diff --git a/dashboard/src/api/index.js b/dashboard/src/api/index.js
index 69241cd..15f3a2e 100644
--- a/dashboard/src/api/index.js
+++ b/dashboard/src/api/index.js
@@ -283,10 +283,22 @@ export function createRoomWS(roomId, callbacks = {}) {
       }
     },
     sendMessage: (content, userId = 'user', userName = 'User') => {
-      ws.send(JSON.stringify({ action: 'send_message', content, user_id: userId, user_name: userName }))
+      const send = () => {
+        if (ws.readyState === WebSocket.OPEN) {
+          ws.send(JSON.stringify({ action: 'send_message', content, user_id: userId, user_name: userName }))
+        } else if (ws.readyState === WebSocket.CONNECTING) {
+          // Wait for connection then retry
+          ws.addEventListener('open', () => {
+            ws.send(JSON.stringify({ action: 'send_message', content, user_id: userId, user_name: userName }))
+          }, { once: true })
+        }
+      }
+      send()
     },
     ping: () => {
-      ws.send(JSON.stringify({ action: 'ping' }))
+      if (ws.readyState === WebSocket.OPEN) {
+        ws.send(JSON.stringify({ action: 'ping' }))
+      }
     },
     close: () => ws.close()
   }
diff --git a/dashboard/src/views/AgentsView.vue b/dashboard/src/views/AgentsView.vue
index 8e588ed..7279c37 100644
--- a/dashboard/src/views/AgentsView.vue
+++ b/dashboard/src/views/AgentsView.vue
@@ -33,6 +33,14 @@
           </div>
         </div>
 
+        <div class="agent-config">
+          <span v-if="getProviderName(agent.provider_id)" class="config-tag provider-tag">
+            Provider: {{ getProviderName(agent.provider_id) }}
+          </span>
+          <span v-if="agent.model" class="config-tag">模型: {{ agent.model }}</span>
+          <span v-if="agent.tools?.length" class="config-tag">工具: {{ agent.tools.length }}个</span>
+        </div>
+
         <p class="agent-prompt">{{ agent.system_prompt?.slice(0, 100) }}...</p>
 
         <div class="agent-actions">
@@ -64,6 +72,47 @@
               placeholder="定义 Agent 的行为和职责..."></textarea>
           </div>
 
+          <div class="form-row">
+            <div class="form-group">
+              <label>LLM Provider</label>
+              <select v-model="form.provider_id">
+                <option :value="null">默认配置</option>
+                <option v-for="p in providers" :key="p.id" :value="p.id">
+                  {{ p.name }} ({{ p.provider_type }})
+                </option>
+              </select>
+            </div>
+            <div class="form-group">
+              <label>模型</label>
+              <input v-model="form.model" type="text" :placeholder="selectedProvider?.default_model || '如: deepseek-chat'" />
+            </div>
+          </div>
+
+          <!-- Provider 详情 -->
+          <div v-if="selectedProvider" class="provider-details">
+            <div class="provider-info">
+              <span class="provider-type">{{ selectedProvider.provider_type }}</span>
+              <span v-if="selectedProvider.default_model" class="provider-model">
+                默认模型: {{ selectedProvider.default_model }}
+              </span>
+              <span v-if="selectedProvider.base_url" class="provider-url">
+                API: {{ selectedProvider.base_url }}
+              </span>
+            </div>
+          </div>
+
+          <div class="form-group">
+            <label>工具</label>
+            <div class="tools-grid">
+              <label v-for="tool in tools" :key="tool.function.name"
+                class="tool-checkbox" :class="{ active: form.tools.includes(tool.function.name) }">
+                <input type="checkbox" :checked="form.tools.includes(tool.function.name)"
+                  @change="toggleTool(tool.function.name)" />
+                {{ tool.function.name }}
+              </label>
+            </div>
+          </div>
+
           <div class="form-row">
             <div class="form-group">
               <label>优先级</label>
@@ -103,10 +152,12 @@
 </template>
 
 <script setup>
-import { ref, reactive, onMounted } from 'vue'
-import { agentsAPI } from '@/api'
+import { ref, reactive, computed, onMounted } from 'vue'
+import { agentsAPI, providersAPI, toolsAPI } from '@/api'
 
 const agents = ref([])
+const providers = ref([])
+const tools = ref([])
 const showCreateModal = ref(false)
 const editingAgent = ref(null)
 
@@ -114,6 +165,9 @@ const form = reactive({
   name: '',
   role: 'helper',
   system_prompt: '',
+  provider_id: null,
+  model: '',
+  tools: [],
   priority: 5,
   temperature: 0.7,
   max_tokens: 2048,
@@ -121,21 +175,58 @@ const form = reactive({
   mention_trigger: false
 })
 
