diff --git a/server/cmd/check-providers/main.go b/server/cmd/check-providers/main.go
new file mode 100644
index 0000000..01079ab
--- /dev/null
+++ b/server/cmd/check-providers/main.go
@@ -0,0 +1,133 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"log"
+	"os"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+	"github.com/joho/godotenv"
+)
+
+func main() {
+	// 加载环境变量
+	_ = godotenv.Load("../../.env")
+
+	dbURL := os.Getenv("DATABASE_URL")
+	if dbURL == "" {
+		dbURL = "postgres://freedak:@localhost:5432/govai_portal?sslmode=disable"
+	}
+
+	pool, err := pgxpool.New(context.Background(), dbURL)
+	if err != nil {
+		log.Fatalf("无法连接数据库: %v", err)
+	}
+	defer pool.Close()
+
+	fmt.Println("=== 数据库中的模型提供商配置 ===")
+
+	rows, err := pool.Query(context.Background(), `
+		SELECT id, name, base_url, 
+		       SUBSTRING(api_key_encrypted, 1, 20) as api_key_prefix,
+		       models, is_active, priority, config
+		FROM model_providers 
+		ORDER BY priority DESC, created_at
+	`)
+	if err != nil {
+		log.Fatalf("查询失败: %v", err)
+	}
+	defer rows.Close()
+
+	count := 0
+	for rows.Next() {
+		var id, name, baseURL, apiKeyPrefix string
+		var modelsJSON, configJSON []byte
+		var isActive bool
+		var priority int
+
+		err := rows.Scan(&id, &name, &baseURL, &apiKeyPrefix, &modelsJSON, &isActive, &priority, &configJSON)
+		if err != nil {
+			log.Printf("扫描行失败: %v", err)
+			continue
+		}
+
+		count++
+		fmt.Printf("【提供商 %d】\n", count)
+		fmt.Printf("ID:           %s\n", id)
+		fmt.Printf("名称:         %s\n", name)
+		fmt.Printf("Base URL:     %s\n", baseURL)
+		fmt.Printf("API Key:      %s...\n", apiKeyPrefix)
+		fmt.Printf("激活状态:     %v\n", isActive)
+		fmt.Printf("优先级:       %d\n", priority)
+
+		// 解析 models
+		var models []map[string]interface{}
+		if err := json.Unmarshal(modelsJSON, &models); err == nil {
+			fmt.Printf("可用模型:     ")
+			for i, m := range models {
+				if i > 0 {
+					fmt.Print(", ")
+				}
+				fmt.Printf("%v", m["id"])
+			}
+			fmt.Println()
+		}
+
+		// 解析 config
+		var config map[string]interface{}
+		if err := json.Unmarshal(configJSON, &config); err == nil {
+			fmt.Printf("Provider 类型: %v\n", config["provider"])
+			fmt.Printf("默认模型:     %v\n", config["default_model"])
+		}
+
+		fmt.Println()
+	}
+
+	if count == 0 {
+		fmt.Println("❌ 数据库中没有任何模型提供商配置")
+		fmt.Println("提示: 运行种子数据脚本来初始化配置")
+	} else {
+		fmt.Printf("✅ 共找到 %d 个提供商配置\n", count)
+	}
+
+	// 查询应用配置中的模型使用情况
+	fmt.Println("\n=== 应用配置中的模型使用情况 ===")
+
+	appRows, err := pool.Query(context.Background(), `
+		SELECT name, 
+		       COALESCE(app_config->>'model', '未配置') as model,
+		       status
+		FROM applications 
+		WHERE status = 'approved'
+		ORDER BY created_at DESC
+		LIMIT 10
+	`)
+	if err != nil {
+		log.Printf("查询应用失败: %v", err)
+		return
+	}
+	defer appRows.Close()
+
+	appCount := 0
+	modelStats := make(map[string]int)
+
+	for appRows.Next() {
+		var name, model, status string
+		if err := appRows.Scan(&name, &model, &status); err != nil {
+			continue
+		}
+		appCount++
+		modelStats[model]++
+		fmt.Printf("应用: %-30s | 模型: %-15s | 状态: %s\n", name, model, status)
+	}
+
+	if appCount > 0 {
+		fmt.Printf("\n✅ 共查询到 %d 个已批准的应用\n\n", appCount)
+		fmt.Println("模型使用统计:")
+		for model, count := range modelStats {
+			fmt.Printf("  - %-15s: %d 个应用\n", model, count)
+		}
+	}
+}
diff --git a/server/cmd/clear-app-models/main.go b/server/cmd/clear-app-models/main.go
