diff --git a/backend/agent_factory.py b/backend/agent_factory.py
index 8c79f02..97a693d 100644
--- a/backend/agent_factory.py
+++ b/backend/agent_factory.py
@@ -15,7 +15,7 @@
 
 from agents._config import set_default_openai_api
 from agents.tracing import set_tracing_disabled
-# from .utils.logger import log
+from .utils.logger import log
 
 # def load_env_config():
 #     """Load environment variables from .env.llm file"""
@@ -37,6 +37,8 @@
 def create_agent(**kwargs) -> Agent:
     # 通过用户配置拿/环境变量
     config = kwargs.pop("config") if "config" in kwargs else {}
+    log.info(f"Agent Factory - Received config: {config}")
+    
     # 避免将 None 写入 headers
     session_id = (config or {}).get("session_id")
     default_headers = {}
@@ -53,11 +55,17 @@ def create_agent(**kwargs) -> Agent:
         if config.get("openai_api_key") and config.get("openai_api_key") != "":
             api_key = config.get("openai_api_key")
     
+    log.info(f"Agent Factory - Using base_url: {base_url}")
+    log.info(f"Agent Factory - Using api_key: {'***' if api_key else 'None'}")
+    
     # Check if this is LMStudio and adjust API key handling
     is_lmstudio = is_lmstudio_url(base_url)
+    log.info(f"Agent Factory - Is LMStudio: {is_lmstudio}")
+    
     if is_lmstudio and not api_key:
         # LMStudio typically doesn't require an API key, use a placeholder
         api_key = "lmstudio-local"
+        log.info("Agent Factory - Set LMStudio placeholder API key")
 
     client = AsyncOpenAI(
         api_key=api_key,
@@ -67,6 +75,7 @@ def create_agent(**kwargs) -> Agent:
 
     default_model_name = os.environ.get("OPENAI_MODEL", "gemini-2.5-flash")
     model_name = kwargs.pop("model") or default_model_name
+    log.info(f"Agent Factory - Using model: {model_name}")
     model = OpenAIChatCompletionsModel(model_name, openai_client=client)
 
     if config.get("max_tokens"):
diff --git a/backend/controller/conversation_api.py b/backend/controller/conversation_api.py
index 887d280..a463437 100644
--- a/backend/controller/conversation_api.py
+++ b/backend/controller/conversation_api.py
@@ -132,11 +132,35 @@ async def invoke_chat(request):
     set_language(language)
     
     # 构建配置信息
+    openai_api_key = request.headers.get('Openai-Api-Key')
+    openai_base_url = request.headers.get('Openai-Base-Url')
+
+    # If base URL header missing, fall back to last successful one (e.g., LMStudio) or default LMStudio
+    if not openai_base_url:
+        from ..utils.globals import get_last_openai_base_url, LMSTUDIO_DEFAULT_BASE_URL
+        remembered = get_last_openai_base_url()
+        if remembered:
+            openai_base_url = remembered
+            log.info(f"Using remembered base URL for chat: {openai_base_url}")
+        else:
+            openai_base_url = LMSTUDIO_DEFAULT_BASE_URL
+            log.info(f"No base URL provided; defaulting to LMStudio: {openai_base_url}")
+
+    # If neither API key nor base URL is provided, assume local LMStudio by default
+    # This enables out-of-the-box local usage without requiring a cloud key
+    if not openai_api_key and not openai_base_url:
+        from ..utils.globals import LMSTUDIO_DEFAULT_BASE_URL
+        openai_base_url = LMSTUDIO_DEFAULT_BASE_URL
+        log.info(f"No OpenAI credentials provided; defaulting base URL to LMStudio: {openai_base_url}")
+    
+    log.info(f"Request headers - Openai-Api-Key: {'***' if openai_api_key else 'None'}")
+    log.info(f"Request headers - Openai-Base-Url: {openai_base_url}")
+    
     config = {
         "session_id": session_id,
         "workflow_checkpoint_id": workflow_checkpoint_id,
-        "openai_api_key": request.headers.get('Openai-Api-Key'),
-        "openai_base_url": request.headers.get('Openai-Base-Url'),
+        "openai_api_key": openai_api_key,
+        "openai_base_url": openai_base_url,
         "model_select": next((x['data'][0] for x in ext if x['type'] == 'model_select' and x.get('data')), None)
     }
     