+const selectedProvider = computed(() => {
+  if (!form.provider_id) return null
+  return providers.value.find(p => p.id === form.provider_id) || null
+})
+
+function getProviderName(providerId) {
+  if (!providerId) return null
+  const p = providers.value.find(p => p.id === providerId)
+  return p ? p.name : null
+}
+
 async function loadAgents() {
   try {
     const res = await agentsAPI.list()
-    agents.value = res.agents || []
+    // Support both {agents: []} and {success: true, data: {agents: []}}
+    agents.value = res.data?.agents || res.agents || []
   } catch (e) {
     console.error('Failed to load agents:', e)
   }
 }
 
+async function loadProviders() {
+  try {
+    const res = await providersAPI.list()
+    // Support both {providers: []} and {success: true, data: {providers: []}}
+    providers.value = res.data?.providers || res.providers || []
+  } catch (e) {
+    console.error('Failed to load providers:', e)
+    providers.value = []
+  }
+}
+
+async function loadTools() {
+  try {
+    const res = await toolsAPI.list()
+    // Support both {tools: []} and {success: true, data: {tools: []}}
+    tools.value = res.data?.tools || res.tools || []
+  } catch (e) {
+    console.error('Failed to load tools:', e)
+    tools.value = []
+  }
+}
+
 function editAgent(agent) {
   editingAgent.value = agent
   Object.assign(form, {
     name: agent.name,
     role: agent.role,
     system_prompt: agent.system_prompt,
+    provider_id: agent.provider_id || null,
+    model: agent.model || '',
+    tools: agent.tools || [],
     priority: agent.priority,
     temperature: agent.temperature,
     max_tokens: agent.max_tokens,
@@ -149,6 +240,7 @@ function closeModal() {
   editingAgent.value = null
   Object.assign(form, {
     name: '', role: 'helper', system_prompt: '',
+    provider_id: null, model: '', tools: [],
     priority: 5, temperature: 0.7, max_tokens: 2048,
     auto_response: true, mention_trigger: false
   })
@@ -156,10 +248,23 @@ function closeModal() {
 
 async function saveAgent() {
   try {
+    const data = { ...form }
+
+    // Handle provider_id: send clear_provider if selecting "default"
+    if (form.provider_id === null) {
+      data.clear_provider = true
+      delete data.provider_id
+    }
+
+    if (!data.model) delete data.model
+    if (data.tools.length === 0) delete data.tools
+
+    console.log('Saving agent with data:', data)
+
     if (editingAgent.value) {
-      await agentsAPI.update(editingAgent.value.id, { ...form })
+      await agentsAPI.update(editingAgent.value.id, data)
     } else {
-      await agentsAPI.create({ ...form })
+      await agentsAPI.create(data)
     }
     closeModal()
     loadAgents()
@@ -179,7 +284,20 @@ async function deleteAgent(id) {
   }
 }
 
-onMounted(loadAgents)
+function toggleTool(toolName) {
+  const idx = form.tools.indexOf(toolName)
+  if (idx >= 0) {
+    form.tools.splice(idx, 1)
+  } else {
+    form.tools.push(toolName)
+  }
+}
+
+onMounted(() => {
+  loadAgents()
+  loadProviders()
+  loadTools()
+})
 </script>
 
 <style scoped>
@@ -266,10 +384,63 @@ onMounted(loadAgents)
 .agent-prompt {
   font-size: 13px;
   color: #666;
-  margin-bottom: 16px;
+  margin-bottom: 12px;
   line-height: 1.5;
 }
 
+.agent-config {
+  display: flex;
+  gap: 8px;
+  margin-bottom: 12px;
+  flex-wrap: wrap;
+}
+
+.config-tag {
+  font-size: 11px;
+  padding: 2px 8px;
+  background: #f0f0f0;
+  border-radius: 10px;
+  color: #666;
+}
+
+.config-tag.provider-tag {
+  background: #667eea20;
+  color: #667eea;
+}
+
+.provider-details {
+  background: #f8f9fa;
+  border: 1px solid #e5e7eb;
+  border-radius: 8px;
+  padding: 12px;
+  margin-bottom: 16px;
+}
+
+.provider-info {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 12px;
+  font-size: 13px;
+}
+
+.provider-type {
+  font-weight: 600;
+  color: #667eea;
+  background: #667eea15;
+  padding: 2px 8px;
+  border-radius: 4px;
+}
+
+.provider-model {
+  color: #666;
+}
+
+.provider-url {
+  color: #888;
+  font-size: 12px;
+  word-break: break-all;
+}
+
 .agent-actions {
   display: flex;
   gap: 8px;
@@ -347,7 +518,8 @@ onMounted(loadAgents)
 }
 