new file mode 100644
index 0000000..edce8fe
--- /dev/null
+++ b/server/cmd/clear-app-models/main.go
@@ -0,0 +1,112 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+	"github.com/joho/godotenv"
+)
+
+func main() {
+	// 加载环境变量
+	_ = godotenv.Load("../../.env")
+
+	dbURL := os.Getenv("DATABASE_URL")
+	if dbURL == "" {
+		dbURL = "postgres://freedak:@localhost:5432/govai_portal?sslmode=disable"
+	}
+
+	pool, err := pgxpool.New(context.Background(), dbURL)
+	if err != nil {
+		log.Fatalf("无法连接数据库: %v", err)
+	}
+	defer pool.Close()
+
+	fmt.Println("=== 清空应用硬编码模型配置 ===")
+	fmt.Println("目标：让所有应用自动使用优先级最高的 Provider 的默认模型")
+
+	// 1. 查看当前状态
+	fmt.Println("【步骤1】查看当前应用模型配置...")
+	var totalApps, configuredApps int
+	err = pool.QueryRow(context.Background(), `
+		SELECT 
+			COUNT(*) as total,
+			COUNT(CASE WHEN app_config->>'model' IS NOT NULL AND app_config->>'model' != '' THEN 1 END) as configured
+		FROM applications 
+		WHERE status = 'approved'
+	`).Scan(&totalApps, &configuredApps)
+
+	if err != nil {
+		log.Fatalf("查询失败: %v", err)
+	}
+
+	fmt.Printf("  - 已批准应用总数: %d\n", totalApps)
+	fmt.Printf("  - 配置了模型的应用: %d\n", configuredApps)
+	fmt.Printf("  - 未配置模型的应用: %d\n\n", totalApps-configuredApps)
+
+	// 2. 清空所有应用的模型配置
+	fmt.Println("【步骤2】清空所有应用的模型配置...")
+	result, err := pool.Exec(context.Background(), `
+		UPDATE applications
+		SET app_config = app_config - 'model'
+		WHERE status = 'approved'
+		  AND app_config ? 'model'
+	`)
+
+	if err != nil {
+		log.Fatalf("更新失败: %v", err)
+	}
+
+	rowsAffected := result.RowsAffected()
+	fmt.Printf("✅ 已清空 %d 个应用的模型配置\n\n", rowsAffected)
+
+	// 3. 验证结果
+	fmt.Println("【步骤3】验证结果...")
+	var remainingConfigured int
+	err = pool.QueryRow(context.Background(), `
+		SELECT COUNT(*)
+		FROM applications 
+		WHERE status = 'approved'
+		  AND app_config->>'model' IS NOT NULL 
+		  AND app_config->>'model' != ''
+	`).Scan(&remainingConfigured)
+
+	if err != nil {
+		log.Fatalf("验证查询失败: %v", err)
+	}
+
+	if remainingConfigured == 0 {
+		fmt.Println("✅ 所有应用的模型配置已清空")
+	} else {
+		fmt.Printf("⚠️  还有 %d 个应用仍有模型配置\n", remainingConfigured)
+	}
+
+	// 4. 显示当前优先级最高的 Provider
+	fmt.Println("\n【步骤4】当前优先级最高的 Provider...")
+	var providerName, defaultModel string
+	var priority int
+	err = pool.QueryRow(context.Background(), `
+		SELECT name, COALESCE(config->>'default_model', '未配置'), priority
+		FROM model_providers
+		WHERE is_active = true
+		ORDER BY priority DESC, created_at
+		LIMIT 1
+	`).Scan(&providerName, &defaultModel, &priority)
+
+	if err != nil {
+		log.Printf("查询 Provider 失败: %v", err)
+	} else {
+		fmt.Printf("  Provider: %s\n", providerName)
+		fmt.Printf("  默认模型: %s\n", defaultModel)
+		fmt.Printf("  优先级: %d\n", priority)
+	}
+
+	fmt.Println("\n=== 配置清理完成 ===")
+	fmt.Println("\n说明：")
+	fmt.Println("  - 现在所有应用将自动使用优先级最高的 Provider 的默认模型")
+	fmt.Println("  - 当前会使用: " + providerName + " / " + defaultModel)
+	fmt.Println("  - 如需特殊应用使用指定模型，可在应用配置中单独设置")
+}
diff --git a/server/cmd/fix-providers/main.go b/server/cmd/fix-providers/main.go
new file mode 100644
index 0000000..f7e4b97
--- /dev/null
+++ b/server/cmd/fix-providers/main.go
@@ -0,0 +1,113 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+	"github.com/joho/godotenv"
+)
+
+func main() {
+	// 加载环境变量
+	_ = godotenv.Load("../../.env")
+
+	dbURL := os.Getenv("DATABASE_URL")
+	if dbURL == "" {
