luxx/services/chat.py

@@ -1,15 +1,9 @@
-"""Chat service module - Refactored with step-by-step flow"""
+"""Chat service module"""
 import json
 import uuid
 import logging
-from typing import List, Dict, Any, AsyncGenerator, Tuple, Optional
-
-# For Python < 3.9 compatibility
-try:
-    from typing import List
-except ImportError:
-    pass
 
+from typing import List, Dict,AsyncGenerator
 from luxx.models import Conversation, Message, LLMProvider
 from luxx.tools.executor import ToolExecutor
 from luxx.tools.core import registry
@@ -17,274 +11,85 @@ from luxx.services.llm_client import LLMClient
 from luxx.database import SessionLocal
 
 logger = logging.getLogger(__name__)
-
+# Maximum iterations to prevent infinite loops
 MAX_ITERATIONS = 20
 
 
 def _sse_event(event: str, data: dict) -> str:
-    """Format SSE event string."""
+    """Format a Server-Sent Event string."""
     return f"event: {event}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
 
 
-def get_llm_client(conversation: Conversation = None) -> Tuple[LLMClient, Optional[int]]:
+def get_llm_client(conversation: Conversation = None):
-    """Get LLM client from conversation's provider."""
+    """Get LLM client, optionally using conversation's provider. Returns (client, max_tokens)"""
+    max_tokens = None
     if conversation and conversation.provider_id:
         db = SessionLocal()
         try:
             provider = db.query(LLMProvider).filter(LLMProvider.id == conversation.provider_id).first()
             if provider:
-                return LLMClient(
+                max_tokens = provider.max_tokens
+                client = LLMClient(
                     api_key=provider.api_key,
                     api_url=provider.base_url,
                     model=provider.default_model
-                ), provider.max_tokens
+                )
+                return client, max_tokens
         finally:
             db.close()
-    return LLMClient(), None
-
-
-class StreamState:
-    """Holds streaming state across iterations."""
     
-    def __init__(self):
+    # Fallback to global config
-        self.messages: List[Dict] = []
+    client = LLMClient()
-        self.all_steps: List[Dict] = []
+    return client, max_tokens
-        self.all_tool_calls: List[Dict] = []
-        self.all_tool_results: List[Dict] = []
-        self.step_index: int = 0
-        self.total_usage: Dict[str, int] = {
-            "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0
-        }
-        # Current iteration state
-        self.full_content: str = ""
-        self.full_thinking: str = ""
-        self.tool_calls_list: List[Dict] = []
-        self.thinking_step_id: Optional[str] = None
-        self.thinking_step_idx: Optional[int] = None
-        self.text_step_id: Optional[str] = None
-        self.text_step_idx: Optional[int] = None
 
 
 class ChatService:
-    """Chat service with step-by-step flow architecture."""
+    """Chat service with tool support"""
     
     def __init__(self):
         self.tool_executor = ToolExecutor()
     
-    # ==================== Step 1: Initialize ====================
+    def build_messages(
-    
+        self,
-    def build_messages(self, conversation: Conversation, user_message: str) -> List[Dict[str, str]]:
+        conversation: Conversation,
-        """Build message list including user message."""
+        include_system: bool = True
+    ) -> List[Dict[str, str]]:
+        """Build message list"""
+        
         messages = []
-        if conversation.system_prompt:
+        
-            messages.append({"role": "system", "content": conversation.system_prompt})
+        if include_system and conversation.system_prompt:
+            messages.append({
+                "role": "system",
+                "content": conversation.system_prompt
+            })
         
         db = SessionLocal()
         try:
-            for msg in db.query(Message).filter(
+            db_messages = db.query(Message).filter(
                 Message.conversation_id == conversation.id
-            ).order_by(Message.created_at).all():
+            ).order_by(Message.created_at).all()
+            
+            for msg in db_messages:
+                # Parse JSON content if possible
                 try:
-                    obj = json.loads(msg.content) if msg.content else {}
+                    content_obj = json.loads(msg.content) if msg.content else {}
-                    content = obj.get("text", msg.content) if isinstance(obj, dict) else msg.content
+                    if isinstance(content_obj, dict):
+                        content = content_obj.get("text", msg.content)
+                    else:
+                        content = msg.content
                 except (json.JSONDecodeError, TypeError):
                     content = msg.content
-                messages.append({"role": msg.role, "content": content})
+                
+                messages.append({
+                    "role": msg.role,
+                    "content": content
+                })
         finally:
             db.close()
         
