dv-pipecat-ai 0.0.85.dev10__py3-none-any.whl → 0.0.85.dev12__py3-none-any.whl

{dv_pipecat_ai-0.0.85.dev10.dist-info → dv_pipecat_ai-0.0.85.dev12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-ai
-Version: 0.0.85.dev10
+Version: 0.0.85.dev12
 Summary: An open source framework for voice (and multimodal) assistants
 License-Expression: BSD-2-Clause
 Project-URL: Source, https://github.com/pipecat-ai/pipecat

{dv_pipecat_ai-0.0.85.dev10.dist-info → dv_pipecat_ai-0.0.85.dev12.dist-info}/RECORD

@@ -1,4 +1,4 @@
-dv_pipecat_ai-0.0.85.dev10.dist-info/licenses/LICENSE,sha256=DWY2QGf2eMCFhuu2ChairtT6CB7BEFffNVhXWc4Od08,1301
+dv_pipecat_ai-0.0.85.dev12.dist-info/licenses/LICENSE,sha256=DWY2QGf2eMCFhuu2ChairtT6CB7BEFffNVhXWc4Od08,1301
 pipecat/__init__.py,sha256=j0Xm6adxHhd7D06dIyyPV_GlBYLlBnTAERVvD_jAARQ,861
 pipecat/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/adapters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -77,7 +77,7 @@ pipecat/extensions/voicemail/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NM
 pipecat/extensions/voicemail/voicemail_detector.py,sha256=g3L1m3cPJzsadeB5a8WRC9klH0D8m7xfPgB2YEaL6Do,29983
 pipecat/frames/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/frames/frames.proto,sha256=JXZm3VXLR8zMOUcOuhVoe2mhM3MQIQGMJXLopdJO_5Y,839
-pipecat/frames/frames.py,sha256=oqoo7p-uJOqak50mxhCGq7S0TusM0I4qp3QAftKHQnw,45428
+pipecat/frames/frames.py,sha256=2aXsBpZB6dU7I1PZRFh8RqALDktxxoRgZPa1pf_cdCM,45804
 pipecat/frames/protobufs/frames_pb2.py,sha256=VHgGV_W7qQ4sfQK6RHb5_DggLm3PiSYMr6aBZ8_p1cQ,2590
 pipecat/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/metrics/metrics.py,sha256=bdZNciEtLTtA-xgoKDz2RJAy6fKrXkTwz3pryVHzc2M,2713
@@ -111,7 +111,7 @@ pipecat/processors/idle_frame_processor.py,sha256=z8AuhGap61lA5K35P6XCaOpn4kkmK_
 pipecat/processors/logger.py,sha256=VGNwxQSc_F0rS3KBmfqas7f5aFyRQKfeljozOxfGXk4,2393
 pipecat/processors/producer_processor.py,sha256=iIIOHZd77APvUGP7JqFbznAHUnCULcq_qYiSEjwXHcc,3265
 pipecat/processors/text_transformer.py,sha256=LnfWJYzntJhZhrQ1lgSSY4D4VbHtrQJgrC227M69ZYU,1718
-pipecat/processors/transcript_processor.py,sha256=CG9yej6WOiy_HhagNXjxkISHkHii0JDfK_V6opseC2E,11740
+pipecat/processors/transcript_processor.py,sha256=SDbqFLzasptZWqeiA6BESJ4hky-Uh-ZUNYBZR0q0Dnw,12508
 pipecat/processors/two_stage_user_idle_processor.py,sha256=uf2aZh_lfW-eMxmFogP3R4taAJ1yXOSqjKsR7oXtD0Y,2938
 pipecat/processors/user_idle_processor.py,sha256=PQItBx5bL1y_lFTtHf0zgLubjCXv60jU1jrVtm-K4bg,9268
 pipecat/processors/aggregators/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -119,7 +119,7 @@ pipecat/processors/aggregators/dtmf_aggregator.py,sha256=nngjLiaOtcZtuCNpYPyfUVL
 pipecat/processors/aggregators/gated.py,sha256=tii0sRrBkRW6y9Xq5iTWPnqlOEejU4VqPIPtdOa61pc,3073
 pipecat/processors/aggregators/gated_openai_llm_context.py,sha256=cr6MT8J6SpPzZbppKPOKe3_pt_5qXC9g6a4wvZDyrec,3005
 pipecat/processors/aggregators/llm_context.py,sha256=eDf1cQElcISLx3onaA9LCWuepzb2G_JGszLzpNXggXo,9723
-pipecat/processors/aggregators/llm_response.py,sha256=0StzYtq7EzlAFSWp10I0yY0pV1jysw1ySEWv5R50h_s,47360
+pipecat/processors/aggregators/llm_response.py,sha256=W0bqc5IZTE1cIB8egBefUOaPZiLklzge5npFWvAbCcw,48248
 pipecat/processors/aggregators/llm_response_universal.py,sha256=fBnB3rZVdxj4iEKIWcnR7yTpqyKupbcg7IUv6XVxrDQ,34287
 pipecat/processors/aggregators/openai_llm_context.py,sha256=cC8DXdVPERRN04i0i-1Ys6kusvnbMALeH-Z8Pu5K684,12999
 pipecat/processors/aggregators/sentence.py,sha256=E7e3knfQl6HEGpYMKPklF1aO_gOn-rr7SnynErwfkQk,2235
@@ -289,6 +289,8 @@ pipecat/services/rime/tts.py,sha256=XHMSnQUi7gMtWF42u4rBVv6oBDor4KkwkL7O-Sj9MPo,
 pipecat/services/riva/__init__.py,sha256=rObSsj504O_TMXhPBg_ymqKslZBhovlR-A0aaRZ0O6A,276
 pipecat/services/riva/stt.py,sha256=dtg8toijmexWB3uipw0EQ7ov3DFgHj40kFFv1Zadmmc,25116
 pipecat/services/riva/tts.py,sha256=idbqx3I2NlWCXtrIFsjEaYapxA3BLIA14ai3aMBh-2w,8158
+pipecat/services/salesforce/__init__.py,sha256=OFvYbcvCadYhcKdBAVLj3ZUXVXQ1HyVyhgxIFf6_Thg,173
+pipecat/services/salesforce/llm.py,sha256=mpozkzldgz3plbMOJcKddiyJxn7x4qqPuJVn22a41Ag,23009
 pipecat/services/sambanova/__init__.py,sha256=oTXExLic-qTcsfsiWmssf3Elclf3IIWoN41_2IpoF18,128
 pipecat/services/sambanova/llm.py,sha256=5XVfPLEk__W8ykFqLdV95ZUhlGGkAaJwmbciLdZYtTc,8976
 pipecat/services/sambanova/stt.py,sha256=ZZgEZ7WQjLFHbCko-3LNTtVajjtfUvbtVLtFcaNadVQ,2536
@@ -361,6 +363,7 @@ pipecat/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/utils/base_object.py,sha256=62e5_0R_rcQe-JdzUM0h1wtv1okw-0LPyG78ZKkDyzE,5963
 pipecat/utils/logger_config.py,sha256=5-RmvReZIINeqSXz3ALhEIiMZ_azmpOxnlIkdyCjWWk,5606
 pipecat/utils/network.py,sha256=RRQ7MmTcbeXBJ2aY5UbMCQ6elm5B8Rxkn8XqkJ9S0Nc,825
+pipecat/utils/redis.py,sha256=JmBaC1yY6e8qygUQkAER3DNFCYSCH18hd7NN9qqjDMU,1677
 pipecat/utils/string.py,sha256=TskK9KxQSwbljct0J6y9ffkRcx4xYjTtPhFjEL4M1i8,6720
 pipecat/utils/time.py,sha256=lirjh24suz9EI1pf2kYwvAYb3I-13U_rJ_ZRg3nRiGs,1741
 pipecat/utils/utils.py,sha256=T2y1Mcd9yWiZiIToUiRkhW-n7EFf8juk3kWX3TF8XOQ,2451
@@ -381,7 +384,7 @@ pipecat/utils/tracing/service_decorators.py,sha256=HwDCqLGijhYD3F8nxDuQmEw-YkRw0
 pipecat/utils/tracing/setup.py,sha256=7TEgPNpq6M8lww8OQvf0P9FzYc5A30xICGklVA-fua0,2892
 pipecat/utils/tracing/turn_context_provider.py,sha256=ikon3plFOx0XbMrH6DdeHttNpb-U0gzMZIm3bWLc9eI,2485
 pipecat/utils/tracing/turn_trace_observer.py,sha256=dma16SBJpYSOE58YDWy89QzHyQFc_9gQZszKeWixuwc,9725
-dv_pipecat_ai-0.0.85.dev10.dist-info/METADATA,sha256=ezbvZ9D9Q9E1aVPhwoNcHu02GKAveWpHvFp0lgahMVc,32858
-dv_pipecat_ai-0.0.85.dev10.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-dv_pipecat_ai-0.0.85.dev10.dist-info/top_level.txt,sha256=kQzG20CxGf-nSsHmtXHx3hY2-8zHA3jYg8jk0TajqXc,8
-dv_pipecat_ai-0.0.85.dev10.dist-info/RECORD,,
+dv_pipecat_ai-0.0.85.dev12.dist-info/METADATA,sha256=pj9DcBZS3A1SFUXm5aZDJS5K9e4YldFoPBcrrBiFxPI,32858
+dv_pipecat_ai-0.0.85.dev12.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+dv_pipecat_ai-0.0.85.dev12.dist-info/top_level.txt,sha256=kQzG20CxGf-nSsHmtXHx3hY2-8zHA3jYg8jk0TajqXc,8
+dv_pipecat_ai-0.0.85.dev12.dist-info/RECORD,,

pipecat/frames/frames.py

@@ -451,12 +451,14 @@ class TranscriptionMessage:
         content: The message content/text.
         user_id: Optional identifier for the user.
         timestamp: Optional timestamp when the message was created.
+        message_id: Optional unique identifier for tracking and dropping messages.
     """
 
     role: Literal["user", "assistant"]
     content: str
     user_id: Optional[str] = None
     timestamp: Optional[str] = None
+    message_id: Optional[int] = None
 
