Enhance conversation history and runtime variable management

- Update ConversationRecorder to include source and nodeId metadata in transcripts for better context tracking.
- Introduce optional variable handling in DynamicVariableStore, allowing for unset variables to be rendered as empty without raising errors.
- Refactor WorkflowBrain to apply turn configurations and manage interaction policies dynamically, improving agent responsiveness.
- Implement tests to ensure proper handling of updated session variables and workflow metadata in various scenarios.

backend/services/pipecat/turn_config.py

@@ -7,6 +7,11 @@ from typing import Any
 from pipecat.audio.turn.smart_turn.local_smart_turn_v3 import LocalSmartTurnAnalyzerV3
 from pipecat.audio.vad.silero import SileroVADAnalyzer
 from pipecat.audio.vad.vad_analyzer import VADParams
+from pipecat.processors.aggregators.llm_context import LLMContext
+from pipecat.processors.aggregators.llm_response_universal import (
+    LLMUserAggregator,
+    LLMUserAggregatorParams,
+)
 from pipecat.turns.user_start import (
     TranscriptionUserTurnStartStrategy,
     VADUserTurnStartStrategy,
@@ -39,18 +44,21 @@ def _value(config: dict[str, Any], snake: str, camel: str, default: Any) -> Any:
     return config.get(snake, config.get(camel, default))
 
 
-def create_vad_analyzer(turn_config: dict[str, Any]) -> SileroVADAnalyzer:
+def create_vad_params(turn_config: dict[str, Any]) -> VADParams:
+    """Translate product settings into Pipecat's runtime VAD parameters."""
     vad = _section(turn_config, "vad", "vad")
-    return SileroVADAnalyzer(
-        params=VADParams(
-            confidence=float(vad.get("confidence", DEFAULT_VAD["confidence"])),
-            start_secs=float(_value(vad, "start_secs", "startSecs", 0.2)),
-            stop_secs=float(_value(vad, "stop_secs", "stopSecs", 0.2)),
-            min_volume=float(_value(vad, "min_volume", "minVolume", 0.6)),
-        )
+    return VADParams(
+        confidence=float(vad.get("confidence", DEFAULT_VAD["confidence"])),
+        start_secs=float(_value(vad, "start_secs", "startSecs", 0.2)),
+        stop_secs=float(_value(vad, "stop_secs", "stopSecs", 0.2)),
+        min_volume=float(_value(vad, "min_volume", "minVolume", 0.6)),
     )
 
 
+def create_vad_analyzer(turn_config: dict[str, Any]) -> SileroVADAnalyzer:
+    return SileroVADAnalyzer(params=create_vad_params(turn_config))
+
+
 def create_user_turn_strategies(
     turn_config: dict[str, Any], *, enable_interruptions: bool
 ) -> UserTurnStrategies:
@@ -87,3 +95,34 @@ def create_user_turn_strategies(
             )
         ]
     return UserTurnStrategies(start=start, stop=stop)
+
+
+class ConfigurableLLMUserAggregator(LLMUserAggregator):
+    """LLM user aggregator with one stable project-level runtime update API.
+
+    Pipecat 1.5 exposes ``UserTurnController.update_strategies`` but does not
+    surface it on ``LLMUserAggregator``. Keeping that version-specific bridge
+    here prevents Workflow orchestration from depending on Pipecat internals.
+    VAD threshold updates still use Pipecat's public ``VADParamsUpdateFrame``.
+    """
+
+    def __init__(
+        self,
+        context: LLMContext,
+        *,
+        params: LLMUserAggregatorParams | None = None,
+        **kwargs: Any,
+    ) -> None:
+        super().__init__(context, params=params, **kwargs)
+
+    async def apply_turn_strategies(
+        self,
+        turn_config: dict[str, Any],
+        *,
+        enable_interruptions: bool,
+    ) -> None:
+        strategies = create_user_turn_strategies(
+            turn_config,
+            enable_interruptions=enable_interruptions,
+        )
+        await self._user_turn_controller.update_strategies(strategies)