+		dbURL = "postgres://freedak:@localhost:5432/govai_portal?sslmode=disable"
+	}
+
+	pool, err := pgxpool.New(context.Background(), dbURL)
+	if err != nil {
+		log.Fatalf("无法连接数据库: %v", err)
+	}
+	defer pool.Close()
+
+	fmt.Println("=== 开始修复数据库配置 ===")
+
+	// 1. 修复本地模型配置
+	fmt.Println("【任务1】修复本地模型配置...")
+	result, err := pool.Exec(context.Background(), `
+		UPDATE model_providers 
+		SET config = '{
+			"provider": "openai",
+			"default_model": "qwen2.5-7b-instruct",
+			"supports_streaming": true,
+			"supports_function_calling": false
+		}'::jsonb,
+		models = '[
+			{"id": "qwen2.5-7b-instruct", "name": "Qwen2.5-7B-Instruct", "type": "chat"}
+		]'::jsonb
+		WHERE name = '本地LLM (Qwen2.5-7B)'
+	`)
+	if err != nil {
+		log.Printf("❌ 修复本地模型失败: %v", err)
+	} else {
+		rowsAffected := result.RowsAffected()
+		if rowsAffected > 0 {
+			fmt.Printf("✅ 成功修复本地模型配置 (影响 %d 行)\n\n", rowsAffected)
+		} else {
+			fmt.Println("⚠️  未找到本地模型记录")
+		}
+	}
+
+	// 2. 为未配置模型的应用设置默认模型
+	fmt.Println("【任务2】为未配置模型的应用设置默认模型...")
+	result, err = pool.Exec(context.Background(), `
+		UPDATE applications
+		SET app_config = jsonb_set(
+			COALESCE(app_config, '{}'::jsonb),
+			'{model}',
+			'"qwen2.5-7b-instruct"'
+		)
+		WHERE (app_config->>'model' IS NULL OR app_config->>'model' = '')
+		  AND status = 'approved'
+	`)
+	if err != nil {
+		log.Printf("❌ 更新应用配置失败: %v", err)
+	} else {
+		rowsAffected := result.RowsAffected()
+		if rowsAffected > 0 {
+			fmt.Printf("✅ 成功为 %d 个应用设置默认模型\n\n", rowsAffected)
+		} else {
+			fmt.Println("ℹ️  所有应用已配置模型")
+		}
+	}
+
+	// 3. 验证修复结果
+	fmt.Println("【验证】检查修复结果...")
+
+	var localConfig, localModels string
+	err = pool.QueryRow(context.Background(), `
+		SELECT 
+			COALESCE(config::text, 'null') as config,
+			COALESCE(models::text, 'null') as models
+		FROM model_providers 
+		WHERE name = '本地LLM (Qwen2.5-7B)'
+	`).Scan(&localConfig, &localModels)
+
+	if err != nil {
+		log.Printf("查询验证失败: %v", err)
+	} else {
+		fmt.Printf("本地模型 config: %s\n", localConfig)
+		fmt.Printf("本地模型 models: %s\n\n", localModels)
+	}
+
+	var unconfiguredCount int
+	err = pool.QueryRow(context.Background(), `
+		SELECT COUNT(*) 
+		FROM applications 
+		WHERE (app_config->>'model' IS NULL OR app_config->>'model' = '')
+		  AND status = 'approved'
+	`).Scan(&unconfiguredCount)
+
+	if err != nil {
+		log.Printf("查询未配置应用数量失败: %v", err)
+	} else {
+		fmt.Printf("未配置模型的应用数量: %d\n", unconfiguredCount)
+	}
+
+	fmt.Println("\n=== 修复完成 ===")
+}
diff --git a/server/cmd/server/router.go b/server/cmd/server/router.go
index a901508..edb0a13 100644
--- a/server/cmd/server/router.go
+++ b/server/cmd/server/router.go
@@ -44,7 +44,7 @@ func newRouter(cfg *config.Config, pool *pgxpool.Pool, rdb *redis.Client) http.H
 	// LLM Manager — direct model calls replacing Dify for chat
 	llmMgr := llm.NewManager()
 	llmMgr.SetPool(pool)
-	
+
 	// 优先从数据库加载 providers（如果失败，回退到环境变量配置）
 	if err := llmMgr.LoadProvidersFromDB(context.Background()); err != nil {
 		// 回退：使用环境变量配置
diff --git a/server/internal/handler/analysis_template.go b/server/internal/handler/analysis_template.go
index 955ff68..6c576b1 100644
--- a/server/internal/handler/analysis_template.go
+++ b/server/internal/handler/analysis_template.go
@@ -27,13 +27,14 @@ func NewAnalysisTemplateHandler(pool *pgxpool.Pool, manager *llm.Manager, provid
 	return &AnalysisTemplateHandler{pool: pool, manager: manager, provider: provider}
 }
 