 .form-group input,
-.form-group textarea {
+.form-group textarea,
+.form-group select {
   width: 100%;
   padding: 10px;
   border: 1px solid var(--border-color);
@@ -356,6 +528,34 @@ onMounted(loadAgents)
   box-sizing: border-box;
 }
 
+.tools-grid {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+}
+
+.tool-checkbox {
+  display: flex;
+  align-items: center;
+  gap: 6px;
+  padding: 6px 12px;
+  border: 1px solid var(--border-color);
+  border-radius: 16px;
+  font-size: 13px;
+  cursor: pointer;
+  transition: all 0.2s;
+}
+
+.tool-checkbox input {
+  display: none;
+}
+
+.tool-checkbox.active {
+  background: #667eea;
+  color: white;
+  border-color: #667eea;
+}
+
 .form-group textarea {
   resize: vertical;
   font-family: inherit;
diff --git a/dashboard/vite.config.js b/dashboard/vite.config.js
index 9694a30..5e1237e 100644
--- a/dashboard/vite.config.js
+++ b/dashboard/vite.config.js
@@ -15,8 +15,13 @@ export default defineConfig({
       '/api': {
         target: 'http://localhost:8000',
         changeOrigin: true,
-        secure: false,
-        rewrite: (path) => path
+        secure: false
+      },
+      '/ws': {
+        target: 'ws://localhost:8000',
+        ws: true,
+        changeOrigin: true,
+        secure: false
       }
     }
   },
diff --git a/luxx/agents/base.py b/luxx/agents/base.py
index 37e38e7..3e3093d 100644
--- a/luxx/agents/base.py
+++ b/luxx/agents/base.py
@@ -1,11 +1,11 @@
 """Base Agent class"""
 import json
-import uuid
 import logging
-from typing import List, Dict, Any, Optional, AsyncGenerator
-from abc import ABC, abstractmethod
+from typing import List, Dict, Any, AsyncGenerator
+from abc import ABC
 
-from luxx.services.llm_client import LLMClient
+from luxx.tools.core import registry
+from luxx.services.chat import chat_service
 
 logger = logging.getLogger(__name__)
 
@@ -42,32 +42,6 @@ class BaseAgent(ABC):
         self.auto_response = auto_response
         self.mention_trigger = mention_trigger
         self.avatar = avatar
-        self.llm_client = None
-
-    def _get_llm_client(self, room_id: str = None):
-        """Get LLM client, optionally using agent's provider"""
-        if self.llm_client:
-            return self.llm_client
-
-        if self.provider_id:
-            from luxx.core.database import SessionLocal
-            from luxx.models import LLMProvider
-            db = SessionLocal()
-            try:
-                provider = db.query(LLMProvider).filter(LLMProvider.id == self.provider_id).first()
-                if provider:
-                    self.llm_client = LLMClient(
-                        api_key=provider.api_key,
-                        api_url=provider.base_url,
-                        model=provider.default_model
-                    )
-                    return self.llm_client
-            finally:
-                db.close()
-
-        # Fallback to global config
-        self.llm_client = LLMClient()
-        return self.llm_client
 
     async def stream_response(
         self,
@@ -78,6 +52,7 @@ class BaseAgent(ABC):
     ) -> AsyncGenerator[Dict[str, Any], None]:
         """
         Generate streaming response for the agent.
+        Reuses ChatService's core logic for consistency.
 
         Args:
             user_message: The user's message
@@ -88,9 +63,18 @@ class BaseAgent(ABC):
         Yields:
             SSE-formatted event dictionaries
         """
-        messages = []
+        logger.info(f"[Agent {self.name}] Starting stream_response, provider_id={self.provider_id}, model={self.model}")
 
-        # Add system prompt
+        # Get tools if enabled
+        enabled_tools = []
+        if self.tools:
+            for tool_name in self.tools:
+                tool = registry.get(tool_name)
+                if tool:
+                    enabled_tools.append(tool)
+
+        # Build messages list
+        messages = []
         final_system_prompt = self._build_system_prompt(context)
         messages.append({"role": "system", "content": final_system_prompt})
 
@@ -98,138 +82,36 @@ class BaseAgent(ABC):
         if conversation_history:
             for msg in conversation_history[-10:]:
                 role = "assistant" if msg["sender_type"] == "agent" else "user"
-                messages.append({
-                    "role": role,
-                    "content": msg["content"]
-                })
+                content = msg["content"]
+                # Handle JSON content format
+                if isinstance(content, str):
+                    try:
+                        content_obj = json.loads(content)
+                        if isinstance(content_obj, dict):
+                            content = content_obj.get("text", content)
+                    except json.JSONDecodeError:
+                        pass
+                messages.append({"role": role, "content": content})
 
