fix: patch some stuff

Author: cpacker

letta/agents/letta_agent_v3.py

@@ -385,7 +385,11 @@ async def _step(
             self.response_messages.extend(persisted_messages[new_message_idx:])
 
             if llm_adapter.supports_token_streaming():
-                if persisted_messages[-1].role != "approval" and tool_call is not None:
+                # Stream the tool return if a tool was actually executed.
+                # In the normal streaming path, the tool call is surfaced via the streaming interface
+                # (llm_adapter.tool_call), so don't rely solely on the local `tool_call` variable.
+                has_tool_return = any(m.role == "tool" for m in persisted_messages)
+                if persisted_messages[-1].role != "approval" and has_tool_return:
                     tool_return = [msg for msg in persisted_messages if msg.role == "tool"][-1].to_letta_messages()[0]
                     if include_return_message_types is None or tool_return.message_type in include_return_message_types:
                         yield tool_return

letta/schemas/letta_message_content.py

@@ -1,6 +1,7 @@
 from enum import Enum
 from typing import Annotated, List, Literal, Optional, Union
 
+from openai.types import Reasoning
 from pydantic import BaseModel, Field
 
 
@@ -274,14 +275,18 @@ class SummarizedReasoningContent(MessageContent):
     encrypted_content: str = Field(default=None, description="The encrypted reasoning content.")
 
     # Temporary stop-gap until the SDKs are updated
-    def to_reasoning_content(self):
+    def to_reasoning_content(self) -> Optional[ReasoningContent]:
         # Merge the summary parts with a '\n' join
-        combined_summary = "\n\n".join([s.text for s in self.summary])
-        return ReasoningContent(
-            is_native=True,
-            reasoning=combined_summary,
-            signature=self.encrypted_content,
-        )
+        parts = [s.text for s in self.summary if s.text != ""]
+        if not parts or len(parts) == 0:
+            return None
+        else:
+            combined_summary = "\n\n".join(parts)
+            return ReasoningContent(
+                is_native=True,
+                reasoning=combined_summary,
+                signature=self.encrypted_content,
+            )
 
 
 LettaMessageContentUnion = Annotated[

letta/schemas/message.py

@@ -389,19 +389,20 @@ def _convert_reasoning_messages(
             elif isinstance(content_part, SummarizedReasoningContent):
                 # TODO remove the cast and just return the native type
                 casted_content_part = content_part.to_reasoning_content()
-                messages.append(
-                    ReasoningMessage(
-                        id=self.id,
-                        date=self.created_at,
-                        reasoning=casted_content_part.reasoning,
-                        source="reasoner_model",  # TODO do we want to tag like this?
-                        signature=casted_content_part.signature,
-                        name=self.name,
-                        otid=otid,
-                        step_id=self.step_id,
-                        is_err=self.is_err,
+                if casted_content_part is not None:
+                    messages.append(
+                        ReasoningMessage(
+                            id=self.id,
+                            date=self.created_at,
+                            reasoning=casted_content_part.reasoning,
+                            source="reasoner_model",  # TODO do we want to tag like this?
+                            signature=casted_content_part.signature,
+                            name=self.name,
+                            otid=otid,
+                            step_id=self.step_id,
+                            is_err=self.is_err,
+                        )
                     )
-                )
 
             elif isinstance(content_part, RedactedReasoningContent):
                 # "native" redacted/hidden COT