OpenAILLMContextFrame):
             logger.debug(f"Context Frame: {frame}")
@@ -96,7 +86,7 @@ class StatementJudgeContextFilter(FrameProcessor):
                     for content in message["content"]:
                         if content["type"] == "text":
                             user_text_messages.insert(0, content["text"])
-            # If we have any user text content, push an LLMMessagesFrame
+            # If we have any user text content, push a context frame with the simplified context.
             if user_text_messages:
                 logger.debug(f"User text messages: {user_text_messages}")
                 user_message = " ".join(reversed(user_text_messages))
@@ -110,7 +100,7 @@ class StatementJudgeContextFilter(FrameProcessor):
                 if last_assistant_message:
                     messages.append(last_assistant_message)
                 messages.append({"role": "user", "content": user_message})
-                await self.push_frame(LLMMessagesFrame(messages))
+                await self.push_frame(OpenAILLMContextFrame(OpenAILLMContext(messages)))
 
 
 class CompletenessCheck(FrameProcessor):
@@ -296,7 +286,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
     # This turns the LLM context into an inference request to classify the user's speech
     # as complete or incomplete.
-    statement_judge_context_filter = StatementJudgeContextFilter(notifier=notifier)
+    statement_judge_context_filter = StatementJudgeContextFilter()
 
     # This sends a UserStoppedSpeakingFrame and triggers the notifier event
     completeness_check = CompletenessCheck(notifier=notifier)
@@ -316,7 +306,6 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     async def pass_only_llm_trigger_frames(frame):
         return (
             isinstance(frame, OpenAILLMContextFrame)
-            or isinstance(frame, LLMMessagesFrame)
             or isinstance(frame, StartInterruptionFrame)
             or isinstance(frame, StopInterruptionFrame)
             or isinstance(frame, FunctionCallInProgressFrame)
@@ -331,14 +320,14 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
             ParallelPipeline(
                 [
                     # Ignore everything except an OpenAILLMContextFrame. Pass a specially constructed
-                    # LLMMessagesFrame to the statement classifier LLM. The only frame this
+                    # simplified context frame to the statement classifier LLM. The only frame this
                     # sub-pipeline will output is a UserStoppedSpeakingFrame.
                     statement_judge_context_filter,
                     statement_llm,
                     completeness_check,
                 ],
                 [
-                    # Block everything except OpenAILLMContextFrame and LLMMessagesFrame
+                    # Block everything except frames that trigger LLM inference.
                     FunctionFilter(filter=pass_only_llm_trigger_frames),
                     llm,
                     bot_output_gate,  # Buffer all llm/tts output until notified.

examples/foundational/22c-natural-conversation-mixed-llms.py

@@ -19,7 +19,6 @@ from pipecat.frames.frames import (
     Frame,
     FunctionCallInProgressFrame,
     FunctionCallResultFrame,
-    LLMMessagesFrame,
     StartFrame,
     StartInterruptionFrame,
     StopInterruptionFrame,
@@ -266,10 +265,6 @@ Please be very concise in your responses. Unless you are explicitly asked to do
 
 
 class StatementJudgeContextFilter(FrameProcessor):
-    def __init__(self, notifier: BaseNotifier, **kwargs):
-        super().__init__(**kwargs)
-        self._notifier = notifier
-
     async def process_frame(self, frame: Frame, direction: FrameDirection):
         await super().process_frame(frame, direction)
         # We must not block system frames.
@@ -277,14 +272,8 @@ class StatementJudgeContextFilter(FrameProcessor):
             await self.push_frame(frame, direction)
             return
 
-        # Just treat an LLMMessagesFrame as complete, no matter what.
-        if isinstance(frame, LLMMessagesFrame):
-            await self._notifier.notify()
-            return
-
-        # Otherwise, we only want to handle OpenAILLMContextFrames, and only want to push a simple
-        # messages frame that contains a system prompt and the most recent user messages,
-        # concatenated.
+        # We only want to handle OpenAILLMContextFrames, and only want to push through a simplified
+        # context frame that contains a system prompt and the most recent user messages,
         if isinstance(frame, OpenAILLMContextFrame):
             # Take text content from the most recent user messages.
             messages = frame.context.messages
@@ -301,7 +290,7 @@ class StatementJudgeContextFilter(FrameProcessor):
                     for content in message["content"]:
                         if content["type"] == "text":
                             user_text_messages.insert(0, content["text"])
-            # If we have any user text content, push an LLMMessagesFrame
+            # If we have any user text content, push a context frame with the simplified context.
             if user_text_messages:
                 user_message = " ".join(reversed(user_text_messages))
                 logger.debug(f"!!! {user_message}")
@@ -314,7 +303,7 @@ class StatementJudgeContextFilter(FrameProcessor):
                 if last_assistant_message:
                     messages.append(last_assistant_message)
                 messages.append({"role": "user", "content": user_message})
-                await self.push_frame(LLMMessagesFrame(messages))
+                await self.push_frame(OpenAILLMContextFrame(OpenAILLMContext(messages)))
 
