dv-pipecat-flows 0.0.22.dev918__tar.gz → 0.0.22.dev1116__tar.gz

{dv_pipecat_flows-0.0.22.dev918/src/dv_pipecat_flows.egg-info → dv_pipecat_flows-0.0.22.dev1116}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.22.dev918
+Version: 0.0.22.dev1116
 Summary: Conversation Flow management for Pipecat AI applications
 License: BSD 2-Clause License
 Project-URL: Source, https://github.com/pipecat-ai/pipecat-flows

{dv_pipecat_flows-0.0.22.dev918 → dv_pipecat_flows-0.0.22.dev1116/src/dv_pipecat_flows.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.22.dev918
+Version: 0.0.22.dev1116
 Summary: Conversation Flow management for Pipecat AI applications
 License: BSD 2-Clause License
 Project-URL: Source, https://github.com/pipecat-ai/pipecat-flows

{dv_pipecat_flows-0.0.22.dev918 → dv_pipecat_flows-0.0.22.dev1116}/src/pipecat_flows/adapters.py

@@ -672,6 +672,12 @@ def create_adapter(llm, context_aggregator) -> LLMAdapter:
     llm_type = type(llm).__name__
     llm_class = type(llm)
 
+    if llm_type == "OpenAIResponsesLLMService":
+        # Responses API service uses universal LLMContext — should have been caught
+        # by the LLMContextAggregatorPair check above, but handle explicitly as a fallback
+        logger.debug("Creating universal adapter for OpenAI Responses API service")
+        return UniversalLLMAdapter()
+
     if llm_type == "OpenAILLMService":
         logger.debug("Creating OpenAI adapter")
         return OpenAIAdapter()

{dv_pipecat_flows-0.0.22.dev918 → dv_pipecat_flows-0.0.22.dev1116}/src/pipecat_flows/manager.py

@@ -1135,12 +1135,23 @@ In all of these cases, you can provide a `name` in your new node's config for de
                 standard_functions, original_configs=functions_list
             )
 
+            # Collect mirror_context messages from tts_say pre-actions
+            mirror_messages = []
+            for action in node_config.get("pre_actions", []):
+                if (
+                    action.get("type") == "tts_say"
+                    and action.get("mirror_context", True)
+                    and action.get("text")
+                ):
+                    mirror_messages.append({"role": "assistant", "content": action["text"]})
+
             # Update LLM context
             await self._update_llm_context(
                 role_messages=node_config.get("role_messages"),
                 task_messages=node_config["task_messages"],
                 functions=formatted_tools,
                 strategy=node_config.get("context_strategy"),
+                mirror_messages=mirror_messages,
             )
             logger.debug("Updated LLM context")
 
@@ -1186,6 +1197,7 @@ In all of these cases, you can provide a `name` in your new node's config for de
         task_messages: List[dict],
         functions: List[dict],
         strategy: Optional[ContextStrategyConfig] = None,
+        mirror_messages: Optional[List[dict]] = None,
     ) -> None:
         """Update LLM context with new messages and functions.
 
@@ -1194,6 +1206,9 @@ In all of these cases, you can provide a `name` in your new node's config for de
             task_messages: Task messages to add to context.
             functions: New functions to make available.
             strategy: Optional context update configuration.
+            mirror_messages: Optional assistant messages from tts_say pre-actions
+                with mirror_context=True. These are appended after task_messages
+                so the LLM sees what was spoken to the user.
 
         Raises:
             FlowError: If context update fails.
@@ -1243,6 +1258,12 @@ In all of these cases, you can provide a `name` in your new node's config for de
             # Add task messages
             messages.extend(task_messages)
 
+            # Add mirror messages (tts_say text mirrored as assistant context)
+            # Placed after task_messages so the LLM sees: [system instructions] + [what was spoken]
+            if mirror_messages:
+                messages.extend(mirror_messages)
+                logger.debug(f"Added {len(mirror_messages)} mirror message(s) to LLM context")
+
             # For first node or RESET/RESET_WITH_SUMMARY strategy, use update frame
             frame_type = (
                 LLMMessagesUpdateFrame