 
 @dataclass
@@ -510,6 +512,17 @@ class TranscriptionUpdateFrame(DataFrame):
         return f"{self.name}(pts: {pts}, messages: {len(self.messages)})"
 
 
+@dataclass
+class TranscriptDropFrame(DataFrame):
+    """Frame indicating previously emitted transcript chunks should be discarded.
+
+    Parameters:
+        transcript_ids: List of frame/message identifiers to drop.
+    """
+
+    transcript_ids: List[int]
+
+
 @dataclass
 class LLMContextFrame(Frame):
     """Frame containing a universal LLM context.

pipecat/processors/aggregators/llm_response.py

@@ -51,6 +51,7 @@ from pipecat.frames.frames import (
     StartFrame,
     StartInterruptionFrame,
     TextFrame,
+    TranscriptDropFrame,
     TranscriptionFrame,
     UserImageRawFrame,
     UserStartedSpeakingFrame,
@@ -446,6 +447,7 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
         self._latest_final_transcript = ""
         self._last_user_speaking_time = 0
         self._last_aggregation_push_time = 0
+        self._pending_transcription_ids: List[int] = []
 
     async def reset(self):
         """Reset the aggregation state and interruption strategies."""
@@ -453,6 +455,7 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
         self._was_bot_speaking = False
         self._seen_interim_results = False
         self._waiting_for_aggregation = False
+        self._pending_transcription_ids.clear()
         [await s.reset() for s in self._interruption_strategies]
 
     async def handle_aggregation(self, aggregation: str):
@@ -548,7 +551,8 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
                     await self._process_aggregation()
                 else:
                     self.logger.debug("Interruption conditions not met - not pushing aggregation")
-                    # Don't process aggregation, just reset it
+                    # Don't process aggregation, discard pending transcriptions and reset
+                    await self._discard_pending_transcriptions("interruption_conditions_not_met")
                     await self.reset()
             else:
                 if trigger_interruption:
@@ -614,10 +618,18 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
         for s in self.interruption_strategies:
             await s.append_audio(frame.audio, frame.sample_rate)
 
+    async def _discard_pending_transcriptions(self, reason: str):
+        """Notify upstream processors that pending transcripts should be dropped."""
+        if self._pending_transcription_ids:
+            drop_frame = TranscriptDropFrame(transcript_ids=list(self._pending_transcription_ids))
+            await self.push_frame(drop_frame, FrameDirection.UPSTREAM)
+        self._pending_transcription_ids.clear()
+
     async def _handle_user_started_speaking(self, frame: UserStartedSpeakingFrame):
         if len(self._aggregation) > 0:
             self.logger.debug(f"Dropping {self._aggregation}")
             self._aggregation = ""
+            await self._discard_pending_transcriptions("user_started_speaking")
         self._latest_final_transcript = ""
         self._last_user_speaking_time = time.time()
         self._user_speaking = True
@@ -662,6 +674,7 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
             return
 
         self._aggregation += f" {text}" if self._aggregation else text
+        self._pending_transcription_ids.append(frame.id)
         # We just got a final result, so let's reset interim results.
         self._seen_interim_results = False
 
@@ -791,6 +804,7 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
             if self._bot_speaking and not self._params.enable_emulated_vad_interruptions:
                 # If emulated VAD interruptions are disabled and bot is speaking, ignore
                 logger.debug("Ignoring user speaking emulation, bot is speaking.")
+                await self._discard_pending_transcriptions("emulated_vad_ignored")
                 await self.reset()
             else:
                 # Either bot is not speaking, or emulated VAD interruptions are enabled

pipecat/processors/transcript_processor.py

@@ -20,6 +20,7 @@ from pipecat.frames.frames import (
     EndFrame,
     Frame,
     StartInterruptionFrame,
+    TranscriptDropFrame,
     TranscriptionFrame,
     TranscriptionMessage,
     TranscriptionUpdateFrame,
@@ -44,6 +45,7 @@ class BaseTranscriptProcessor(FrameProcessor):
         super().__init__(**kwargs)
         self._processed_messages: List[TranscriptionMessage] = []
         self._register_event_handler("on_transcript_update")
+        self._register_event_handler("on_transcript_drop")
 
     async def _emit_update(self, messages: List[TranscriptionMessage]):
         """Emit transcript updates for new messages.
@@ -57,6 +59,19 @@ class BaseTranscriptProcessor(FrameProcessor):
             await self._call_event_handler("on_transcript_update", update_frame)
             await self.push_frame(update_frame)
 