-        messages.append({"role": "user", "content": json.dumps({"text": user_message, "attachments": []})})
         return messages
     
-    def init_stream_state(self, conversation: Conversation, user_message: str, enabled_tools: list) -> Tuple[StreamState, LLMClient, Dict, str, Optional[int]]:
-        """Initialize streaming state. Returns: (state, llm, tools, model, max_tokens)"""
-        state = StreamState()
-        state.messages = self.build_messages(conversation, user_message)
-        
-        tools = [t for t in registry.list_all() if t.get("function", {}).get("name") in enabled_tools] if enabled_tools else []
-        
-        llm, max_tokens = get_llm_client(conversation)
-        model = conversation.model or llm.default_model or "gpt-4"
-        
-        tool_context = {"workspace": None, "user_id": None, "username": None, "user_permission_level": 1}
-        
-        return state, llm, tools, model, max_tokens, tool_context
-    
-    # ==================== Step 2: Stream LLM ====================
-    
-    def parse_sse_line(self, sse_line: str) -> Tuple[Optional[str], Optional[str]]:
-        """Parse SSE line into (event_type, data_str)."""
-        event_type = data_str = None
-        for line in sse_line.strip().split('\n'):
-            if line.startswith('event: '):
-                event_type = line[7:].strip()
-            elif line.startswith('data: '):
-                data_str = line[6:].strip()
-        return event_type, data_str
-    
-    def stream_llm_response(self, llm, model: str, messages: List[Dict], tools: list,
-                           temperature: float, max_tokens: int, thinking_enabled: bool):
-        """
-        Stream LLM response and yield (sse_line, parsed_chunk) pairs.
-        """
-        for sse_line in llm.stream_call(
-            model=model, messages=messages, tools=tools,
-            temperature=temperature, max_tokens=max_tokens or 8192,
-            thinking_enabled=thinking_enabled
-        ):
-            _, data_str = self.parse_sse_line(sse_line)
-            chunk = None
-            if data_str:
-                try:
-                    chunk = json.loads(data_str)
-                except json.JSONDecodeError:
-                    pass
-            yield sse_line, chunk
-    
-    def process_delta(self, state: StreamState, delta: dict) -> List[str]:
-        """
-        Process a single delta, return list of SSE event strings.
-        """
-        events = []
-        
-        # Handle thinking/reasoning
-        reasoning = delta.get("reasoning_content", "")
-        if reasoning:
-            if not state.full_thinking:
-                state.thinking_step_idx = state.step_index
-                state.thinking_step_id = f"step-{state.step_index}"
-                state.step_index += 1
-            state.full_thinking += reasoning
-            events.append(_sse_event("process_step", {
-                "step": {"id": state.thinking_step_id, "index": state.thinking_step_idx, "type": "thinking", "content": state.full_thinking}
-            }))
-        
-        # Handle content
-        content = delta.get("content", "")
-        if content:
-            if not state.full_content:
-                state.text_step_idx = state.step_index
-                state.text_step_id = f"step-{state.step_index}"
-                state.step_index += 1
-            state.full_content += content
-            events.append(_sse_event("process_step", {
-                "step": {"id": state.text_step_id, "index": state.text_step_idx, "type": "text", "content": state.full_content}
-            }))
-        
-        # Handle tool calls
-        for tc in delta.get("tool_calls", []):
-            idx = tc.get("index", 0)
-            if idx >= len(state.tool_calls_list):
-                state.tool_calls_list.append({"id": tc.get("id", ""), "type": "function", "function": {"name": "", "arguments": ""}})
-            func = tc.get("function", {})
-            if func.get("name"):
-                state.tool_calls_list[idx]["function"]["name"] += func["name"]
-            if func.get("arguments"):
-                state.tool_calls_list[idx]["function"]["arguments"] += func["arguments"]
-        
-        return events
-    
-    def save_steps(self, state: StreamState):
-        """Save current iteration steps to all_steps."""
-        if state.thinking_step_id:
-            state.all_steps.append({"id": state.thinking_step_id, "index": state.thinking_step_idx, "type": "thinking", "content": state.full_thinking})
-        if state.text_step_id:
-            state.all_steps.append({"id": state.text_step_id, "index": state.text_step_idx, "type": "text", "content": state.full_content})
-    
-    # ==================== Step 3: Execute Tools ====================
-    
-    def execute_tools(self, state: StreamState, tool_context: Dict) -> Tuple[List[Dict], List[str]]:
-        """
-        Execute tools and return (results, events).