@@ -165,6 +189,25 @@ async def invoke_chat(request):
     # 不再需要创建用户消息存储到后端，前端负责消息存储
 
     try:
+        # Validate API key presence unless using LMStudio/local base URL
+        from ..utils.globals import is_lmstudio_url
+        if not openai_api_key and not (openai_base_url and is_lmstudio_url(openai_base_url)):
+            warning_msg = (
+                "No OpenAI API key provided. Please click the gear icon (⚙️) to configure your key, "
+                "or set a local LMStudio base URL in settings."
+            )
+            log.error(warning_msg)
+            chat_response = ChatResponse(
+                session_id=session_id,
+                text=warning_msg,
+                finished=True,
+                type="message",
+                format="text",
+                ext=None
+            )
+            await response.write(json.dumps(chat_response).encode() + b"\n")
+            await response.write_eof()
+            return response
         # Call the MCP client to get streaming response with historical messages and image support
         # Pass OpenAI-formatted messages and processed images to comfyui_agent_invoke
         accumulated_text = ""
diff --git a/backend/controller/llm_api.py b/backend/controller/llm_api.py
index 4887cc8..25c1a94 100644
--- a/backend/controller/llm_api.py
+++ b/backend/controller/llm_api.py
@@ -12,7 +12,7 @@
 import json
 from typing import List, Dict, Any
 from aiohttp import web
-from ..utils.globals import LLM_DEFAULT_BASE_URL, LMSTUDIO_DEFAULT_BASE_URL, is_lmstudio_url
+from ..utils.globals import LLM_DEFAULT_BASE_URL, LMSTUDIO_DEFAULT_BASE_URL, is_lmstudio_url, set_last_openai_base_url
 import server
 import requests
 from ..utils.logger import log
@@ -57,6 +57,11 @@ async def list_models(request):
                     "name": model['id'],
                     "image_enable": True
                 })
+            # Remember a working base URL for later chat calls
+            set_last_openai_base_url(openai_base_url)
+        else:
+            # Surface error info to logs to help users diagnose (e.g., 401)
+            log.error(f"Model list request failed: HTTP {response.status_code} - {response.text}")
         
         return web.json_response({
                 "models": llm_config
diff --git a/backend/service/debug_agent.py b/backend/service/debug_agent.py
index 9ac5360..8f4dcbd 100644
--- a/backend/service/debug_agent.py
+++ b/backend/service/debug_agent.py
@@ -209,6 +209,10 @@ async def debug_workflow_errors(workflow_data: Dict[str, Any]):
         if not session_id:
             session_id = str(uuid.uuid4())  # Fallback if no context
         
+        # Use user's selected model if available, otherwise fall back to WORKFLOW_MODEL_NAME
+        selected_model = config.get('model_select') if config else None
+        model_to_use = selected_model if selected_model else WORKFLOW_MODEL_NAME
+        
         # 1. 保存工作流数据到数据库
         log.info(f"Saving workflow data for session {session_id}")
         save_result = save_workflow_data(
@@ -252,16 +256,20 @@ async def debug_workflow_errors(workflow_data: Dict[str, Any]):
 **Note**: The workflow validation is done using ComfyUI's internal functions, not actual execution, so it's fast and safe.
 
 Start by validating the workflow to see its current state.""",
-            model=WORKFLOW_MODEL_NAME,
+            model=model_to_use,
             tools=[run_workflow, analyze_error_type, save_current_workflow],
             config={
                 "max_tokens": 8192
             }
         )
         
+        # Use user's selected model if available, otherwise fall back to WORKFLOW_MODEL_NAME
+        selected_model = config.get('model_select') if config else None
+        model_to_use = selected_model if selected_model else WORKFLOW_MODEL_NAME
+        
         workflow_bugfix_default_agent = create_agent(
             name="Workflow Bugfix Default Agent",
-            model=WORKFLOW_MODEL_NAME,
+            model=model_to_use,
             handoff_description="""
             I am the Workflow Bugfix Default Agent. I specialize in fixing structural issues in ComfyUI workflows.
             
@@ -310,7 +318,7 @@ async def debug_workflow_errors(workflow_data: Dict[str, Any]):
         
         link_agent = create_agent(
             name="Link Agent",
-            model=WORKFLOW_MODEL_NAME,
+            model=model_to_use,
             handoff_description="""
             I am the Link Agent. I specialize in analyzing and fixing workflow connection issues.
             