 
 class CompletenessCheck(FrameProcessor):
@@ -499,7 +488,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
     # This turns the LLM context into an inference request to classify the user's speech
     # as complete or incomplete.
-    statement_judge_context_filter = StatementJudgeContextFilter(notifier=notifier)
+    statement_judge_context_filter = StatementJudgeContextFilter()
 
     # This sends a UserStoppedSpeakingFrame and triggers the notifier event
     completeness_check = CompletenessCheck(notifier=notifier)
@@ -522,7 +511,6 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     async def pass_only_llm_trigger_frames(frame):
         return (
             isinstance(frame, OpenAILLMContextFrame)
-            or isinstance(frame, LLMMessagesFrame)
             or isinstance(frame, StartInterruptionFrame)
             or isinstance(frame, StopInterruptionFrame)
             or isinstance(frame, FunctionCallInProgressFrame)
@@ -542,14 +530,14 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
                 ],
                 [
                     # Ignore everything except an OpenAILLMContextFrame. Pass a specially constructed
-                    # LLMMessagesFrame to the statement classifier LLM. The only frame this
+                    # simplified context frame to the statement classifier LLM. The only frame this
                     # sub-pipeline will output is a UserStoppedSpeakingFrame.
                     statement_judge_context_filter,
                     statement_llm,
                     completeness_check,
                 ],
                 [
-                    # Block everything except OpenAILLMContextFrame and LLMMessagesFrame
+                    # Block everything except frames that trigger LLM inference.
                     FunctionFilter(filter=pass_only_llm_trigger_frames),
                     llm,
                     bot_output_gate,  # Buffer all llm/tts output until notified.

pyproject.toml

@@ -33,7 +33,9 @@ dependencies = [
     "pyloudnorm~=0.1.1",
     "resampy~=0.4.3",
     "soxr~=0.5.0",
-    "openai>=1.74.0,<2",
+    "openai>=1.74.0,<=1.99.1",
+    # Pinning numba (resampy dep) to resolve a package dependency
+    "numba==0.61.2",
 ]
 
 [project.urls]

scripts/evals/run-release-evals.py

@@ -93,7 +93,7 @@ TESTS_14 = [
     ("14p-function-calling-gemini-vertex-ai.py", PROMPT_WEATHER, EVAL_WEATHER),
     ("14q-function-calling-qwen.py", PROMPT_WEATHER, EVAL_WEATHER),
     ("14r-function-calling-aws.py", PROMPT_WEATHER, EVAL_WEATHER),
-    ("14v-function-calling-openai.py.py", PROMPT_WEATHER, EVAL_WEATHER),
+    ("14v-function-calling-openai.py", PROMPT_WEATHER, EVAL_WEATHER),
     # Currently not working.
     # ("14c-function-calling-together.py", PROMPT_WEATHER, EVAL_WEATHER),
     # ("14k-function-calling-cerebras.py", PROMPT_WEATHER, EVAL_WEATHER),

src/pipecat/frames/frames.py

@@ -478,6 +478,12 @@ class TranscriptionUpdateFrame(DataFrame):
 class LLMMessagesFrame(DataFrame):
     """Frame containing LLM messages for chat completion.
 
+    .. deprecated:: 0.0.79
+        This class is deprecated and will be removed in a future version.
+        Instead, use either:
+        - `LLMMessagesUpdateFrame` with `run_llm=True`
+        - `OpenAILLMContextFrame` with desired messages in a new context
+
     A frame containing a list of LLM messages. Used to signal that an LLM
     service should run a chat completion and emit an LLMFullResponseStartFrame,
     TextFrames and an LLMFullResponseEndFrame. Note that the `messages`
@@ -490,6 +496,20 @@ class LLMMessagesFrame(DataFrame):
 
     messages: List[dict]
 