-        """
-        if not state.tool_calls_list:
-            return [], []
-        
-        state.all_tool_calls.extend(state.tool_calls_list)
-        tool_call_ids = []
-        events = []
-        
-        # Yield tool_call steps
-        for tc in state.tool_calls_list:
-            step_id = f"step-{state.step_index}"
-            tool_call_ids.append(step_id)
-            state.step_index += 1
-            step = {
-                "id": step_id, "index": len(state.all_steps), "type": "tool_call",
-                "id_ref": tc.get("id", ""), "name": tc["function"]["name"], "arguments": tc["function"]["arguments"]
-            }
-            state.all_steps.append(step)
-            events.append(_sse_event("process_step", {"step": step}))
-        
-        # Execute tools
-        results = self.tool_executor.process_tool_calls_parallel(state.tool_calls_list, tool_context)
-        
-        # Yield tool_result steps
-        for i, tr in enumerate(results):
-            ref_id = tool_call_ids[i] if i < len(tool_call_ids) else f"tool-{i}"
-            step_id = f"step-{state.step_index}"
-            state.step_index += 1
-            
-            content = tr.get("content", "")
-            success = True
-            try:
-                obj = json.loads(content)
-                if isinstance(obj, dict):
-                    success = obj.get("success", True)
-            except:
-                pass
-            
-            step = {
-                "id": step_id, "index": len(state.all_steps), "type": "tool_result",
-                "id_ref": ref_id, "name": tr.get("name", ""), "content": content, "success": success
-            }
-            state.all_steps.append(step)
-            events.append(_sse_event("process_step", {"step": step}))
-            
-            state.all_tool_results.append({"role": "tool", "tool_call_id": tr.get("tool_call_id", ""), "content": content})
-        
-        return results, events
-    
-    def update_messages_for_next_iteration(self, state: StreamState, results: List[Dict]):
-        """Update messages list with assistant response and tool results for next iteration."""
-        state.messages.append({"role": "assistant", "content": state.full_content or "", "tool_calls": state.tool_calls_list})
-        if results:
-            state.messages.extend(state.all_tool_results[-len(results):])
-        state.all_tool_results = []
-    
-    def reset_iteration_state(self, state: StreamState):
-        """Reset state for next iteration."""
-        state.full_content = state.full_thinking = ""
-        state.tool_calls_list = []
-        state.thinking_step_id = state.thinking_step_idx = state.text_step_id = state.text_step_idx = None
-    
-    # ==================== Step 4: Finalize ====================
-    
-    def save_message(self, conversation_id: str, state: StreamState, token_count: int):
-        """Save assistant message to database."""
-        content_json = {"text": state.full_content, "steps": state.all_steps}
-        if state.all_tool_calls:
-            content_json["tool_calls"] = state.all_tool_calls
-        
-        db = SessionLocal()
-        try:
-            db.add(Message(
-                id=str(uuid.uuid4()),
-                conversation_id=conversation_id,
-                role="assistant",
-                content=json.dumps(content_json, ensure_ascii=False),
-                token_count=token_count,
-                usage=json.dumps(state.total_usage) if state.total_usage else None
-            ))
-            db.commit()
-        except Exception:
-            db.rollback()
-            raise
-        finally:
-            db.close()
-    
-    # ==================== Main Orchestrator ====================
-    
     async def stream_response(
         self,
         conversation: Conversation,
@@ -295,100 +100,360 @@ class ChatService:
         username: str = None,
         workspace: str = None,
         user_permission_level: int = 1
-    ) -> AsyncGenerator[str, None]:
+    ) -> AsyncGenerator[Dict[str, str], None]:
-        """Main streaming orchestrator - step-by-step flow."""
+        """
+        Streaming response generator
+        
+        Yields raw SSE event strings for direct forwarding.
+        """
         try:
-            # Step 1: Initialize
+            messages = self.build_messages(conversation)
-            state, llm, tools, model, max_tokens, tool_context = self.init_stream_state(
+            
-                conversation, user_message, enabled_tools or []
+            messages.append({
-            )
+                "role": "user",
-            tool_context.update
+                "content": json.dumps({"text": user_message, "attachments": []})
-            ({
-                "user_id": user_id, 
-                "username": username, 
-                "workspace": workspace, 
-                "user_permission_level": user_permission_level
             })
             