         # Add current user message
         messages.append({"role": "user", "content": user_message})
 
-        # Get LLM client
-        llm = self._get_llm_client()
-
-        # Get tools if enabled
-        enabled_tools = []
-        if self.tools:
-            from luxx.tools.core import registry
-            for tool_name in self.tools:
-                tool = registry.get(tool_name)
-                if tool:
-                    enabled_tools.append(tool)
-
-        # Stream response
-        step_index = 0
-        full_content = ""
-
-        try:
-            async for sse_line in llm.stream_call(
-                model=self.model or llm.default_model,
-                messages=messages,
-                tools=enabled_tools if enabled_tools else None,
-                temperature=self.temperature,
-                max_tokens=self.max_tokens,
-                thinking_enabled=thinking_enabled
-            ):
-                # Parse SSE line
-                event_type = None
-                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()
-
-                if data_str is None:
-                    continue
-
-                # Handle error events
-                if event_type == 'error':
-                    try:
-                        error_data = json.loads(data_str)
-                        yield {
-                            "event": "error",
-                            "data": {"content": error_data.get("content", "Unknown error")}
-                        }
-                    except json.JSONDecodeError:
-                        yield {
-                            "event": "error",
-                            "data": {"content": data_str}
-                        }
-                    return
-
-                # Parse the data
-                try:
-                    chunk = json.loads(data_str)
-                except json.JSONDecodeError:
-                    continue
-
-                # Check for error in response
-                if "error" in chunk:
-                    error_msg = chunk["error"].get("message", str(chunk["error"]))
-                    yield {
-                        "event": "error",
-                        "data": {"content": f"API Error: {error_msg}"}
-                    }
-                    return
-
-                # Get delta
-                choices = chunk.get("choices", [])
-                if not choices:
-                    continue
-
-                delta = choices[0].get("delta", {})
-
-                # Handle reasoning (thinking)
-                reasoning = delta.get("reasoning_content", "")
-                if reasoning:
-                    step_index += 1
-                    yield {
-                        "event": "process_step",
-                        "data": {
-                            "step": {
-                                "id": f"{self.agent_id}-step-{step_index}",
-                                "type": "thinking",
-                                "content": reasoning
-                            }
-                        }
-                    }
-
-                # Handle content
-                content = delta.get("content", "")
-                if content:
-                    step_index += 1
-                    full_content += content
-                    yield {
-                        "event": "process_step",
-                        "data": {
-                            "step": {
-                                "id": f"{self.agent_id}-step-{step_index}",
-                                "type": "text",
-                                "content": full_content
-                            }
-                        }
-                    }
-
-            # Final message
-            yield {
-                "event": "done",
-                "data": {
-                    "message_id": str(uuid.uuid4()),
-                    "agent_id": self.agent_id,
-                    "agent_name": self.name,
-                    "content": full_content,
-                    "token_count": len(full_content) // 4
-                }
-            }
-
-        except Exception as e:
-            logger.error(f"Agent {self.name} stream error: {e}")
-            yield {
-                "event": "error",
-                "data": {"content": str(e)}
-            }
+        # Delegate to ChatService's core logic
+        async for sse_str in chat_service.stream_response_for_agent(
+            messages=messages,
+            model=self.model,
+            tools=enabled_tools if enabled_tools else None,
+            temperature=self.temperature,
+            max_tokens=self.max_tokens,
+            thinking_enabled=thinking_enabled,
+            provider_id=self.provider_id,
+            workspace=context.get("workspace") if context else None,
+            user_id=context.get("user_id") if context else None,
+            username=context.get("username") if context else None,
+            user_permission_level=context.get("user_permission_level", 1) if context else 1
+        ):
+            # Forward the SSE string with agent context appended
+            yield sse_str
 
     def _build_system_prompt(self, context: Dict = None) -> str:
         """Build the final system prompt with context"""
@@ -251,6 +133,7 @@ class BaseAgent(ABC):
             "role": self.role,
             "avatar": self.avatar,
             "system_prompt": self.system_prompt,
+            "provider_id": self.provider_id,
             "model": self.model,
             "tools": self.tools,
             "priority": self.priority,
diff --git a/luxx/api/agents.py b/luxx/api/agents.py
index d2d5926..697eb4f 100644
--- a/luxx/api/agents.py
+++ b/luxx/api/agents.py
@@ -37,6 +37,7 @@ class UpdateAgentRequest(BaseModel):
     max_tokens: Optional[int] = None
     is_active: Optional[bool] = None
     avatar: Optional[str] = None
+    clear_provider: bool = False
 