+    async def _handle_transcript_drop(self, frame: TranscriptDropFrame):
+        """Handle transcript drop notifications by removing stored messages."""
+        if not frame.transcript_ids:
+            return
+
+        await self._call_event_handler("on_transcript_drop", frame)
+
+        drop_ids = set(frame.transcript_ids)
+        if drop_ids:
+            self._processed_messages = [
+                msg for msg in self._processed_messages if msg.message_id not in drop_ids
+            ]
+
 
 class UserTranscriptProcessor(BaseTranscriptProcessor):
     """Processes user transcription frames into timestamped conversation messages."""
@@ -72,9 +87,15 @@ class UserTranscriptProcessor(BaseTranscriptProcessor):
 
         if isinstance(frame, TranscriptionFrame):
             message = TranscriptionMessage(
-                role="user", user_id=frame.user_id, content=frame.text, timestamp=frame.timestamp
+                role="user",
+                user_id=frame.user_id,
+                content=frame.text,
+                timestamp=frame.timestamp,
+                message_id=frame.id,
             )
             await self._emit_update([message])
+        elif isinstance(frame, TranscriptDropFrame):
+            await self._handle_transcript_drop(frame)
 
         await self.push_frame(frame, direction)
 

pipecat/services/salesforce/__init__.py

@@ -0,0 +1,9 @@
+#
+# Copyright (c) 2024–2025, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+from .llm import SalesforceAgentLLMService
+
+__all__ = ["SalesforceAgentLLMService"]