-            # ReAct loop
+            # Get tools based on enabled_tools filter
+            if enabled_tools:
+                tools = [t for t in registry.list_all() if t.get("function", {}).get("name") in enabled_tools]
+            else:
+                tools = []
+            
+            llm, provider_max_tokens = get_llm_client(conversation)
+            model = conversation.model or llm.default_model or "gpt-4"
+            # 直接使用 provider 的 max_tokens
+            max_tokens = provider_max_tokens
+            
+            # State tracking
+            all_steps = []
+            all_tool_calls = []
+            all_tool_results = []
+            step_index = 0
+            
+            # Token usage tracking
+            total_usage = {
+                "prompt_tokens": 0,
+                "completion_tokens": 0,
+                "total_tokens": 0
+            }
+            
+            # Global step IDs for thinking and text (persist across iterations)
+            thinking_step_id = None
+            thinking_step_idx = None
+            text_step_id = None
+            text_step_idx = None
+            
             for _ in range(MAX_ITERATIONS):
-                self.reset_iteration_state(state)
+                # Stream from LLM
+                full_content = ""
+                full_thinking = ""
+                tool_calls_list = []
                 
-                # Step 2: Stream LLM
+                # Step tracking - use unified step-{index} format
-                async for sse_line, chunk in self.stream_llm_response(
+                thinking_step_id = None
-                    llm, model, state.messages, tools,
+                thinking_step_idx = None
-                    conversation.temperature, max_tokens,
+                text_step_id = None
-                    thinking_enabled or conversation.thinking_enabled
+                text_step_idx = None
+                
+                async for sse_line in llm.stream_call(
+                    model=model,
+                    messages=messages,
+                    tools=tools,
+                    temperature=conversation.temperature,
+                    max_tokens=max_tokens or 8192,
+                    thinking_enabled=thinking_enabled or conversation.thinking_enabled
                 ):
-                    # Handle error events
+                    # Parse SSE line
-                    event_type, data_str = self.parse_sse_line(sse_line)
+                    # Format: "event: xxx\ndata: {...}\n\n"
-                    if event_type == 'error':
+                    event_type = None
-                        error_data = json.loads(data_str) if data_str else {}
+                    data_str = None
-                        yield _sse_event("error", {"content": error_data.get("content", "Unknown error")})
-                        return
                     
-                    if not chunk:
+                    for line in sse_line.strip().split('\n'):
+                        if line.startswith('event: '):
+                            event_type = line[7:].strip()
+                        elif line.startswith('data: '):
+                            data_str = line[6:].strip()
+                    
+                    if data_str is None:
                         continue
                     
-                    # Extract usage
+                    # Handle error events from LLM
-                    if "usage" in chunk:
+                    if event_type == 'error':
-                        u = chunk["usage"]
+                        try:
-                        state.total_usage = {
+                            error_data = json.loads(data_str)
-                            "prompt_tokens": u.get("prompt_tokens", 0),
+                            yield _sse_event("error", {"content": error_data.get("content", "Unknown error")})
-                            "completion_tokens": u.get("completion_tokens", 0),
+                        except json.JSONDecodeError:
-                            "total_tokens": u.get("total_tokens", 0)
+                            yield _sse_event("error", {"content": data_str})
-                        }
+                        return
                     
-                    # Check for API errors
+                    # Parse the data
+                    try:
+                        chunk = json.loads(data_str)
+                    except json.JSONDecodeError:
+                        yield _sse_event("error", {"content": f"Failed to parse response: {data_str}"})
+                        return
+                    
+                    # 提取 API 返回的 usage 信息
+                    if "usage" in chunk:
+                        usage = chunk["usage"]
+                        total_usage["prompt_tokens"] = usage.get("prompt_tokens", 0)
+                        total_usage["completion_tokens"] = usage.get("completion_tokens", 0)
+                        total_usage["total_tokens"] = usage.get("total_tokens", 0)
+                    
+                    # Check for error in response
                     if "error" in chunk:
-                        yield _sse_event("error", {"content": f"API Error: {chunk['error'].get('message', str(chunk['error']))}"})
+                        error_msg = chunk["error"].get("message", str(chunk["error"]))
+                        yield _sse_event("error", {"content": f"API Error: {error_msg}"})
                         return
                     
