Update code for tool call

engine/services/base.py

@@ -7,7 +7,7 @@ StreamEngine pattern.
 
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
-from typing import AsyncIterator, Optional, List, Dict, Any
+from typing import AsyncIterator, Optional, List, Dict, Any, Literal
 from enum import Enum
 
 
@@ -52,6 +52,15 @@ class LLMMessage:
         return d
 
 
+@dataclass
+class LLMStreamEvent:
+    """Structured LLM stream event."""
+
+    type: Literal["text_delta", "tool_call", "done"]
+    text: Optional[str] = None
+    tool_call: Optional[Dict[str, Any]] = None
+
+
 @dataclass
 class TTSChunk:
     """TTS audio chunk."""
@@ -170,7 +179,7 @@ class BaseLLMService(ABC):
         messages: List[LLMMessage],
         temperature: float = 0.7,
         max_tokens: Optional[int] = None
-    ) -> AsyncIterator[str]:
+    ) -> AsyncIterator[LLMStreamEvent]:
         """
         Generate response in streaming mode.
         
@@ -180,7 +189,7 @@ class BaseLLMService(ABC):
             max_tokens: Maximum tokens to generate
             
         Yields:
-            Text chunks as they are generated
+            Stream events (text delta/tool call/done)
         """
         pass