-// getProvider 获取当前激活的 provider
-func (h *AnalysisTemplateHandler) getProvider(ctx context.Context) (llm.Provider, error) {
-	provider, err := h.manager.GetActiveProvider(ctx)
+// getProviderWithModel 获取当前激活的 provider 及其默认模型
+func (h *AnalysisTemplateHandler) getProviderWithModel(ctx context.Context) (llm.Provider, string, error) {
+	provider, defaultModel, err := h.manager.GetActiveProviderWithModel(ctx)
 	if err == nil {
-		return provider, nil
+		return provider, defaultModel, nil
 	}
-	return h.manager.GetProvider(h.provider)
+	provider, err = h.manager.GetProvider(h.provider)
+	return provider, "", err
 }
 
 func (h *AnalysisTemplateHandler) ListTemplates(w http.ResponseWriter, r *http.Request) {
@@ -234,8 +235,21 @@ func (h *AnalysisTemplateHandler) GenerateReport(w http.ResponseWriter, r *http.
 		}
 	}
 
+	// 获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.getProviderWithModel(r.Context())
+	if err != nil {
+		response.InternalError(w, "获取模型提供商失败")
+		return
+	}
+
+	// 使用优先级：Provider 默认模型 > 应用配置模型
+	modelToUse := defaultModel
+	if modelToUse == "" {
+		modelToUse = appModel // 降级时使用应用配置
+	}
+
 	llmReq := &llm.ChatRequest{
-		Model: appModel,
+		Model: modelToUse,
 		Messages: []llm.Message{
 			{Role: llm.RoleSystem, Content: "你是一位资深的政务数据分析和研判专家。要求：1)基于提供的真实数据库数据进行分析 2)报告必须完整输出到最后一个章节，不能中途截断 3)保持精炼，每个章节500字以内 4)使用Markdown格式排版 5)表格数据必须引用真实数据 6)报告输出完毕后，必须在最后一行单独输出「---\\n\\n**【完稿】**」作为完成标记"},
 			{Role: llm.RoleUser, Content: prompt},
@@ -246,14 +260,7 @@ func (h *AnalysisTemplateHandler) GenerateReport(w http.ResponseWriter, r *http.
 	}
 
 	startTime := time.Now()
-	
-	// 获取激活的 provider
-	provider, err := h.getProvider(r.Context())
-	if err != nil {
-		response.InternalError(w, "获取模型提供商失败")
-		return
-	}
-	
+
 	body, err := h.manager.ChatStream(r.Context(), provider, llmReq)
 	if err != nil {
 		response.Error(w, http.StatusBadGateway, 50202, "模型服务不可用: "+err.Error())
diff --git a/server/internal/handler/chat_llm.go b/server/internal/handler/chat_llm.go
index 5088318..8ffde92 100644
--- a/server/internal/handler/chat_llm.go
+++ b/server/internal/handler/chat_llm.go
@@ -34,16 +34,17 @@ func NewLLMChatHandler(pool *pgxpool.Pool, manager *llm.Manager, defaultProvider
 	return &LLMChatHandler{pool: pool, manager: manager, provider: defaultProvider, rdb: rdb, workerURL: workerURL, embedder: embedder}
 }
 