                     # Get delta
-                    delta = None
                     choices = chunk.get("choices", [])
+                    delta = None
+                    
                     if choices:
                         delta = choices[0].get("delta", {})
+                        # If no delta but has message (non-streaming response)
                         if not delta:
-                            content = choices[0].get("message", {}).get("content")
+                            message = choices[0].get("message", {})
-                            if content:
+                            if message.get("content"):
-                                delta = {"content": content}
+                                delta = {"content": message["content"]}
                     
                     if not delta:
+                        # Check if there's any content in the response (for non-standard LLM responses)
                         content = chunk.get("content") or chunk.get("message", {}).get("content", "")
                         if content:
                             delta = {"content": content}
                     
                     if delta:
-                        # Step 2b: Process delta
+                        # Handle reasoning (thinking)
-                        for event in self.process_delta(state, delta):
+                        reasoning = delta.get("reasoning_content", "")
-                            yield event
+                        if reasoning:
+                            prev_thinking_len = len(full_thinking)
+                            full_thinking += reasoning
+                            if prev_thinking_len == 0:  # New thinking stream started
+                                thinking_step_idx = step_index
+                                thinking_step_id = f"step-{step_index}"
+                                step_index += 1
+                            yield _sse_event("process_step", {
+                                "step": {
+                                    "id": thinking_step_id,
+                                    "index": thinking_step_idx,
+                                    "type": "thinking",
+                                    "content": full_thinking
+                                }
+                            })
+                        
+                        # Handle content
+                        content = delta.get("content", "")
+                        if content:
+                            prev_content_len = len(full_content)
+                            full_content += content
+                            if prev_content_len == 0:  # New text stream started
+                                text_step_idx = step_index
+                                text_step_id = f"step-{step_index}"
+                                step_index += 1
+                            yield _sse_event("process_step", {
+                                "step": {
+                                    "id": text_step_id,
+                                    "index": text_step_idx,
+                                    "type": "text",
+                                    "content": full_content
+                                }
+                            })
+                        
+                        # Accumulate tool calls
+                        tool_calls_delta = delta.get("tool_calls", [])
+                        for tc in tool_calls_delta:
+                            idx = tc.get("index", 0)
+                            if idx >= len(tool_calls_list):
+                                tool_calls_list.append({
+                                    "id": tc.get("id", ""),
+                                    "type": "function",
+                                    "function": {"name": "", "arguments": ""}
+                                })
+                            func = tc.get("function", {})
+                            if func.get("name"):
+                                tool_calls_list[idx]["function"]["name"] += func["name"]
+                            if func.get("arguments"):
+                                tool_calls_list[idx]["function"]["arguments"] += func["arguments"]
                 
-                # Save steps after streaming
+                # Save thinking step
-                self.save_steps(state)
+                if thinking_step_id is not None:
+                    all_steps.append({
+                        "id": thinking_step_id,
+                        "index": thinking_step_idx,
+                        "type": "thinking",
+                        "content": full_thinking
+                    })
                 