@@ -401,7 +409,7 @@ async def debug_workflow_errors(workflow_data: Dict[str, Any]):
 
         parameter_agent = create_agent(
             name="Parameter Agent",
-            model=WORKFLOW_MODEL_NAME,
+            model=model_to_use,
             handoff_description="""
             I am the Parameter Agent. I specialize in handling parameter-related errors in ComfyUI workflows.
             
diff --git a/backend/service/mcp_client.py b/backend/service/mcp_client.py
index 7973b22..7ae5c27 100644
--- a/backend/service/mcp_client.py
+++ b/backend/service/mcp_client.py
@@ -169,6 +169,7 @@ async def on_handoff(ctx: RunContextWrapper[None], input_data: HandoffRewriteDat
             # Enhanced retry mechanism for OpenAI streaming errors
             max_retries = 3
             retry_count = 0
+            last_non_retryable_error_message: Optional[str] = None
             
             async def process_stream_events(stream_result):
                 """Process stream events with enhanced error handling"""
@@ -335,6 +336,8 @@ async def process_stream_events(stream_result):
                     else:
                         log.error(f"Non-retryable streaming error or max retries reached: {error_msg}")
                         log.error(f"Traceback: {traceback.format_exc()}")
+                        # Capture the error to surface it to the caller if no content was streamed
+                        last_non_retryable_error_message = error_msg
                         if isinstance(stream_error, RateLimitError):
                             default_error_msg = 'Rate limit exceeded, please try again later.'
                             error_body = stream_error.body
@@ -359,6 +362,10 @@ async def process_stream_events(stream_result):
                         await asyncio.sleep(1)
                         continue
 
+            # If we encountered a non-retryable error and produced no text, surface the error message
+            if last_non_retryable_error_message and not current_text:
+                current_text = last_non_retryable_error_message
+
             # Add detailed debugging info about tool results
             log.info(f"Total tool results: {len(tool_results)}")
             for tool_name, result in tool_results.items():
diff --git a/backend/service/workflow_rewrite_agent.py b/backend/service/workflow_rewrite_agent.py
index a49301c..7f00ffa 100644
--- a/backend/service/workflow_rewrite_agent.py
+++ b/backend/service/workflow_rewrite_agent.py
@@ -19,7 +19,7 @@
 
 from ..agent_factory import create_agent
 from ..utils.globals import WORKFLOW_MODEL_NAME, get_language
-from ..utils.request_context import get_session_id
+from ..utils.request_context import get_session_id, get_config
 
 from ..service.workflow_rewrite_tools import *
 
@@ -43,10 +43,19 @@ def create_workflow_rewrite_agent():
     
     language = get_language()
     session_id = get_session_id() or "unknown_session"
+    config = get_config() or {}
+    
+    # Use user's selected model if available, otherwise fall back to WORKFLOW_MODEL_NAME
+    selected_model = config.get('model_select') if config else None
+    model_to_use = selected_model if selected_model else WORKFLOW_MODEL_NAME
+    
+    # Merge max_tokens into config
+    agent_config = {**config, "max_tokens": 8192} if config else {"max_tokens": 8192}
     
     return create_agent(
         name="Workflow Rewrite Agent",
-        model=WORKFLOW_MODEL_NAME,
+        model=model_to_use,
+        config=agent_config,
         handoff_description="""
         我是工作流改写代理，专门负责根据用户需求修改和优化当前画布上的ComfyUI工作流。
         """,
@@ -104,10 +113,7 @@ def create_workflow_rewrite_agent():
 
         始终以用户的实际需求为导向，提供专业、准确、高效的工作流改写服务。
         """,
-        tools=[get_rewrite_expert_by_name, get_current_workflow, get_node_info, update_workflow, remove_node],
-        config={
-            "max_tokens": 8192
-        }
+        tools=[get_rewrite_expert_by_name, get_current_workflow, get_node_info, update_workflow, remove_node]
     )
 
 # 注意：工作流改写代理现在需要在有session context的环境中创建
diff --git a/backend/service/workflow_rewrite_agent_simple.py b/backend/service/workflow_rewrite_agent_simple.py
index 54462fe..5635ae2 100644
--- a/backend/service/workflow_rewrite_agent_simple.py
+++ b/backend/service/workflow_rewrite_agent_simple.py
@@ -88,9 +88,14 @@ def rewrite_workflow_simple(rewrite_context: RewriteContext) -> str:
             api_key = get_comfyui_copilot_api_key() or ""
         )
 
+        # Use user's selected model if available, otherwise fall back to WORKFLOW_MODEL_NAME
+        config = get_config()
+        selected_model = config.get('model_select') if config else None
+        model_to_use = selected_model if selected_model else WORKFLOW_MODEL_NAME
+
         # 调用LLM
         completion = client.chat.completions.parse(
-            model=WORKFLOW_MODEL_NAME,
+            model=model_to_use,
             messages=[
                 {"role": "system", "content": system_prompt},
                 {"role": "user", "content": context_info}
diff --git a/backend/utils/globals.py b/backend/utils/globals.py
index 4d3b231..4a5ca4b 100644
--- a/backend/utils/globals.py
+++ b/backend/utils/globals.py
@@ -21,6 +21,7 @@ def __init__(self):
         self._lock = threading.RLock()
         self._state: Dict[str, Any] = {
             'LANGUAGE': 'en',  # Default language
+            'LAST_OPENAI_BASE_URL': None,
         }
     
     def get(self, key: str, default: Any = None) -> Any:
@@ -91,6 +92,16 @@ def set_comfyui_copilot_api_key(api_key: str) -> None:
     _global_state.set('comfyui_copilot_api_key', api_key)
 