-// getProvider 获取当前激活的 provider，优先从数据库读取
-func (h *LLMChatHandler) getProvider(ctx context.Context) (llm.Provider, error) {
-	// 尝试从数据库获取激活的 provider
-	provider, err := h.manager.GetActiveProvider(ctx)
+// getProviderWithModel 获取当前激活的 provider 及其默认模型
+func (h *LLMChatHandler) getProviderWithModel(ctx context.Context) (llm.Provider, string, error) {
+	// 尝试从数据库获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.manager.GetActiveProviderWithModel(ctx)
 	if err == nil {
-		return provider, nil
+		return provider, defaultModel, nil
 	}
-	
+
 	// 降级到配置的 fallback provider
-	return h.manager.GetProvider(h.provider)
+	provider, err = h.manager.GetProvider(h.provider)
+	return provider, "", err // 降级时返回空模型，使用应用配置
 }
 
 type llmChatRequest struct {
@@ -568,8 +569,21 @@ func (h *LLMChatHandler) Chat(w http.ResponseWriter, r *http.Request) {
 		history = h.loadConversationHistory(r.Context(), appID, userID.String(), convID, 10)
 	}
 
+	// 获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.getProviderWithModel(r.Context())
+	if err != nil {
+		response.InternalError(w, "获取模型提供商失败")
+		return
+	}
+
+	// 使用优先级：Provider 默认模型 > 应用配置模型
+	modelToUse := defaultModel
+	if modelToUse == "" {
+		modelToUse = cfg.Model // 降级时使用应用配置
+	}
+
 	llmReq := &llm.ChatRequest{
-		Model:       cfg.Model,
+		Model:       modelToUse,
 		Messages:    h.buildMessages(cfg.SystemPrompt, knowledgeCtx, hasKB, history, req.Message, orgApps),
 		Temperature: cfg.Temp,
 		MaxTokens:   cfg.MaxTok,
@@ -577,14 +591,7 @@ func (h *LLMChatHandler) Chat(w http.ResponseWriter, r *http.Request) {
 	}
 
 	startTime := time.Now()
-	
-	// 获取激活的 provider
-	provider, err := h.getProvider(r.Context())
-	if err != nil {
-		response.InternalError(w, "获取模型提供商失败")
-		return
-	}
-	
+
 	body, err := h.manager.ChatStream(r.Context(), provider, llmReq)
 	if err != nil {
 		response.Error(w, http.StatusBadGateway, 50202, "模型服务不可用: "+err.Error())
@@ -673,8 +680,21 @@ func (h *LLMChatHandler) Completion(w http.ResponseWriter, r *http.Request) {
 	// 加载同机构应用列表，用于超范围引导跳转
 	orgApps := h.loadSameOrgApps(r.Context(), cfg.OrgID, appID)
 
+	// 获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.getProviderWithModel(r.Context())
+	if err != nil {
+		response.InternalError(w, "获取模型提供商失败")
+		return
+	}
+
+	// 使用优先级：Provider 默认模型 > 应用配置模型
+	modelToUse := defaultModel
+	if modelToUse == "" {
+		modelToUse = cfg.Model // 降级时使用应用配置
+	}
+
 	llmReq := &llm.ChatRequest{
-		Model:       cfg.Model,
+		Model:       modelToUse,
 		Messages:    h.buildMessages(cfg.SystemPrompt, knowledgeCtx, hasKB, nil, req.Message, orgApps),
 		Temperature: cfg.Temp,
 		MaxTokens:   cfg.MaxTok,
@@ -684,13 +704,6 @@ func (h *LLMChatHandler) Completion(w http.ResponseWriter, r *http.Request) {
 	startTime := time.Now()
 	convID := uuid.New().String()
 