pipecat/services/salesforce/llm.py

@@ -0,0 +1,587 @@
+#
+# Copyright (c) 2024–2025, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+"""Salesforce Agent API LLM service implementation."""
+
+import asyncio
+import json
+import time
+from dataclasses import dataclass
+from typing import AsyncGenerator, Dict, Optional
+
+import httpx
+from loguru import logger
+
+from pipecat.frames.frames import (
+    Frame,
+    LLMFullResponseEndFrame,
+    LLMFullResponseStartFrame,
+    LLMMessagesFrame,
+    LLMTextFrame,
+    LLMUpdateSettingsFrame,
+)
+from pipecat.processors.aggregators.openai_llm_context import (
+    OpenAILLMContext,
+    OpenAILLMContextFrame,
+)
+from pipecat.processors.frame_processor import FrameDirection
+from pipecat.services.llm_service import LLMService
+from pipecat.services.openai.llm import (
+    OpenAIAssistantContextAggregator,
+    OpenAIContextAggregatorPair,
+    OpenAIUserContextAggregator,
+)
+from pipecat.processors.aggregators.llm_response import (
+    LLMAssistantAggregatorParams,
+    LLMUserAggregatorParams,
+)
+from env_config import api_config
+from pipecat.utils.redis import create_async_redis_client
+
+
+@dataclass
+class SalesforceSessionInfo:
+    """Information about an active Salesforce Agent session."""
+
+    session_id: str
+    agent_id: str
+    created_at: float
+    last_used: float
+
+
+class SalesforceAgentLLMService(LLMService):
+    """Salesforce Agent API LLM service implementation.
+
+    This service integrates with Salesforce Agent API to provide conversational
+    AI capabilities using Salesforce's Agentforce platform.
+    """
+
+    def __init__(
+        self,
+        *,
+        model: str = "salesforce-agent",
+        session_timeout_secs: float = 3600.0,
+        agent_id: str = api_config.SALESFORCE_AGENT_ID,
+        org_domain: str = api_config.SALESFORCE_ORG_DOMAIN,
+        client_id: str = api_config.SALESFORCE_CLIENT_ID,
+        client_secret: str = api_config.SALESFORCE_CLIENT_SECRET,
+        api_host: str = api_config.SALESFORCE_API_HOST,
+        redis_url: Optional[str] = None,
+        **kwargs,
+    ):
+        """Initialize Salesforce Agent LLM service.
+
+        Reads configuration from environment variables:
+        - SALESFORCE_AGENT_ID: The Salesforce agent ID to interact with
+        - SALESFORCE_ORG_DOMAIN: Salesforce org domain (e.g., https://myorg.my.salesforce.com)
+        - SALESFORCE_CLIENT_ID: Connected app client ID for OAuth
+        - SALESFORCE_CLIENT_SECRET: Connected app client secret for OAuth
+        - SALESFORCE_API_HOST: Salesforce API host base URL (e.g., https://api.salesforce.com)
+
+        Args:
+            model: The model name (defaults to "salesforce-agent").
+            session_timeout_secs: Session timeout in seconds (default: 1 hour).
+            agent_id: Salesforce agent ID. Defaults to SALESFORCE_AGENT_ID.
+            org_domain: Salesforce org domain. Defaults to SALESFORCE_ORG_DOMAIN.
+            client_id: Salesforce connected app client ID. Defaults to SALESFORCE_CLIENT_ID.
+            client_secret: Salesforce connected app client secret. Defaults to SALESFORCE_CLIENT_SECRET.
+            api_host: Salesforce API host base URL. Defaults to SALESFORCE_API_HOST.
+            redis_url: Optional Redis URL override for token caching.
+            **kwargs: Additional arguments passed to parent LLMService.
+        """
+        # Initialize parent LLM service
+        super().__init__(**kwargs)
+
+        self._agent_id = agent_id
+        self._org_domain = org_domain 
+        self._client_id = client_id
+        self._client_secret = client_secret
+        self._api_host = api_host
+
+
+        # Validate required environment variables
+        required_vars = {
+            "SALESFORCE_AGENT_ID": self._agent_id,
+            "SALESFORCE_ORG_DOMAIN": self._org_domain,
+            "SALESFORCE_API_HOST": self._api_host,
+            "SALESFORCE_CLIENT_ID": self._client_id,
+            "SALESFORCE_CLIENT_SECRET": self._client_secret,
+        }
+
+        missing_vars = [var for var, value in required_vars.items() if not value]
+        if missing_vars:
+            raise ValueError(f"Missing required environment variables: {', '.join(missing_vars)}")
+
+        logger.info(f"Salesforce LLM initialized - Agent ID: {self._agent_id}")
+
+        self._session_timeout_secs = session_timeout_secs
+
+        if redis_url is not None:
+            self._redis_url = redis_url
+        else:
+            self._redis_url = getattr(api_config, "REDIS_URL", None)
+        self._redis_client = None
+        self._redis_client_init_attempted = False
+        self._token_cache_key = f"salesforce_agent_access_token:{self._agent_id}"
+        self._token_cache_leeway_secs = 300
+        self._sequence_counter = 0
+        self._warmup_task: Optional[asyncio.Task] = None
+
+        # Session management
+        self._sessions: Dict[str, SalesforceSessionInfo] = {}
+        self._current_session_id: Optional[str] = None
+
+        # HTTP client for API calls
+        self._http_client = httpx.AsyncClient(
+            timeout=30.0,
+            limits=httpx.Limits(
+                max_keepalive_connections=10,
+                max_connections=100,
+                keepalive_expiry=None,
+            ),
+        )
+
+        self._schedule_session_warmup()
+        
+
+    async def __aenter__(self):
+        """Async context manager entry."""
+        await self.ensure_session_ready()
+        return self
+
+    async def __aexit__(self, exc_type, exc_val, exc_tb):
+        """Async context manager exit."""
+        if self._warmup_task:
+            try:
+                await asyncio.shield(self._warmup_task)
+            except Exception as exc:  # pragma: no cover - warmup best effort
+                logger.debug(f"Salesforce warmup task failed during exit: {exc}")