 
+def set_last_openai_base_url(base_url: Optional[str]) -> None:
+    """Remember the last used OpenAI base URL (e.g., LMStudio), for fallback in chat."""
+    _global_state.set('LAST_OPENAI_BASE_URL', base_url)
+
+
+def get_last_openai_base_url() -> Optional[str]:
+    """Get the last remembered OpenAI base URL."""
+    return _global_state.get('LAST_OPENAI_BASE_URL')
+
+
 BACKEND_BASE_URL = "https://comfyui-copilot-server.onrender.com"
 LMSTUDIO_DEFAULT_BASE_URL = "http://localhost:1234/v1"
 WORKFLOW_MODEL_NAME = "us.anthropic.claude-sonnet-4-20250514-v1:0"
diff --git a/backend/utils/logger.py b/backend/utils/logger.py
index 23cee18..de890a6 100644
--- a/backend/utils/logger.py
+++ b/backend/utils/logger.py
@@ -1,5 +1,6 @@
 """
-Logging utility module using Python standard library logging with file location info.
+Logging utility module using Python standard library logging with file location info
+and robust Unicode handling for Windows consoles and log files.
 """
 
 import logging
@@ -9,6 +10,46 @@
 import inspect
 from datetime import datetime
 
+# Set UTF-8 encoding for Windows console output
+if os.name == 'nt':  # Windows
+    try:
+        # Set environment variable for UTF-8 encoding
+        os.environ['PYTHONIOENCODING'] = 'utf-8'
+        # Try to reconfigure stdout and stderr
+        if hasattr(sys.stdout, 'reconfigure'):
+            sys.stdout.reconfigure(encoding='utf-8', errors='replace')
+        if hasattr(sys.stderr, 'reconfigure'):
+            sys.stderr.reconfigure(encoding='utf-8', errors='replace')
+    except Exception:
+        pass  # Continue with default encoding if reconfiguration fails
+
+
+class SafeStreamHandler(logging.StreamHandler):
+    """Stream handler that safely handles Unicode encoding errors."""
+    
+    def emit(self, record):
+        try:
+            super().emit(record)
+        except UnicodeEncodeError:
+            # If encoding fails, try to emit with safe encoding
+            try:
+                # Temporarily modify the record to use safe encoding
+                original_msg = record.msg
+                record.msg = str(original_msg).encode('utf-8', errors='replace').decode('utf-8')
+                super().emit(record)
+                # Restore original message
+                record.msg = original_msg
+            except Exception:
+                # If that also fails, use a very basic approach
+                try:
+                    msg = f"[ENCODING ERROR] {record.levelname}: {str(record.msg)[:100]}..."
+                    if getattr(self, 'stream', None):
+                        self.stream.write(msg + self.terminator)
+                    self.flush()
+                except Exception:
+                    # Last resort: silently ignore
+                    pass
+
 
 class LocationFormatter(logging.Formatter):
     """Custom formatter that adds file location information."""
@@ -22,7 +63,20 @@ def format(self, record):
             line_number = record.lineno if record.lineno else 0
             record.location = f"{filename}:{function_name}:{line_number}"
         
-        return super().format(record)
+        try:
+            return super().format(record)
+        except UnicodeEncodeError:
+            # Handle encoding errors by safely encoding problematic characters
+            original_msg = str(record.msg)
+            record.msg = original_msg.encode('utf-8', errors='replace').decode('utf-8')
+            # Also handle any other string attributes that might have encoding issues
+            if hasattr(record, 'location'):
+                record.location = str(record.location).encode('utf-8', errors='replace').decode('utf-8')
+            return super().format(record)
+        except Exception:
+            # Fallback for any other formatting errors
+            record.msg = f"[Encoding Error] {str(record.msg)[:200]}..."
+            return super().format(record)
 
 
 def setup_logger():
@@ -35,10 +89,17 @@ def setup_logger():
     if logger.handlers:
         return logger
     
-    # Console handler with color support
-    console_handler = logging.StreamHandler(sys.stderr)
+    # Console handler with UTF-8 encoding and safe Unicode handling
+    console_handler = SafeStreamHandler(sys.stderr)
     console_handler.setLevel(logging.DEBUG)
     
