diff --git a/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue b/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
index 393ce4e..cd1810f 100644
--- a/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
+++ b/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
@@ -49,6 +49,8 @@ const agentForm = reactive({
   presencePenalty: null as number | null,
   frequencyPenalty: null as number | null,
   maxToken: 8192 as number | null,
+  streamEnabled: 1,
+  reasoningVisible: 0,
   question: "",
 });
 const runtimeConfig = ref<AgentRuntimeConfig | null>(null);
@@ -130,6 +132,8 @@ function applyDebugModelDefaults(modelId?: number) {
     presencePenalty: model.presencePenalty ?? null,
     frequencyPenalty: model.frequencyPenalty ?? null,
     maxToken: model.maxToken ?? (model.enabled === 1 ? runtimeConfig.value?.maxToken : null) ?? 8192,
+    streamEnabled: model.streamEnabled ?? 1,
+    reasoningVisible: runtimeConfig.value?.reasoningVisible ?? 0,
   });
 }
 
@@ -293,6 +297,8 @@ async function debugAgent() {
         presencePenalty: agentForm.presencePenalty,
         frequencyPenalty: agentForm.frequencyPenalty,
         maxToken: agentForm.maxToken,
+        streamEnabled: agentForm.streamEnabled,
+        reasoningVisible: agentForm.reasoningVisible,
       },
       (chunk) => {
         currentAssistant().content += chunk;
@@ -469,6 +475,34 @@ function errorMessage(error: unknown, fallback: string) {
                 v-model:frequency-penalty="agentForm.frequencyPenalty"
                 v-model:max-token="agentForm.maxToken"
               />
+              <el-form-item class="agent-stream-setting">
+                <div class="agent-stream-setting-copy">
+                  <strong>流式输出</strong>
+                  <span>开启后后台预览会边生成边展示；关闭后等待模型完成再一次性显示。</span>
+                </div>
+                <el-switch
+                  v-model="agentForm.streamEnabled"
+                  :active-value="1"
+                  :inactive-value="0"
+                  active-text="开启"
+                  inactive-text="关闭"
+                  inline-prompt
+                />
+              </el-form-item>
+              <el-form-item class="agent-stream-setting">
+                <div class="agent-stream-setting-copy">
+                  <strong>展示思考过程</strong>
+                  <span>开启后本次后台预览可展开查看模型思考内容；不会修改用户端正式配置。</span>
+                </div>
+                <el-switch
+                  v-model="agentForm.reasoningVisible"
+                  :active-value="1"
+                  :inactive-value="0"
+                  active-text="开启"
+                  inactive-text="关闭"
+                  inline-prompt
+                />
+              </el-form-item>
             </section>
           </el-form>
         </el-tab-pane>
diff --git a/ai_knowledge_base_v2/apps/backend/app/schemas/admin.py b/ai_knowledge_base_v2/apps/backend/app/schemas/admin.py
index 80813fc..7894d58 100644
--- a/ai_knowledge_base_v2/apps/backend/app/schemas/admin.py
+++ b/ai_knowledge_base_v2/apps/backend/app/schemas/admin.py
@@ -89,6 +89,8 @@ class AgentDebugRequest(BaseModel):
     presencePenalty: float | None = Field(default=None, ge=-2, le=2)
     frequencyPenalty: float | None = Field(default=None, ge=-2, le=2)
     maxToken: int | None = Field(default=8192, ge=1, le=100000)
+    streamEnabled: int = Field(default=1, ge=0, le=1)
+    reasoningVisible: int = Field(default=0, ge=0, le=1)
 
 
 class AgentRuntimeConfigSaveRequest(BaseModel):
diff --git a/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py b/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py
index a9b6bea..60fd731 100644
--- a/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py
@@ -48,6 +48,7 @@ class AgentDebugService:
             "presence_penalty": payload.presencePenalty,
             "frequency_penalty": payload.frequencyPenalty,
             "max_token": payload.maxToken,
+            "stream_enabled": payload.streamEnabled,
         }
 
     @classmethod
@@ -62,7 +63,7 @@ class AgentDebugService:
             yield {"type": "error", "message": "模型不存在"}
             return
         try:
-            reasoning_visible = ReasoningPolicyService.is_visible(db)
+            reasoning_visible = payload.reasoningVisible == 1
             yield {
                 "type": "status",
                 "stage": "retrieving",
diff --git a/ai_knowledge_base_v2/apps/backend/app/services/model_service.py b/ai_knowledge_base_v2/apps/backend/app/services/model_service.py
index e438073..5a409bc 100644
--- a/ai_knowledge_base_v2/apps/backend/app/services/model_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/model_service.py
@@ -169,7 +169,7 @@ def _copy_model_with_overrides(model: ModelConfig, overrides: dict[str, Any]) ->
         frequency_penalty=overrides.get("frequency_penalty"),
         max_token=overrides.get("max_token"),
         context_window=model.context_window,
-        stream_enabled=model.stream_enabled,
+        stream_enabled=overrides.get("stream_enabled", model.stream_enabled),
         response_format=model.response_format,
         extra_params=model.extra_params,
         remark=model.remark,
diff --git a/ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py b/ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py
index 797519d..69cdf5a 100644
--- a/ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py
+++ b/ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py
@@ -13,12 +13,13 @@ from app.models import Base
 from app.models.admin import Admin
 from app.models.ai_config import ModelConfig, SystemConfig
 from app.schemas.admin import AgentDebugRequest, AgentRuntimeConfigSaveRequest
+from app.services.agent_debug_service import AgentDebugService
 from app.services.model_stream_service import (
     AsyncStreamingModelResponse,
     _openai_stream_payload,
     _stream_configured_model_async,
 )
-from app.services.model_service import ModelClientService, _max_output_tokens
+from app.services.model_service import ModelClientService, _copy_model_with_overrides, _max_output_tokens
 from app.services.rag_service import RagResult
 
 
@@ -146,6 +147,23 @@ def test_disabled_stream_returns_one_complete_chunk():
     assert chunks == [answer]
 
 
+def test_debug_stream_setting_overrides_model_without_changing_it():
+    model = _model()
+    debug_model = _copy_model_with_overrides(model, {"stream_enabled": 0})
+
+    assert model.stream_enabled == 1
+    assert debug_model.stream_enabled == 0
+
+    payload = AgentDebugRequest(
+        promptContent="测试",
+        modelId=1,
+        question="测试",
+        streamEnabled=0,
+        reasoningVisible=1,
+    )
+    assert AgentDebugService.overrides(payload)["stream_enabled"] == 0
+
+
 @pytest.mark.parametrize("reasoning_visible", [True, False])
 def test_agent_debug_stream_respects_reasoning_visibility(reasoning_visible):
     async def chunks():
@@ -179,7 +197,7 @@ def test_agent_debug_stream_respects_reasoning_visibility(reasoning_visible):
         db.add(SystemConfig(
             id=1,
             config_key="show_model_reasoning",
-            config_value="true" if reasoning_visible else "false",
+            config_value="false" if reasoning_visible else "true",
         ))
         db.commit()
         payload = AgentDebugRequest(
@@ -188,6 +206,7 @@ def test_agent_debug_stream_respects_reasoning_visibility(reasoning_visible):
             knowledgeIds=[],
             question="怎么冷静",
             maxToken=8192,
+            reasoningVisible=1 if reasoning_visible else 0,
         )
         with (
             patch(