+            finally:
+                self._warmup_task = None
+
+        await self._cleanup_sessions()
+        await self._http_client.aclose()
+
+        if self._redis_client:
+            close_coro = getattr(self._redis_client, "close", None)
+            if callable(close_coro):
+                try:
+                    await close_coro()
+                except Exception as exc:  # pragma: no cover - best effort cleanup
+                    logger.debug(f"Failed to close Redis client cleanly: {exc}")
+        self._redis_client = None
+        self._redis_client_init_attempted = False
+
+    def can_generate_metrics(self) -> bool:
+        """Check if this service can generate processing metrics."""
+        return True
+
+    def _schedule_session_warmup(self):
+        """Kick off background warm-up if an event loop is running."""
+        try:
+            loop = asyncio.get_running_loop()
+        except RuntimeError:
+            return
+
+        if loop.is_closed():
+            return
+
+        async def _warmup():
+            try:
+                await self.ensure_session_ready()
+            except Exception as exc:  # pragma: no cover - warmup best effort
+                logger.warning(f"Salesforce warmup failed: {exc}")
+                raise
+
+        task = loop.create_task(_warmup())
+
+        def _on_done(warmup_task: asyncio.Task):
+            if warmup_task.cancelled():
+                logger.debug("Salesforce warmup task cancelled")
+            elif warmup_task.exception():
+                logger.warning(f"Salesforce warmup task error: {warmup_task.exception()}")
+            self._warmup_task = None
+
+        task.add_done_callback(_on_done)
+        self._warmup_task = task
+
+    def _get_redis_client(self):
+        """Return a Redis client for token caching if configured."""
+        if self._redis_client is None and not self._redis_client_init_attempted:
+            self._redis_client_init_attempted = True
+            self._redis_client = create_async_redis_client(
+                self._redis_url, decode_responses=True, encoding="utf-8", logger=logger
+            )
+
+        return self._redis_client
+
+    async def _get_cached_access_token(self) -> Optional[str]:
+        """Return cached access token from Redis."""
+        redis_client = self._get_redis_client()
+        if not redis_client:
+            return None
+
+        try:
+            return await redis_client.get(self._token_cache_key)
+        except Exception as exc:  # pragma: no cover - cache failures shouldn't break flow
+            logger.warning(f"Failed to read Salesforce token from Redis: {exc}")
+            return None
+
+    async def _set_cached_access_token(self, token: str, expires_in: Optional[int]):
+        """Persist access token in Redis with TTL matching Salesforce expiry."""
+        redis_client = self._get_redis_client()
+        if not redis_client:
+            return
+
+        ttl_seconds = 3600
+        if expires_in is not None:
+            try:
+                ttl_seconds = max(int(expires_in) - self._token_cache_leeway_secs, 30)
+            except (TypeError, ValueError):
+                logger.debug("Unable to parse Salesforce token expiry; falling back to default TTL")
+
+        try:
+            await redis_client.set(self._token_cache_key, token, ex=ttl_seconds)
+        except Exception as exc:  # pragma: no cover - cache failures shouldn't break flow
+            logger.warning(f"Failed to store Salesforce token in Redis: {exc}")
+
+    async def _get_access_token(self) -> str:
+        """Get OAuth access token using client credentials."""
+        cached_token = await self._get_cached_access_token()
+        if cached_token:
+            return cached_token
+
+        token_url = f"{self._org_domain}/services/oauth2/token"
+        data = {
+            "grant_type": "client_credentials",
+            "client_id": self._client_id,
+            "client_secret": self._client_secret,
+        }
+
+        try:
+            response = await self._http_client.post(token_url, data=data)
+            response.raise_for_status()
+            token_data = response.json()
+            access_token = token_data["access_token"]
+            await self._set_cached_access_token(access_token, token_data.get("expires_in"))
+            return access_token
+        except Exception as e:
+            logger.error(f"Failed to get access token: {e}")
+            raise
+
+    async def _create_session(self) -> str:
+        """Create a new Salesforce Agent session."""
+        access_token = await self._get_access_token()
+        session_url = f"{self._api_host}/einstein/ai-agent/v1/agents/{self._agent_id}/sessions"
+
+        headers = {
+            "Authorization": f"Bearer {access_token}",
+            "Content-Type": "application/json",
+        }
+
+        external_session_key = f"pipecat-{int(time.time())}-{id(self)}"
+
+        payload = {
+            "externalSessionKey": external_session_key,
+            "instanceConfig": {"endpoint": self._org_domain},
+            "tz": "America/Los_Angeles",
+            "variables": [{"name": "$Context.EndUserLanguage", "type": "Text", "value": "en_US"}],
+            "featureSupport": "Streaming",
+            "streamingCapabilities": {"chunkTypes": ["Text"]},
+            "bypassUser": True,
+        }
+
+        try:
+            response = await self._http_client.post(session_url, headers=headers, json=payload)
+            response.raise_for_status()
+            session_data = response.json()
+            session_id = session_data["sessionId"]
+
+            # Store session info
+            current_time = time.time()
+            self._sessions[session_id] = SalesforceSessionInfo(
+                session_id=session_id,
+                agent_id=self._agent_id,
+                created_at=current_time,