+    def __post_init__(self):
+        super().__post_init__()
+        import warnings
+
+        warnings.simplefilter("always")
+        warnings.warn(
+            "LLMMessagesFrame is deprecated and will be removed in a future version. "
+            "Instead, use either "
+            "`LLMMessagesUpdateFrame` with `run_llm=True`, or "
+            "`OpenAILLMContextFrame` with desired messages in a new context",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+
 
 @dataclass
 class LLMMessagesAppendFrame(DataFrame):

src/pipecat/processors/aggregators/llm_response.py

@@ -12,6 +12,7 @@ LLM processing, and text-to-speech components in conversational AI pipelines.
 """
 
 import asyncio
+import warnings
 from abc import abstractmethod
 from dataclasses import dataclass
 from typing import Dict, List, Literal, Optional, Set
@@ -994,6 +995,10 @@ class LLMAssistantContextAggregator(LLMContextResponseAggregator):
 class LLMUserResponseAggregator(LLMUserContextAggregator):
     """User response aggregator that outputs LLMMessagesFrame instead of context frames.
 
+    .. deprecated:: 0.0.79
+        This class is deprecated and will be removed in a future version.
+        Use `LLMUserContextAggregator` or another LLM-specific subclass instead.
+
     This aggregator extends LLMUserContextAggregator but pushes LLMMessagesFrame
     objects downstream instead of OpenAILLMContextFrame objects. This is useful
     when you need message-based output rather than context-based output.
@@ -1013,6 +1018,12 @@ class LLMUserResponseAggregator(LLMUserContextAggregator):
             params: Configuration parameters for aggregation behavior.
             **kwargs: Additional arguments passed to parent class.
         """
+        warnings.warn(
+            "LLMUserResponseAggregator is deprecated and will be removed in a future version. "
+            "Use LLMUserContextAggregator or another LLM-specific subclass instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
         super().__init__(context=OpenAILLMContext(messages), params=params, **kwargs)
 
     async def _process_aggregation(self):
@@ -1027,6 +1038,10 @@ class LLMUserResponseAggregator(LLMUserContextAggregator):
 class LLMAssistantResponseAggregator(LLMAssistantContextAggregator):
     """Assistant response aggregator that outputs LLMMessagesFrame instead of context frames.
 
+    .. deprecated:: 0.0.79
+        This class is deprecated and will be removed in a future version.
+        Use `LLMAssistantContextAggregator` or another LLM-specific subclass instead.
+
     This aggregator extends LLMAssistantContextAggregator but pushes LLMMessagesFrame
     objects downstream instead of OpenAILLMContextFrame objects. This is useful
     when you need message-based output rather than context-based output.
@@ -1046,6 +1061,12 @@ class LLMAssistantResponseAggregator(LLMAssistantContextAggregator):
             params: Configuration parameters for aggregation behavior.
             **kwargs: Additional arguments passed to parent class.
         """
+        warnings.warn(
+            "LLMAssistantResponseAggregator is deprecated and will be removed in a future version. "
+            "Use LLMAssistantContextAggregator or another LLM-specific subclass instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
         super().__init__(context=OpenAILLMContext(messages), params=params, **kwargs)
 
     async def push_aggregation(self):

src/pipecat/processors/aggregators/user_response.py

@@ -12,13 +12,14 @@ in conversational pipelines.
 """
 
 from pipecat.frames.frames import TextFrame
-from pipecat.processors.aggregators.llm_response import LLMUserResponseAggregator
+from pipecat.processors.aggregators.llm_response import LLMUserContextAggregator
+from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
 
 
-class UserResponseAggregator(LLMUserResponseAggregator):
+class UserResponseAggregator(LLMUserContextAggregator):
     """Aggregates user responses into TextFrame objects.
 
-    This aggregator extends LLMUserResponseAggregator to specifically handle
+    This aggregator extends LLMUserContextAggregator to specifically handle
     user input by collecting text responses and outputting them as TextFrame
     objects when the aggregation is complete.
     """
@@ -27,9 +28,9 @@ class UserResponseAggregator(LLMUserResponseAggregator):
         """Initialize the user response aggregator.
 