 
 @router.get("")
@@ -92,7 +93,8 @@ async def update_agent(agent_id: str, request: UpdateAgentRequest):
         temperature=request.temperature,
         max_tokens=request.max_tokens,
         is_active=request.is_active,
-        avatar=request.avatar
+        avatar=request.avatar,
+        clear_provider=request.clear_provider
     )
     if not agent:
         raise HTTPException(status_code=404, detail="Agent not found")
diff --git a/luxx/services/agent.py b/luxx/services/agent.py
index 247f9e1..8a2da2f 100644
--- a/luxx/services/agent.py
+++ b/luxx/services/agent.py
@@ -67,7 +67,8 @@ class AgentManager:
     def update_agent(self, agent_id: str, name: str = None, role: str = None, system_prompt: str = None,
                      provider_id: int = None, model: str = None, tools: List[str] = None, priority: int = None,
                      auto_response: bool = None, mention_trigger: bool = None, temperature: float = None,
-                     max_tokens: int = None, is_active: bool = None, avatar: str = None) -> Optional[Dict]:
+                     max_tokens: int = None, is_active: bool = None, avatar: str = None,
+                     clear_provider: bool = False) -> Optional[Dict]:
         """Update an agent"""
         db = SessionLocal()
         try:
@@ -81,12 +82,14 @@ class AgentManager:
                 agent.role = role
             if system_prompt is not None:
                 agent.system_prompt = system_prompt
-            if provider_id is not None:
+            if clear_provider:
+                agent.provider_id = None
+            elif provider_id is not None:
                 agent.provider_id = provider_id
             if model is not None:
                 agent.model = model
             if tools is not None:
-                agent.tools = json.dumps(tools)
+                agent.tools = json.dumps(tools) if tools else None
             if priority is not None:
                 agent.priority = priority
             if auto_response is not None:
diff --git a/luxx/services/chat.py b/luxx/services/chat.py
index 501c7e9..44c5efc 100644
--- a/luxx/services/chat.py
+++ b/luxx/services/chat.py
@@ -4,11 +4,10 @@ import uuid
 import logging
 from typing import List, Dict, Any, AsyncGenerator, Optional
 
-from luxx.models import Conversation, Message
+from luxx.models import Conversation
 from luxx.tools.executor import ToolExecutor
 from luxx.tools.core import registry
 from luxx.services.llm_client import LLMClient
-from luxx.core.config import config
 
 logger = logging.getLogger(__name__)
 # Maximum iterations to prevent infinite loops
@@ -20,15 +19,23 @@ def _sse_event(event: str, data: dict) -> str:
     return f"event: {event}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
 
 
-def get_llm_client(conversation: Conversation = None):
-    """Get LLM client, optionally using conversation's provider. Returns (client, max_tokens)"""
+def get_llm_client(conversation=None, provider_id: int = None):
+    """Get LLM client, optionally using conversation's or provider_id's settings. Returns (client, max_tokens)"""
     max_tokens = None
-    if conversation and conversation.provider_id:
+    target_provider_id = None
+
+    # Determine provider_id
+    if conversation and hasattr(conversation, 'provider_id'):
+        target_provider_id = conversation.provider_id
+    if provider_id:
+        target_provider_id = provider_id
+
+    if target_provider_id:
         from luxx.models import LLMProvider
         from luxx.core.database import SessionLocal
         db = SessionLocal()
         try:
-            provider = db.query(LLMProvider).filter(LLMProvider.id == conversation.provider_id).first()
+            provider = db.query(LLMProvider).filter(LLMProvider.id == target_provider_id).first()
             if provider:
                 max_tokens = provider.max_tokens
                 client = LLMClient(
@@ -39,7 +46,7 @@ def get_llm_client(conversation: Conversation = None):
                 return client, max_tokens
         finally:
             db.close()
-    
+
     # Fallback to global config
     client = LLMClient()
     return client, max_tokens
@@ -198,10 +205,10 @@ class StreamContext:
 