+                last_used=current_time,
+            )
+
+            logger.debug(f"Created Salesforce Agent session: {session_id}")
+            return session_id
+
+        except Exception as e:
+            logger.error(f"Failed to create Salesforce Agent session: {e}")
+            raise
+
+    async def _get_or_create_session(self) -> str:
+        """Get existing session or create a new one."""
+        current_time = time.time()
+
+        # Check if current session is still valid
+        if self._current_session_id and self._current_session_id in self._sessions:
+            session = self._sessions[self._current_session_id]
+            if current_time - session.last_used < self._session_timeout_secs:
+                session.last_used = current_time
+                return self._current_session_id
+            else:
+                # Session expired, remove it
+                self._sessions.pop(self._current_session_id, None)
+                self._current_session_id = None
+
+        # Create new session
+        self._current_session_id = await self._create_session()
+        return self._current_session_id
+
+    async def ensure_session_ready(self) -> str:
+        """Ensure a Salesforce session is ready for use."""
+        return await self._get_or_create_session()
+
+    async def _cleanup_sessions(self):
+        """Clean up expired sessions."""
+        current_time = time.time()
+        expired_sessions = []
+
+        for session_id, session in self._sessions.items():
+            if current_time - session.last_used > self._session_timeout_secs:
+                expired_sessions.append(session_id)
+
+        for session_id in expired_sessions:
+            try:
+                # End the session via API
+                access_token = await self._get_access_token()
+                url = f"{self._api_host}/einstein/ai-agent/v1/sessions/{session_id}"
+                headers = {
+                    "Authorization": f"Bearer {access_token}",
+                    "x-session-end-reason": "UserRequest",
+                }
+                await self._http_client.delete(url, headers=headers)
+            except Exception as e:
+                logger.warning(f"Failed to end session {session_id}: {e}")
+            finally:
+                self._sessions.pop(session_id, None)
+                if self._current_session_id == session_id:
+                    self._current_session_id = None
+
+    def _extract_user_message(self, context: OpenAILLMContext) -> str:
+        """Extract the last user message from context.
+        
+        Similar to Vistaar pattern - extract only the most recent user message.
+        
+        Args:
+            context: The OpenAI LLM context containing messages.
+            
+        Returns:
+            The last user message as a string.
+        """
+        messages = context.get_messages()
+        
+        # Find the last user message (iterate in reverse for efficiency)
+        for message in reversed(messages):
+            if message.get("role") == "user":
+                content = message.get("content", "")
+                
+                # Handle content that might be a list (for multimodal messages)
+                if isinstance(content, list):
+                    text_parts = [
+                        item.get("text", "") for item in content if item.get("type") == "text"
+                    ]
+                    content = " ".join(text_parts)
+                
+                if isinstance(content, str):
+                    return content.strip()
+        
+        return ""
+
+    def _generate_sequence_id(self) -> int:
+        """Generate a sequence ID for the message."""
+        self._sequence_counter += 1
+        return self._sequence_counter
+
+    async def _stream_salesforce_response(self, session_id: str, user_message: str) -> AsyncGenerator[str, None]:
+        """Stream response from Salesforce Agent API."""
+        access_token = await self._get_access_token()
+        url = f"{self._api_host}/einstein/ai-agent/v1/sessions/{session_id}/messages/stream"
+        
+        headers = {
+            "Authorization": f"Bearer {access_token}",
+            "Content-Type": "application/json",
+            "Accept": "text/event-stream",
+        }
+
+        message_data = {
+            "message": {
+                "sequenceId": self._generate_sequence_id(),
+                "type": "Text",
+                "text": user_message
+            },
+            "variables": [
+                {
+                    "name": "$Context.EndUserLanguage",
+                    "type": "Text",
+                    "value": "en_US"
+                }
+            ]
+        }
+
+        try:
+            logger.info(f"🌐 Salesforce API request: {user_message[:50]}...")
+            async with self._http_client.stream("POST", url, headers=headers, json=message_data) as response:
+                response.raise_for_status()
+                
+                async for line in response.aiter_lines():
+                    if not line:
+                        continue
+
+                    # Parse SSE format
+                    if line.startswith("data: "):
+                        try:
+                            data = json.loads(line[6:])
+                            message = data.get("message", {})
+                            message_type = message.get("type")
+
+                            if message_type == "TextChunk":
+                                content = message.get("text", "") or message.get("message", "")
+                                if content:
+                                    yield content
+                            elif message_type == "EndOfTurn":
+                                logger.info("🏁 Salesforce response complete")
+                                break