-	// 获取激活的 provider
-	provider, err := h.getProvider(r.Context())
-	if err != nil {
-		response.InternalError(w, "获取模型提供商失败")
-		return
-	}
-
 	body, err := h.manager.ChatStream(r.Context(), provider, llmReq)
 	if err != nil {
 		response.Error(w, http.StatusBadGateway, 50202, "模型服务不可用: "+err.Error())
@@ -1176,13 +1189,18 @@ func (h *LLMChatHandler) generateConversationName(appID, userID, convID, userMes
 		Stream:      false,
 	}
 
-	// 获取激活的 provider
-	provider, err := h.getProvider(ctx)
+	// 获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.getProviderWithModel(ctx)
 	if err != nil {
 		log.Warn().Err(err).Msg("get provider for title generation failed")
 		return
 	}
 
+	// 使用 Provider 默认模型
+	if defaultModel != "" {
+		nameReq.Model = defaultModel
+	}
+
 	result, err := h.manager.Chat(ctx, provider, nameReq)
 	if err != nil {
 		return
diff --git a/server/internal/handler/doc_template.go b/server/internal/handler/doc_template.go
index b30bbf8..60d184d 100644
--- a/server/internal/handler/doc_template.go
+++ b/server/internal/handler/doc_template.go
@@ -27,13 +27,14 @@ func NewDocTemplateHandler(pool *pgxpool.Pool, manager *llm.Manager, provider st
 	return &DocTemplateHandler{pool: pool, manager: manager, provider: provider}
 }
 
-// getProvider 获取当前激活的 provider
-func (h *DocTemplateHandler) getProvider(ctx context.Context) (llm.Provider, error) {
-	provider, err := h.manager.GetActiveProvider(ctx)
+// getProviderWithModel 获取当前激活的 provider 及其默认模型
+func (h *DocTemplateHandler) getProviderWithModel(ctx context.Context) (llm.Provider, string, error) {
+	provider, defaultModel, err := h.manager.GetActiveProviderWithModel(ctx)
 	if err == nil {
-		return provider, nil
+		return provider, defaultModel, nil
 	}
-	return h.manager.GetProvider(h.provider)
+	provider, err = h.manager.GetProvider(h.provider)
+	return provider, "", err
 }
 
 func (h *DocTemplateHandler) ListTemplates(w http.ResponseWriter, r *http.Request) {
@@ -171,8 +172,21 @@ func (h *DocTemplateHandler) GenerateDocument(w http.ResponseWriter, r *http.Req
 	prompt = strings.ReplaceAll(prompt, "{{", "")
 	prompt = strings.ReplaceAll(prompt, "}}", "")
 
+	// 获取激活的 provider 和默认模型
+	provider, defaultModel, err := h.getProviderWithModel(r.Context())
+	if err != nil {
+		response.InternalError(w, "获取模型提供商失败")
+		return
+	}
+
+	// 使用优先级：Provider 默认模型 > 应用配置模型
+	modelToUse := defaultModel
+	if modelToUse == "" {
+		modelToUse = appModel // 降级时使用应用配置
+	}
+
 	llmReq := &llm.ChatRequest{
-		Model: appModel,
+		Model: modelToUse,
 		Messages: []llm.Message{
 			{Role: llm.RoleSystem, Content: "你是一个专业的政务公文写作专家，精通《党政机关公文格式》国家标准（GB/T 9704）和《党政机关公文处理工作条例》。请严格按照规范格式生成公文，确保行文庄重、严谨、准确。输出完整公文内容，使用Markdown格式排版。"},
 			{Role: llm.RoleUser, Content: prompt},
@@ -183,14 +197,7 @@ func (h *DocTemplateHandler) GenerateDocument(w http.ResponseWriter, r *http.Req
 	}
 
 	startTime := time.Now()
-	
-	// 获取激活的 provider
-	provider, err := h.getProvider(r.Context())
-	if err != nil {
-		response.InternalError(w, "获取模型提供商失败")
-		return
-	}
-	
+
 	body, err := h.manager.ChatStream(r.Context(), provider, llmReq)
 	if err != nil {
 		response.Error(w, http.StatusBadGateway, 50202, "模型服务不可用: "+err.Error())
diff --git a/server/pkg/llm/manager.go b/server/pkg/llm/manager.go
index bef7c37..ed6e9f4 100644
--- a/server/pkg/llm/manager.go
+++ b/server/pkg/llm/manager.go
@@ -116,12 +116,18 @@ func (m *Manager) LoadProvidersFromDB(ctx context.Context) error {
 