         Args:
-            **kwargs: Additional arguments passed to parent LLMUserResponseAggregator.
+            **kwargs: Additional arguments passed to parent LLMUserContextAggregator.
         """
-        super().__init__(**kwargs)
+        super().__init__(context=OpenAILLMContext(), **kwargs)
 
     async def push_aggregation(self):
         """Push the aggregated user response as a TextFrame.

src/pipecat/processors/frameworks/langchain.py

@@ -14,9 +14,9 @@ from pipecat.frames.frames import (
     Frame,
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
-    LLMMessagesFrame,
     TextFrame,
 )
+from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContextFrame
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 
 try:
@@ -64,11 +64,11 @@ class LangchainProcessor(FrameProcessor):
         """
         await super().process_frame(frame, direction)
 
-        if isinstance(frame, LLMMessagesFrame):
+        if isinstance(frame, OpenAILLMContextFrame):
             # Messages are accumulated on the context as a list of messages.
             # The last one by the human is the one we want to send to the LLM.
             logger.debug(f"Got transcription frame {frame}")
-            text: str = frame.messages[-1]["content"]
+            text: str = frame.context.messages[-1]["content"]
 
             await self._ainvoke(text.strip())
         else:

src/pipecat/services/anthropic/llm.py

@@ -448,14 +448,16 @@ class AnthropicLLMContext(OpenAILLMContext):
             system: System message content.
         """
         super().__init__(messages=messages, tools=tools, tool_choice=tool_choice)
+        self.__setup_local()
+        self.system = system
 
+    def __setup_local(self):
         # For beta prompt caching. This is a counter that tracks the number of turns
         # we've seen above the cache threshold. We reset this when we reset the
         # messages list. We only care about this number being 0, 1, or 2. But
         # it's easiest just to treat it as a counter.
         self.turns_above_cache_threshold = 0
-
-        self.system = system
+        return
 
     @staticmethod
     def upgrade_to_anthropic(obj: OpenAILLMContext) -> "AnthropicLLMContext":
@@ -472,6 +474,7 @@ class AnthropicLLMContext(OpenAILLMContext):
         logger.debug(f"Upgrading to Anthropic: {obj}")
         if isinstance(obj, OpenAILLMContext) and not isinstance(obj, AnthropicLLMContext):
             obj.__class__ = AnthropicLLMContext
+            obj.__setup_local()
             obj._restructure_from_openai_messages()
         return obj
 

src/pipecat/services/tts_service.py

@@ -269,9 +269,20 @@ class TTSService(AIService):
     async def say(self, text: str):
         """Immediately speak the provided text.
 
+        .. deprecated:: 0.0.79
+            Push a `TTSSpeakFrame` instead to ensure frame ordering is maintained.
+
         Args:
             text: The text to speak.
         """
+        import warnings
+
+        warnings.warn(
+            "`TTSService.say()` is deprecated. Push a `TTSSpeakFrame` instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+
         await self.queue_frame(TTSSpeakFrame(text))
 
     async def process_frame(self, frame: Frame, direction: FrameDirection):

tests/test_langchain.py

@@ -12,7 +12,7 @@ from langchain_core.language_models import FakeStreamingListLLM
 from pipecat.frames.frames import (
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
-    LLMMessagesFrame,
+    OpenAILLMContextAssistantTimestampFrame,
     TextFrame,
     TranscriptionFrame,
     UserStartedSpeakingFrame,
@@ -21,8 +21,12 @@ from pipecat.frames.frames import (
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.processors.aggregators.llm_response import (
     LLMAssistantAggregatorParams,
-    LLMAssistantResponseAggregator,
-    LLMUserResponseAggregator,
+    LLMAssistantContextAggregator,
+    LLMUserContextAggregator,
+)
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
 )
 from pipecat.processors.frame_processor import FrameProcessor
 from pipecat.processors.frameworks.langchain import LangchainProcessor
@@ -63,9 +67,10 @@ class TestLangchain(unittest.IsolatedAsyncioTestCase):
         proc = LangchainProcessor(chain=chain)
         self.mock_proc = self.MockProcessor("token_collector")
 
-        tma_in = LLMUserResponseAggregator(messages)
-        tma_out = LLMAssistantResponseAggregator(
-            messages, params=LLMAssistantAggregatorParams(expect_stripped_words=False)
+        context = OpenAILLMContext()
+        tma_in = LLMUserContextAggregator(context)
+        tma_out = LLMAssistantContextAggregator(
+            context, params=LLMAssistantAggregatorParams(expect_stripped_words=False)
         )
 
         pipeline = Pipeline([tma_in, proc, self.mock_proc, tma_out])
@@ -79,7 +84,8 @@ class TestLangchain(unittest.IsolatedAsyncioTestCase):
         expected_down_frames = [
             UserStartedSpeakingFrame,
             UserStoppedSpeakingFrame,
-            LLMMessagesFrame,
+            OpenAILLMContextFrame,
+            OpenAILLMContextAssistantTimestampFrame,
         ]
         await run_test(
             pipeline,