+                            elif message_type == "Inform":
+                                # Skip INFORM events to avoid duplication
+                                continue
+
+                        except json.JSONDecodeError as e:
+                            logger.warning(f"JSON decode error: {e}, line: {line}")
+                            continue
+
+        except Exception as e:
+            logger.error(f"Failed to stream from Salesforce Agent API: {e}")
+            raise
+
+    async def _process_context(self, context: OpenAILLMContext):
+        """Process the LLM context and generate streaming response.
+
+        Args:
+            context: The OpenAI LLM context containing messages to process.
+        """
+        logger.info(f"🔄 Salesforce processing context with {len(context.get_messages())} messages")
+        
+        # Extract user message from context first
+        user_message = self._extract_user_message(context)
+        
+        if not user_message:
+            logger.warning("Salesforce: No user message found in context")
+            return
+        
+        try:
+            logger.info(f"🎯 Salesforce extracted query: {user_message}")
+            
+            # Start response 
+            await self.push_frame(LLMFullResponseStartFrame())
+            await self.push_frame(LLMFullResponseStartFrame(),FrameDirection.UPSTREAM)
+            await self.start_processing_metrics()
+            await self.start_ttfb_metrics()
+            
+            # Get or create session
+            session_id = await self._get_or_create_session()
+            
+            first_chunk = True
+            
+            # Stream the response
+            async for text_chunk in self._stream_salesforce_response(session_id, user_message):
+                if first_chunk:
+                    await self.stop_ttfb_metrics()
+                    first_chunk = False
+                
+                # Push each text chunk as it arrives
+                await self.push_frame(LLMTextFrame(text=text_chunk))
+            
+        except Exception as e:
+            logger.error(f"Salesforce context processing error: {type(e).__name__}: {str(e)}")
+            import traceback
+            logger.error(f"Salesforce traceback: {traceback.format_exc()}")
+            raise
+        finally:
+            await self.stop_processing_metrics()
+            await self.push_frame(LLMFullResponseEndFrame())
+            await self.push_frame(LLMFullResponseEndFrame(), FrameDirection.UPSTREAM)
+
+    async def process_frame(self, frame: Frame, direction: FrameDirection):
+        """Process frames for LLM completion requests.
+
+        Args:
+            frame: The frame to process.
+            direction: The direction of frame processing.
+        """
+        context = None
+        if isinstance(frame, OpenAILLMContextFrame):
+            context = frame.context
+            logger.info(f"🔍 Received OpenAILLMContextFrame with {len(context.get_messages())} messages")
+        elif isinstance(frame, LLMMessagesFrame):
+            context = OpenAILLMContext.from_messages(frame.messages)
+            logger.info(f"🔍 Received LLMMessagesFrame with {len(frame.messages)} messages")
+        elif isinstance(frame, LLMUpdateSettingsFrame):
+            # Call super for settings frames and update settings
+            await super().process_frame(frame, direction)
+            settings = frame.settings
+            logger.debug(f"Updated Salesforce settings: {settings}")
+        else:
+            # For non-context frames, call super and push them downstream
+            await super().process_frame(frame, direction)
+            await self.push_frame(frame, direction)
+
+        if context:
+            try:
+                await self._process_context(context)
+            except httpx.TimeoutException:
+                logger.error("Timeout while processing Salesforce request")
+                await self._call_event_handler("on_completion_timeout")
+            except Exception as e:
+                logger.error(f"Error processing Salesforce request: {e}")
+                raise
+
+    def create_context_aggregator(
+        self,
+        context: OpenAILLMContext,
+        *,
+        user_params: LLMUserAggregatorParams = LLMUserAggregatorParams(),
+        assistant_params: LLMAssistantAggregatorParams = LLMAssistantAggregatorParams(),
+    ) -> OpenAIContextAggregatorPair:
+        """Create context aggregators for Salesforce LLM.
+
+        Since Salesforce uses OpenAI-compatible message format, we reuse OpenAI's
+        context aggregators directly
+
+        Args:
+            context: The LLM context to create aggregators for.
+            user_params: Parameters for user message aggregation.
+            assistant_params: Parameters for assistant message aggregation.
+
+        Returns:
+            OpenAIContextAggregatorPair: A pair of OpenAI context aggregators,
+            compatible with Salesforce's OpenAI-like message format.
+        """
+        context.set_llm_adapter(self.get_llm_adapter())
+        user = OpenAIUserContextAggregator(context, params=user_params)
+        assistant = OpenAIAssistantContextAggregator(context, params=assistant_params)
+        return OpenAIContextAggregatorPair(_user=user, _assistant=assistant)
+
+    def get_llm_adapter(self):
+        """Get the LLM adapter for this service."""
+        from pipecat.adapters.services.open_ai_adapter import OpenAILLMAdapter
+        return OpenAILLMAdapter()
+
+    async def close(self):
+        """Close the HTTP client when the service is destroyed."""
+        await self._cleanup_sessions()
+        await self._http_client.aclose()
+
+    def __del__(self):
+        """Ensure the client is closed on deletion."""
+        try:
+            asyncio.create_task(self._http_client.aclose())
+        except:
+            pass