 class ChatService:
     """Chat service with tool support"""
-    
+
     def __init__(self):
         self.tool_executor = ToolExecutor()
-    
+
     def build_messages(
         self,
         conversation: Conversation,
@@ -210,21 +217,21 @@ class ChatService:
         """Build message list"""
         from luxx.core.database import SessionLocal
         from luxx.models import Message
-        
+
         messages = []
-        
+
         if include_system and conversation.system_prompt:
             messages.append({
                 "role": "system",
                 "content": conversation.system_prompt
             })
-        
+
         db = SessionLocal()
         try:
             db_messages = db.query(Message).filter(
                 Message.conversation_id == conversation.id
             ).order_by(Message.created_at).all()
-            
+
             for msg in db_messages:
                 # Parse JSON content if possible
                 try:
@@ -235,16 +242,16 @@ class ChatService:
                         content = msg.content
                 except (json.JSONDecodeError, TypeError):
                     content = msg.content
-                
+
                 messages.append({
                     "role": msg.role,
                     "content": content
                 })
         finally:
             db.close()
-        
+
         return messages
-    
+
     async def stream_response(
         self,
         conversation: Conversation,
@@ -257,51 +264,147 @@ class ChatService:
         user_permission_level: int = 1
     ) -> AsyncGenerator[Dict[str, str], None]:
         """
-        Streaming response generator
-        
+        Streaming response generator for user conversations.
+
         Yields raw SSE event strings for direct forwarding.
         """
+        messages = self.build_messages(conversation)
+
+        messages.append({
+            "role": "user",
+            "content": json.dumps({"text": user_message, "attachments": []})
+        })
+
+        # 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"
+        max_tokens = provider_max_tokens
+
+        async for event in self._stream_response_core(
+            messages=messages,
+            model=model,
+            tools=tools,
+            temperature=conversation.temperature,
+            max_tokens=max_tokens or 8192,
+            thinking_enabled=thinking_enabled or conversation.thinking_enabled,
+            conversation_id=conversation.id,
+            user_id=user_id,
+            username=username,
+            workspace=workspace,
+            user_permission_level=user_permission_level
+        ):
+            yield event
+
+    async def stream_response_for_agent(
+        self,
+        messages: List[Dict],
+        model: str = None,
+        tools: list = None,
+        temperature: float = 0.7,
+        max_tokens: int = 2048,
+        thinking_enabled: bool = False,
+        provider_id: int = None,
+        workspace: str = None,
+        user_id: int = None,
+        username: str = None,
+        user_permission_level: int = 1
+    ) -> AsyncGenerator[Dict[str, str], None]:
+        """
+        Streaming response generator for agents (reuses user chat logic).
+
+        Args:
+            messages: Pre-built message list (should include system prompt and history)
+            model: Model name
+            tools: List of tool definitions
+            temperature: Sampling temperature
+            max_tokens: Maximum tokens
+            thinking_enabled: Enable reasoning
+            provider_id: LLM provider ID
+            workspace: Workspace path
+            user_id: User ID
+            username: Username
+            user_permission_level: Permission level
+
+        Yields raw SSE event strings.
+        """
+        llm, provider_max_tokens = get_llm_client(provider_id=provider_id)
+        model = model or llm.default_model or "gpt-4"
+        effective_max_tokens = provider_max_tokens or max_tokens
+
+        async for event in self._stream_response_core(
+            messages=messages,
+            model=model,
+            tools=tools or [],
+            temperature=temperature,
+            max_tokens=effective_max_tokens,
+            thinking_enabled=thinking_enabled,
+            conversation_id=None,  # Agent doesn't save to conversation
+            provider_id=provider_id,
+            user_id=user_id,
+            username=username,
+            workspace=workspace,
+            user_permission_level=user_permission_level
+        ):
+            yield event
+
+    async def _stream_response_core(
+        self,
+        messages: List[Dict],
+        model: str,
+        tools: list,
+        temperature: float,
+        max_tokens: int,
+        thinking_enabled: bool,
+        conversation_id: str = None,
+        provider_id: int = None,
+        user_id: int = None,
+        username: str = None,
+        workspace: str = None,
+        user_permission_level: int = 1
+    ) -> AsyncGenerator[Dict[str, str], None]:
+        """
+        Core streaming response logic (shared by user chat and agents).
+        """
+        # Get LLM client
+        target_provider_id = provider_id if not conversation_id else None
+        llm, _ = get_llm_client(provider_id=target_provider_id)
+
+        # Token usage tracking
+        total_usage = {
+            "prompt_tokens": 0,
+            "completion_tokens": 0,
+            "total_tokens": 0
+        }
+        actual_token_count = 0
+
+        # Streaming context for state management
+        ctx = StreamContext()
+
+        # Build tool context
+        tool_context = {
+            "workspace": workspace,
+            "user_id": user_id,
+            "username": username,
+            "user_permission_level": user_permission_level
+        }
+
         try:
-            messages = self.build_messages(conversation)
-            
-            messages.append({
-                "role": "user",
-                "content": json.dumps({"text": user_message, "attachments": []})
-            })
-            
-            # 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
-            
-            # Token usage tracking
-            total_usage = {
-                "prompt_tokens": 0,
-                "completion_tokens": 0,
-                "total_tokens": 0
-            }
-            actual_token_count = 0
-            
-            # Streaming context for state management
-            ctx = StreamContext()
-            
-            for iteration in range(MAX_ITERATIONS):
+            for _ in range(MAX_ITERATIONS):
                 # Reset streaming context for this iteration
                 ctx.reset_iteration()
-                
+
                 async for sse_line in llm.stream_call(
                     model=model,
                     messages=messages,
                     tools=tools,
-                    temperature=conversation.temperature,
+                    temperature=temperature,
                     max_tokens=max_tokens or 8192,
-                    thinking_enabled=thinking_enabled or conversation.thinking_enabled
+                    thinking_enabled=thinking_enabled
                 ):
                     # Parse SSE line
                     # Format: "event: xxx\ndata: {...}\n\n"
@@ -444,34 +547,36 @@ class ChatService:
                 