 // GetActiveProvider 获取优先级最高的激活 provider
 func (m *Manager) GetActiveProvider(ctx context.Context) (Provider, error) {
+	provider, _, err := m.GetActiveProviderWithModel(ctx)
+	return provider, err
+}
+
+// GetActiveProviderWithModel 获取优先级最高的激活 provider 及其默认模型
+func (m *Manager) GetActiveProviderWithModel(ctx context.Context) (Provider, string, error) {
 	if m.pool == nil {
 		// 如果没有数据库连接，使用注册的 fallback
 		if p, ok := m.providers[m.fallback]; ok {
-			return p, nil
+			return p, "", nil // 降级时返回空模型名，使用应用配置
 		}
-		return nil, fmt.Errorf("no active provider")
+		return nil, "", fmt.Errorf("no active provider")
 	}
 
 	// 从数据库获取优先级最高的激活 provider
@@ -135,21 +141,25 @@ func (m *Manager) GetActiveProvider(ctx context.Context) (Provider, error) {
 		LIMIT 1
 	`).Scan(&id, &baseURL, &apiKey, &configJSON)
 	if err != nil {
-		return nil, fmt.Errorf("query active provider: %w", err)
+		return nil, "", fmt.Errorf("query active provider: %w", err)
 	}
 
 	// 检查缓存
 	m.cacheMu.RLock()
 	if entry, ok := m.cache[id]; ok && time.Since(entry.updatedAt) < m.cacheTTL {
 		m.cacheMu.RUnlock()
-		return entry.provider, nil
+		// 从 config 中提取默认模型
+		var config map[string]any
+		json.Unmarshal(configJSON, &config)
+		defaultModel, _ := config["default_model"].(string)
+		return entry.provider, defaultModel, nil
 	}
 	m.cacheMu.RUnlock()
 
 	// 创建新的 provider
 	var config map[string]any
 	if err := json.Unmarshal(configJSON, &config); err != nil {
-		return nil, fmt.Errorf("unmarshal config: %w", err)
+		return nil, "", fmt.Errorf("unmarshal config: %w", err)
 	}
 
 	providerType, _ := config["provider"].(string)
@@ -157,6 +167,9 @@ func (m *Manager) GetActiveProvider(ctx context.Context) (Provider, error) {
 		providerType = "openai"
 	}
 
+	// 获取默认模型
+	defaultModel, _ := config["default_model"].(string)
+
 	var provider Provider
 	switch providerType {
 	case "openai":
@@ -164,7 +177,7 @@ func (m *Manager) GetActiveProvider(ctx context.Context) (Provider, error) {
 	case "anthropic":
 		provider = NewAnthropicProvider(apiKey, baseURL, "")
 	default:
-		return nil, fmt.Errorf("unsupported provider type: %s", providerType)
+		return nil, "", fmt.Errorf("unsupported provider type: %s", providerType)
 	}
 
 	// 更新缓存
@@ -175,7 +188,7 @@ func (m *Manager) GetActiveProvider(ctx context.Context) (Provider, error) {
 	}
 	m.cacheMu.Unlock()
 
-	return provider, nil
+	return provider, defaultModel, nil
 }
 
 func (m *Manager) GetProvider(name string) (Provider, error) {
@@ -208,9 +221,9 @@ type StreamEvent struct {
 }
 
 type Usage struct {
-	PromptTokens     int `json:"prompt_tokens"`
-	CompletionTokens int `json:"completion_tokens"`
-	TotalTokens      int `json:"total_tokens"`
+	PromptTokens     int    `json:"prompt_tokens"`
+	CompletionTokens int    `json:"completion_tokens"`
+	TotalTokens      int    `json:"total_tokens"`
 	Model            string `json:"model"`
 }