Apply ensure_ascii=False to remaining LLM services and fix changelog format

changelog/3457.changed.md

@@ -1,16 +1 @@
-# Reduce Call Tool Result Context Size by Allowing UTF-8 in JSON Serialization
+- Changed tool result JSON serialization to use `ensure_ascii=False`, preserving UTF-8 characters instead of escaping them. This reduces context size and token usage for non-English languages.
-
-This PR changes tool result serialization to prevent UTF-8 code points from being escaped during serialization. This drastically reduces the context size when returning a response that contains languages other than English.
-
-We have been running a monkey-patched version in production and it helped us improve the agent accuracy and control cost better.
-
-```
->>> data = { "message": "أهلًا بالعالم" }
->>> json.dumps(data)
-'{"message": "\\u0623\\u0647\\u0644\\u064b\\u0627 \\u0628\\u0627\\u0644\\u0639\\u0627\\u0644\\u0645"}'
->>> 
->>> 
->>> 
->>> json.dumps(data, ensure_ascii=False)
-'{"message": "أهلًا بالعالم"}'
-```

src/pipecat/services/aws/nova_sonic/llm.py

@@ -1044,7 +1044,9 @@ class AWSNovaSonicLLMService(LLMService):
                     "toolResult": {
                         "promptName": self._prompt_name,
                         "contentName": content_name,
-                        "content": json.dumps(result) if isinstance(result, dict) else result,
+                        "content": json.dumps(result, ensure_ascii=False)
+                        if isinstance(result, dict)
+                        else result,
                     }
                 }
             }

src/pipecat/services/google/llm.py

@@ -200,7 +200,9 @@ class GoogleAssistantContextAggregator(OpenAIAssistantContextAggregator):
             if message.role == "user":
                 for part in message.parts:
                     if part.function_response and part.function_response.id == tool_call_id:
-                        part.function_response.response = {"value": json.dumps(result)}
+                        part.function_response.response = {
+                            "value": json.dumps(result, ensure_ascii=False)
+                        }
 
 
 @dataclass

src/pipecat/services/grok/realtime/llm.py

@@ -939,7 +939,7 @@ class GrokRealtimeLLMService(LLMService):
         item = events.ConversationItem(
             type="function_call_output",
             call_id=tool_call_id,
-            output=json.dumps(result),
+            output=json.dumps(result, ensure_ascii=False),
         )
         await self.send_client_event(events.ConversationItemCreateEvent(item=item))
 

src/pipecat/services/openai/realtime/llm.py

@@ -1128,7 +1128,7 @@ class OpenAIRealtimeLLMService(LLMService):
         item = events.ConversationItem(
             type="function_call_output",
             call_id=tool_call_id,
-            output=json.dumps(result),
+            output=json.dumps(result, ensure_ascii=False),
         )
         await self.send_client_event(events.ConversationItemCreateEvent(item=item))