                 # No tool calls - final iteration, save message
                 msg_id = str(uuid.uuid4())
-                
+
                 # 使用 API 返回的真实 completion_tokens，如果 API 没返回则降级使用估算值
                 actual_token_count = total_usage.get("completion_tokens", 0) or len(ctx.full_content) // 4
                 logger.info(f"[TOKEN] total_usage: {total_usage}, actual_token_count: {actual_token_count}")
-                
-                self._save_message(
-                    conversation.id,
-                    msg_id,
-                    ctx.full_content,
-                    ctx.all_tool_calls,
-                    ctx.all_tool_results,
-                    ctx.all_steps,
-                    actual_token_count,
-                    total_usage
-                )
-                
+
+                # Only save to DB if conversation_id is provided
+                if conversation_id:
+                    self._save_message(
+                        conversation_id,
+                        msg_id,
+                        ctx.full_content,
+                        ctx.all_tool_calls,
+                        ctx.all_tool_results,
+                        ctx.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 - save message before error
-            if ctx.full_content or ctx.all_tool_calls:
+            if conversation_id and (ctx.full_content or ctx.all_tool_calls):
                 msg_id = str(uuid.uuid4())
                 self._save_message(
-                    conversation.id,
+                    conversation_id,
                     msg_id,
                     ctx.full_content,
                     ctx.all_tool_calls,
diff --git a/luxx/services/llm_client.py b/luxx/services/llm_client.py
index b509948..bb67934 100644
--- a/luxx/services/llm_client.py
+++ b/luxx/services/llm_client.py
@@ -14,7 +14,7 @@ class LLMResponse:
     content: str
     tool_calls: Optional[List[Dict]] = None
     usage: Optional[Dict] = None
-    
+
     def __init__(
         self,
         content: str = "",
@@ -28,14 +28,14 @@ class LLMResponse:
 
 class LLMClient:
     """LLM API client with multi-provider support"""
-    
+
     def __init__(self, api_key: str = None, api_url: str = None, model: str = None):
         self.api_key = api_key or config.llm_api_key
         self.api_url = api_url or config.llm_api_url
         self.default_model = model
         self.provider = self._detect_provider()
         self._client: Optional[httpx.AsyncClient] = None
-    
+
     def _detect_provider(self) -> str:
         """Detect provider from URL"""
         url = self.api_url.lower()
@@ -46,20 +46,26 @@ class LLMClient:
         elif "openai" in url:
             return "openai"
         return "openai"
-    
+
     async def close(self):
         """Close client"""
         if self._client:
             await self._client.aclose()
             self._client = None
-    
+
     def _build_headers(self) -> Dict[str, str]:
         """Build request headers"""
+        if not self.api_key:
+            raise ValueError(
+                "LLM API key is not configured. "
+                "Please set DEEPSEEK_API_KEY environment variable or configure a provider with API key."
+            )
+
         return {
             "Content-Type": "application/json",
             "Authorization": f"Bearer {self.api_key}"
         }
-    
+
     def _build_body(
         self,
         model: str,
@@ -74,47 +80,47 @@ class LLMClient:
             "messages": messages,
             "stream": stream
         }
-        
+
         if "temperature" in kwargs:
             body["temperature"] = kwargs["temperature"]
-        
+
         if "max_tokens" in kwargs:
             body["max_tokens"] = kwargs["max_tokens"]
-        
+
         if "thinking_enabled" in kwargs and kwargs["thinking_enabled"]:
             body["thinking_enabled"] = True
-        
+
         if tools:
             body["tools"] = tools
-        
+
         return body
-    
+
     def _parse_response(self, data: Dict) -> LLMResponse:
         """Parse response"""
         content = ""
         tool_calls = None
         usage = None
-        
+
         if "choices" in data:
             choice = data["choices"][0]
             content = choice.get("message", {}).get("content", "")
             tool_calls = choice.get("message", {}).get("tool_calls")
-        
+
         if "usage" in data:
             usage = data["usage"]
-        
+
         return LLMResponse(
             content=content,
             tool_calls=tool_calls,
             usage=usage
         )
-    
+
     async def client(self) -> httpx.AsyncClient:
         """Get HTTP client"""
         if self._client is None:
             self._client = httpx.AsyncClient(timeout=120.0)
         return self._client
-    
+
     async def sync_call(
         self,
         model: str,
@@ -124,18 +130,19 @@ class LLMClient:
     ) -> LLMResponse:
         """Call LLM API (non-streaming)"""
         body = self._build_body(model, messages, tools, stream=False, **kwargs)
-        
+
         async with httpx.AsyncClient(timeout=120.0) as client:
             response = await client.post(
                 self.api_url,
                 headers=self._build_headers(),
                 json=body
             )
+
             response.raise_for_status()
             data = response.json()
-        
+
         return self._parse_response(data)
-    
+
     async def stream_call(
         self,
         model: str,
@@ -144,14 +151,14 @@ class LLMClient:
         **kwargs
     ) -> AsyncGenerator[str, None]:
         """Stream call LLM API - yields raw SSE event lines
-        
+
         Yields:
             str: Raw SSE event lines for direct forwarding
         """
         body = self._build_body(model, messages, tools, stream=True, **kwargs)
-        
+
         logger.info(f"Starting stream_call for model: {model}, messages count: {len(messages)}")
-        
+
         try:
             async with httpx.AsyncClient(timeout=120.0) as client:
                 logger.info(f"Sending request to {self.api_url}")
@@ -163,7 +170,7 @@ class LLMClient:
                 ) as response:
                     logger.info(f"Response status: {response.status_code}")
                     response.raise_for_status()
-                    
+
                     async for line in response.aiter_lines():
                         if line.strip():
                             yield line + "\n"
diff --git a/luxx/services/room.py b/luxx/services/room.py
index 1e7c798..b0209c9 100644
--- a/luxx/services/room.py
+++ b/luxx/services/room.py
@@ -80,18 +80,72 @@ class ResponseAggregator:
         if not agent_streams:
             return
 