-                # Step 3: Execute tools if present
+                # Save text step
-                if state.tool_calls_list:
+                if text_step_id is not None:
-                    results, events = self.execute_tools(state, tool_context)
+                    all_steps.append({
-                    for event in events:
+                        "id": text_step_id,
-                        yield event
+                        "index": text_step_idx,
-                    self.update_messages_for_next_iteration(state, results)
+                        "type": "text",
+                        "content": full_content
+                    })
+                
+                # Handle tool calls
+                if tool_calls_list:
+                    all_tool_calls.extend(tool_calls_list)
+                    
+                    # Yield tool_call steps - use unified step-{index} format
+                    tool_call_step_ids = []  # Track step IDs for tool calls
+                    for tc in tool_calls_list:
+                        call_step_idx = step_index
+                        call_step_id = f"step-{step_index}"
+                        tool_call_step_ids.append(call_step_id)
+                        step_index += 1
+                        call_step = {
+                            "id": call_step_id,
+                            "index": call_step_idx,
+                            "type": "tool_call",
+                            "id_ref": tc.get("id", ""),
+                            "name": tc["function"]["name"],
+                            "arguments": tc["function"]["arguments"]
+                        }
+                        all_steps.append(call_step)
+                        yield _sse_event("process_step", {"step": call_step})
+                    
+                    # Execute tools
+                    tool_context = {
+                        "workspace": workspace,
+                        "user_id": user_id,
+                        "username": username,
+                        "user_permission_level": user_permission_level
+                    }
+                    tool_results = self.tool_executor.process_tool_calls_parallel(
+                        tool_calls_list, tool_context
+                    )
+                    
+                    # Yield tool_result steps - use unified step-{index} format
+                    for i, tr in enumerate(tool_results):
+                        tool_call_step_id = tool_call_step_ids[i] if i < len(tool_call_step_ids) else f"step-{i}"
+                        result_step_idx = step_index
+                        result_step_id = f"step-{step_index}"
+                        step_index += 1
+                        
+                        content = tr.get("content", "")
+                        success = True
+                        try:
+                            content_obj = json.loads(content)
+                            if isinstance(content_obj, dict):
+                                success = content_obj.get("success", True)
+                        except:
+                            pass
+                        
+                        result_step = {
+                            "id": result_step_id,
+                            "index": result_step_idx,
+                            "type": "tool_result",
+                            "id_ref": tool_call_step_id,  # Reference to the tool_call step
+                            "name": tr.get("name", ""),
+                            "content": content,
+                            "success": success
+                        }
+                        all_steps.append(result_step)
+                        yield _sse_event("process_step", {"step": result_step})
+                        
+                        all_tool_results.append({
+                            "role": "tool",
+                            "tool_call_id": tr.get("tool_call_id", ""),
+                            "content": tr.get("content", "")
+                        })
+                    
+                    # Add assistant message with tool calls for next iteration
+                    messages.append({
+                        "role": "assistant",
+                        "content": full_content or "",
+                        "tool_calls": tool_calls_list
+                    })
+                    messages.extend(all_tool_results[-len(tool_results):])
+                    all_tool_results = []
                     continue
                 
-                # Step 4: Finalize (no tool calls)
+                # No tool calls - final iteration, save message
-                token_count = state.total_usage.get("completion_tokens", 0)
+                msg_id = str(uuid.uuid4())
-                self.save_message(conversation.id, state, token_count)
+                
-                yield _sse_event("done", {"message_id": str(uuid.uuid4()), "token_count": token_count, "usage": state.total_usage})
+                actual_token_count = total_usage.get("completion_tokens", 0)
+                logger.info(f"total_usage: {total_usage}")
+                
+                self._save_message(
+                    conversation.id,
+                    msg_id,
+                    full_content,
+                    all_tool_calls,
+                    all_tool_results,
+                    all_steps,
+                    actual_token_count,
+                    total_usage
+                )
+                
+                yield _sse_event("done", {
+                    "message_id": msg_id,
+                    "token_count": actual_token_count,
+                    "usage": total_usage
+                })
                 return
             
-            # Max iterations exceeded
+            # Max iterations exceeded - save message before error
-            if state.full_content or state.all_tool_calls:
+            if full_content or all_tool_calls:
-                self.save_message(conversation.id, state, state.total_usage.get("completion_tokens", 0))
+                msg_id = str(uuid.uuid4())
+                self._save_message(
+                    conversation.id,
+                    msg_id,
+                    full_content,
+                    all_tool_calls,
+                    all_tool_results,
+                    all_steps,
+                    actual_token_count,
+                    total_usage
+                )
             yield _sse_event("error", {"content": "Exceeded maximum tool call iterations"})
             
         except Exception as e:
             yield _sse_event("error", {"content": str(e)})
+    
+    def _save_message(
+        self,
+        conversation_id: str,
+        msg_id: str,
+        full_content: str,
+        all_tool_calls: list,
+        all_tool_results: list,
+        all_steps: list,
+        token_count: int = 0,
+        usage: dict = None
+    ):
+        """Save the assistant message to database."""
+
+        
+        content_json = {
+            "text": full_content,
+            "steps": all_steps
+        }
+        if all_tool_calls:
+            content_json["tool_calls"] = all_tool_calls
+        
+        db = SessionLocal()
+        try:
+            msg = Message(
+                id=msg_id,
+                conversation_id=conversation_id,
+                role="assistant",
+                content=json.dumps(content_json, ensure_ascii=False),
+                token_count=token_count,
+                usage=json.dumps(usage) if usage else None
+            )
+            db.add(msg)
+            db.commit()
+        except Exception as e:
+            db.rollback()
+            raise
+        finally:
+            db.close()
 
 
-# Global service
+# Global chat service
 chat_service = ChatService()