+    # Set UTF-8 encoding for Windows compatibility
+    if hasattr(console_handler.stream, 'reconfigure'):
+        try:
+            console_handler.stream.reconfigure(encoding='utf-8', errors='replace')
+        except Exception:
+            pass  # Fallback to default encoding if reconfigure fails
+    
     # Console formatter with colors (simple format for better compatibility)
     console_format = '%(asctime)s | %(levelname)-8s | %(location)s | %(message)s'
     console_formatter = LocationFormatter(console_format, datefmt='%Y-%m-%d %H:%M:%S')
@@ -51,7 +112,8 @@ def setup_logger():
     file_handler = logging.handlers.RotatingFileHandler(
         os.path.join(log_dir, "comfyui_copilot.log"),
         maxBytes=10*1024*1024,  # 10MB
-        backupCount=7
+        backupCount=7,
+        encoding='utf-8'
     )
     file_handler.setLevel(logging.DEBUG)
     
@@ -93,11 +155,14 @@ def _log_with_location(self, level, message, *args, **kwargs):
             function_name = frame.f_code.co_name
             line_number = frame.f_lineno
             
+            # Safely encode the message to prevent Unicode issues
+            safe_message = safe_encode_message(message)
+            
             # Create a log record manually to ensure no duplicate processing
             if self._logger.isEnabledFor(level):
                 record = self._logger.makeRecord(
                     self._logger.name, level, frame.f_code.co_filename, line_number,
-                    message, args, None, function_name
+                    safe_message, args, None, function_name
                 )
                 record.location = f"{filename}:{function_name}:{line_number}"
                 
@@ -149,7 +214,17 @@ def exception(self, message, *args, **kwargs):
             del frame
 
 
-# Create default logger instance
+def safe_encode_message(message):
+    """Safely encode a message to handle Unicode characters."""
+    if isinstance(message, str):
+        try:
+            # Try to encode and decode to ensure it's safe
+            return message.encode('utf-8', errors='replace').decode('utf-8')
+        except Exception:
+            return str(message)[:500] + "..." if len(str(message)) > 500 else str(message)
+    return str(message)
+
+# Create a default logger instance for convenience
 log = Logger()
 
 # For backward compatibility and convenience
@@ -168,6 +243,6 @@ def get_logger(name=None):
 
 
 __all__ = [
-    'log', 'Logger', 'get_logger',
+    'log', 'Logger', 'get_logger', 'safe_encode_message',
     'debug', 'info', 'warning', 'warn', 'error', 'critical', 'exception'
 ]
\ No newline at end of file
diff --git a/ui/src/apis/workflowChatApi.ts b/ui/src/apis/workflowChatApi.ts
index e584de9..e39f261 100644
--- a/ui/src/apis/workflowChatApi.ts
+++ b/ui/src/apis/workflowChatApi.ts
@@ -240,11 +240,16 @@ export namespace WorkflowChatAPI {
       };
       
       // Add OpenAI configuration headers if available
-      if (openaiApiKey && openaiApiKey.trim() !== '' && rsaPublicKey) {
+      if ((openaiApiKey && openaiApiKey.trim() !== '' && rsaPublicKey) || openaiBaseUrl) {
         try {
           headers['Openai-Base-Url'] = openaiBaseUrl;
+          if (openaiApiKey && openaiApiKey.trim() !== '') {
+            headers['Openai-Api-Key'] = openaiApiKey;
+          } else {
+            headers['Openai-Api-Key'] = '';
+          }
         } catch (error) {
-          console.error('Error encrypting OpenAI API key:', error);
+          console.error('Error setting OpenAI headers:', error);
         }
       }
       
@@ -264,7 +269,10 @@ export namespace WorkflowChatAPI {
       if(intent && intent !== '') {
         chatUrl = `${BASE_URL}/api/chat/invoke`
       } else {
-        headers['Openai-Api-Key'] = openaiApiKey;
+        // Only set API key header if not already set above
+        if (!headers['Openai-Api-Key']) {
+          headers['Openai-Api-Key'] = openaiApiKey || '';
+        }
       }
       const response = await fetch(chatUrl, {
         method: 'POST',
@@ -439,7 +447,7 @@ export namespace WorkflowChatAPI {
         console.log('Using cached messages from localStorage');
         return JSON.parse(cachedMessages) as Message[];
       }
-      return [];
+      // No cache available; fall back to server request
       const apiKey = getApiKey();
       const browserLanguage = getBrowserLanguage();
       const { openaiApiKey, openaiBaseUrl, rsaPublicKey } = getOpenAiConfig();