+        import asyncio
+
+        def parse_sse(event_str: str) -> Dict[str, Any]:
+            """Parse SSE string to dict."""
+            lines = event_str.strip().split('\n')
+            result = {"event": None, "data": {}}
+            for line in lines:
+                if line.startswith('event: '):
+                    result["event"] = line[7:].strip()
+                elif line.startswith('data: '):
+                    try:
+                        result["data"] = json.loads(line[6:].strip())
+                    except json.JSONDecodeError:
+                        result["data"] = {"content": line[6:].strip()}
+            return result
+
         async def collect_agent_stream(agent_id: str, stream):
+            """Collect all events from a single agent stream."""
             try:
                 async for event in stream:
-                    event["agent_id"] = agent_id
-                    yield event
+                    # Event is SSE string from BaseAgent
+                    parsed = parse_sse(event)
+                    parsed["agent_id"] = agent_id
+                    yield parsed
             except Exception as e:
                 logger.error(f"Agent {agent_id} stream error: {e}")
                 yield {"event": "error", "agent_id": agent_id, "data": {"content": str(e)}}
 
-        tasks = [collect_agent_stream(agent_id, stream) for agent_id, stream in agent_streams.items()]
-        async for event in asyncio.merge(*tasks):
-            yield event
+        # Use a queue-based approach for merging
+        queue = asyncio.Queue()
+
+        async def producer(agent_id: str, stream):
+            try:
+                async for event in stream:
+                    # Parse SSE string to dict if needed
+                    if isinstance(event, str):
+                        parsed = parse_sse(event)
+                        parsed["agent_id"] = agent_id
+                        await queue.put((agent_id, parsed))
+                    else:
+                        # Already a dict, just add agent_id
+                        if isinstance(event, dict):
+                            event["agent_id"] = agent_id
+                        await queue.put((agent_id, event))
+            except Exception as e:
+                logger.error(f"Agent {agent_id} stream error: {e}")
+                await queue.put((agent_id, {"event": "error", "agent_id": agent_id, "data": {"content": str(e)}}))
+            finally:
+                await queue.put((agent_id, None))  # Signal done
+
+        # Start all producers
+        producers = [
+            asyncio.create_task(producer(agent_id, stream))
+            for agent_id, stream in agent_streams.items()
+        ]
+
+        active = len(producers)
+        while active > 0:
+            agent_id, event = await queue.get()
+            if event is None:
+                active -= 1
+            else:
+                yield event
+
+        # Wait for all producers to complete
+        await asyncio.gather(*producers, return_exceptions=True)
 
     def aggregate_final(self, responses: Dict[str, Dict[str, Any]]) -> List[Dict[str, Any]]:
         """Aggregate final responses from agents."""