pipecat/utils/redis.py

@@ -0,0 +1,58 @@
+"""Async Redis helper utilities."""
+
+from __future__ import annotations
+
+from typing import Any, Optional, TYPE_CHECKING
+
+from urllib.parse import urlparse
+
+try:
+    import redis.asyncio as redis
+except ImportError:  # pragma: no cover - Redis is optional
+    redis = None
+
+
+if TYPE_CHECKING:  # pragma: no cover - typing aid
+    from redis.asyncio import Redis
+
+
+def create_async_redis_client(
+    url: Optional[str],
+    *,
+    decode_responses: bool = True,
+    encoding: str = "utf-8",
+    logger: Optional[Any] = None,
+    **kwargs,
+) -> Optional["Redis"]:
+    """Return a configured async Redis client or None if unavailable.
+
+    Args:
+        url: Redis connection URL.
+        decode_responses: Whether to decode responses to str.
+        encoding: Character encoding to use with decoded responses.
+        logger: Optional logger supporting .warning() for diagnostics.
+        **kwargs: Additional keyword arguments forwarded to Redis.from_url.
+    """
+    if redis is None:
+        return None
+
+    if not url or url in {"redis_url", "REDIS_URL"}:
+        return None
+
+    parsed = urlparse(url)
+    connection_kwargs = {
+        "decode_responses": decode_responses,
+        "encoding": encoding,
+    }
+    connection_kwargs.update(kwargs)
+
+    if parsed.scheme == "rediss":
+        connection_kwargs.setdefault("ssl_cert_reqs", "none")
+        connection_kwargs.setdefault("ssl_check_hostname", False)
+
+    try:
+        return redis.Redis.from_url(url, **connection_kwargs)
+    except Exception as exc:  # pragma: no cover - best effort logging
+        if logger is not None:
+            logger.warning(f"Failed to create Redis client: {exc}")
+        return None