dv-pipecat-flows 0.0.0.dev2072__tar.gz → 0.0.0.dev2087__tar.gz

{dv_pipecat_flows-0.0.0.dev2072/src/dv_pipecat_flows.egg-info → dv_pipecat_flows-0.0.0.dev2087}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.0.dev2072
+Version: 0.0.0.dev2087
 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-0.0.0.dev2072 → dv_pipecat_flows-0.0.0.dev2087/src/dv_pipecat_flows.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.0.dev2072
+Version: 0.0.0.dev2087
 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-0.0.0.dev2072 → dv_pipecat_flows-0.0.0.dev2087}/src/pipecat_flows/actions.py

@@ -312,11 +312,16 @@ class ActionManager:
 
         Args:
             action: Action configuration containing:
-                - text: Text to speak (required)
+                - text: Text to speak (required, resolved)
+                - text_raw: Raw template with {{var}} markers (optional)
                 - use_cache: Use caching for TTS audio (optional, default: False)
 
-        Note: Caching functionality must be provided via dependency injection through
-        flow_manager.state['say_with_cache_handler'] if use_cache=True.
+        Caching is opt-in via dependency injection: when use_cache=True and
+        flow_manager.state['say_with_cache_handler'] is set, the handler is
+        invoked as `await handler(text=..., text_raw=...) -> bool`. Returning
+        True means the handler served the audio downstream (and pushed an
+        ActionFinishedFrame downstream itself). Returning False (or raising)
+        falls back to live TTS via TTSSpeakFrame.
 
         To prevent the LLM from immediately responding after TTS in pre-actions,
         set respond_immediately=false in the node configuration.
@@ -327,18 +332,32 @@ class ActionManager:
             return
 
         use_cache = action.get("use_cache", False)
+        cache_handler = (
+            self._flow_manager.state.get("say_with_cache_handler") if use_cache else None
+        )
 
         try:
-            # For now, always use the standard TTS path to avoid pipeline issues
-            # Caching can be implemented at the TTS service level instead
-
             # Mark that we're starting the action
             self._increment_ongoing_actions_count()
 
-            # Queue the TTS frame
-            await self._task.queue_frame(TTSSpeakFrame(text=text))
-
-            # Queue frame marking the end of the action
+            served_from_cache = False
+            if cache_handler is not None:
+                try:
+                    text_raw = action.get("text_raw") or text
+                    served_from_cache = bool(await cache_handler(text=text, text_raw=text_raw))
+                except Exception as e:
+                    logger.error(f"tts_say cache handler raised, falling back to live TTS: {e}")
+                    served_from_cache = False
+
+            if not served_from_cache:
+                # Live TTS path
+                await self._task.queue_frame(TTSSpeakFrame(text=text))
+
+            # Queue frame marking the end of the action. Even on the cached
+            # path this is safe: the frame is queued at the source and has to
+            # traverse the entire pipeline to reach our downstream filter,
+            # while the cached audio was pushed downstream of the TTS service
+            # and reaches the wire sooner.
             await self._task.queue_frame(ActionFinishedFrame())
 
         except Exception as e: