dv-pipecat-flows 0.0.23.dev2__tar.gz → 0.0.23.dev4__tar.gz

{dv_pipecat_flows-0.0.23.dev2/src/dv_pipecat_flows.egg-info → dv_pipecat_flows-0.0.23.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.23.dev2
+Version: 0.0.23.dev4
 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-0.0.23.dev2 → dv_pipecat_flows-0.0.23.dev4/src/dv_pipecat_flows.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-flows
-Version: 0.0.23.dev2
+Version: 0.0.23.dev4
 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-0.0.23.dev2 → dv_pipecat_flows-0.0.23.dev4}/src/pipecat_flows/adapters.py

@@ -31,6 +31,27 @@ from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
 from pipecat_flows.types import FlowsDirectFunctionWrapper, FlowsFunctionSchema
 
 
+def _ensure_array_items(schema: Any) -> None:
+    """Recursively default `items` to string for array properties missing it.
+
+    Gemini rejects the whole request (400 INVALID_ARGUMENT) if any array-typed
+    parameter lacks `items` — a schema shape OpenAI/Azure silently tolerate, so
+    dashboard-authored tools can carry it. Adding `items` is valid JSON Schema
+    for every provider, so this is safe to apply universally.
+    """
+    if not isinstance(schema, dict):
+        return
+    schema_type = schema.get("type")
+    is_array = schema_type == "array" or (
+        isinstance(schema_type, list) and "array" in schema_type
+    )
+    if is_array and "items" not in schema:
+        schema["items"] = {"type": "string"}
+    for prop in (schema.get("properties") or {}).values():
+        _ensure_array_items(prop)
+    _ensure_array_items(schema.get("items"))
+
+
 class LLMAdapter:
     """Base adapter for LLM-specific format handling.
 
@@ -439,6 +460,12 @@ class UniversalLLMAdapter(LLMAdapter):
         # Pipecat's universal LLMContext.
         if not formatted_functions:
             return NOT_GIVEN
+        # The universal path bypasses provider adapters (e.g. GeminiAdapter),
+        # so array params must be sanitized here before pipecat core's
+        # provider adapters serialize them verbatim.
+        for func in formatted_functions.standard_tools or []:
+            for prop in (func.properties or {}).values():
+                _ensure_array_items(prop)
         return formatted_functions
 
     def format_summary_message(self, summary: str) -> dict:
@@ -965,26 +992,11 @@ class GeminiAdapter(LLMAdapter):
         # array-typed parameter lacks `items` — a schema shape OpenAI/Azure
         # silently tolerate, so dashboard-authored tools can carry it.
         for func in gemini_functions:
-            self._ensure_array_items(func.get("parameters"))
+            _ensure_array_items(func.get("parameters"))
 
         # Format as Gemini expects - an array with a single object containing function_declarations
         return [{"function_declarations": gemini_functions}]
 
-    @classmethod
-    def _ensure_array_items(cls, schema: Any) -> None:
-        """Recursively default `items` to string for array properties missing it."""
-        if not isinstance(schema, dict):
-            return
-        schema_type = schema.get("type")
-        is_array = schema_type == "array" or (
-            isinstance(schema_type, list) and "array" in schema_type
-        )
-        if is_array and "items" not in schema:
-            schema["items"] = {"type": "string"}
-        for prop in (schema.get("properties") or {}).values():
-            cls._ensure_array_items(prop)
-        cls._ensure_array_items(schema.get("items"))
-
     def format_summary_message(self, summary: str) -> dict:
         """Format summary as a user message for Gemini.
 

{dv_pipecat_flows-0.0.23.dev2 → dv_pipecat_flows-0.0.23.dev4}/src/pipecat_flows/manager.py

@@ -832,9 +832,15 @@ class FlowManager:
                     )
                 else:
                     # Node function - defer to pipecat's default run_llm logic
-                    # This waits for all parallel function calls to complete
+                    # This waits for all parallel function calls to complete.
+                    # A tool (e.g. a widget send) can opt out of the follow-up LLM
+                    # completion by tagging its result with _suppress_followup_llm;
+                    # pop it so the sentinel never leaks into the LLM context.
+                    suppress_followup_llm = (
+                        isinstance(result, dict) and result.pop("_suppress_followup_llm", False)
+                    )
                     properties = FunctionCallResultProperties(
-                        run_llm=None,
+                        run_llm=False if suppress_followup_llm else None,
                         on_context_updated=None,
                     )