From d8b124fc7c1fbfa5e9f854e85f55cf54925a324c Mon Sep 17 00:00:00 2001
From: "Logan.Li" <admin@attacker.club>
Date: Thu, 30 Jul 2026 21:34:33 +0800
Subject: [PATCH] =?UTF-8?q?feat:=20=E6=9B=B4=E6=96=B0=E4=B8=BB=E6=B5=81?=
 =?UTF-8?q?=E5=A4=A7=E6=A8=A1=E5=9E=8B=E4=BE=9B=E5=BA=94=E5=95=86=E7=9B=AE?=
 =?UTF-8?q?=E5=BD=95?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 按云服务、Coding Plan、API 网关和本地模型分组

- 新增兼容模型发现与供应商协议校验
---
 backend/controller/llm.go                     |  77 ++++--
 backend/controller/llm_chat.go                |  10 +-
 backend/controller/llm_models.go              | 181 +++++++++++++
 backend/controller/llm_models_test.go         |  96 +++++++
 backend/controller/llm_provider_test.go       |  70 +++++
 backend/model/llm.go                          |  34 +--
 backend/pkg/agentcli/llm/anthropic.go         |  12 +-
 backend/pkg/agentcli/llm/openai.go            |  46 ++--
 .../pkg/agentcli/llm/provider_compat_test.go  |  39 +++
 backend/router/routes.go                      |   1 +
 frontend/src/api/aiops/assistant.ts           |   8 +
 frontend/src/components/aiops/LLMConfig.tsx   | 254 ++++++------------
 frontend/src/components/aiops/llmProviders.ts | 193 +++++++++++++
 frontend/src/locales/en-US.json               |  27 +-
 frontend/src/locales/zh-CN.json               |  27 +-
 15 files changed, 838 insertions(+), 237 deletions(-)
 create mode 100644 backend/controller/llm_models.go
 create mode 100644 backend/controller/llm_models_test.go
 create mode 100644 backend/controller/llm_provider_test.go
 create mode 100644 backend/pkg/agentcli/llm/provider_compat_test.go
 create mode 100644 frontend/src/components/aiops/llmProviders.ts

diff --git a/backend/controller/llm.go b/backend/controller/llm.go
index b524ff2..ca927bd 100644
--- a/backend/controller/llm.go
+++ b/backend/controller/llm.go
@@ -34,10 +34,20 @@ func maskAPIKey(key string) string {
 	return strings.Repeat("*", len(key)-4) + key[len(key)-4:]
 }
 
+// canReuseMaskedAPIKey 仅允许原提供商与原 Endpoint 继续使用数据库中的密钥。 // Logan.Li
+func canReuseMaskedAPIKey(provider, endpoint string, saved model.LLMConfig) bool {
+	if saved.ID == 0 || saved.Provider != provider {
+		return false
+	}
+	requestedEndpoint := resolveEndpoint(provider, strings.TrimSpace(endpoint))
+	savedEndpoint := resolveEndpoint(saved.Provider, strings.TrimSpace(saved.Endpoint))
+	return requestedEndpoint != "" && requestedEndpoint == savedEndpoint
+}
+
 // getActiveConfig 读取启用的配置（脱敏后） // Logan.Li
 func (h *LLMHandler) getActiveConfig(maskKey bool) (model.LLMConfig, error) {
 	var cfg model.LLMConfig
-	if err := h.db.First(&cfg).Error; err != nil {
+	if err := h.db.Take(&cfg).Error; err != nil {
 		return cfg, err
 	}
 	if maskKey {
@@ -60,17 +70,17 @@ func (h *LLMHandler) GetLLMConfig(c *gin.Context) {
 		return
 	}
 	c.JSON(http.StatusOK, model.Response{Code: 200, Data: gin.H{
-		"id":               cfg.ID,
-		"provider":         cfg.Provider,
-		"model":            cfg.Model,
-		"api_key":          cfg.APIKey,
-		"endpoint":         cfg.Endpoint,
-		"temperature":      cfg.Temperature,
-		"max_tokens":       cfg.MaxTokens,
-		"system_prompt":    cfg.SystemPrompt,
-		"enabled":          cfg.Enabled,
+		"id":                cfg.ID,
+		"provider":          cfg.Provider,
+		"model":             cfg.Model,
+		"api_key":           cfg.APIKey,
+		"endpoint":          cfg.Endpoint,
+		"temperature":       cfg.Temperature,
+		"max_tokens":        cfg.MaxTokens,
+		"system_prompt":     cfg.SystemPrompt,
+		"enabled":           cfg.Enabled,
 		"reasoning_enabled": cfg.ReasoningEnabled,
-		"configured":       true,
+		"configured":        true,
 	}})
 }
 
@@ -83,9 +93,13 @@ func (h *LLMHandler) SaveLLMConfig(c *gin.Context) {
 		return
 	}
 	var cfg model.LLMConfig
-	h.db.First(&cfg)
-	// 若前端回传脱敏值（含 ****），保留原 Key
-	if strings.Contains(req.APIKey, "****") && cfg.ID != 0 {
+	h.db.Take(&cfg)
+	// 脱敏密钥只能在提供商和 Endpoint 均未改变时复用，避免凭证被转发到其他地址。 // Logan.Li
+	if strings.Contains(req.APIKey, "****") {
+		if !canReuseMaskedAPIKey(req.Provider, req.Endpoint, cfg) {
+			c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "提供商或 Endpoint 已变化，请重新填写 API Key"})
+			return
+		}
 		req.APIKey = cfg.APIKey
 	}
 	cfg.Provider = req.Provider
@@ -137,12 +151,18 @@ func resolveEndpoint(provider, endpoint string) string {
 		return "https://open.bigmodel.cn/api/paas/v4"
 	case "qwen":
 		return "https://dashscope.aliyuncs.com/compatible-mode/v1"
+	case "hunyuan": // 腾讯混元 OpenAI 兼容端点 // Logan.Li
+		return "https://tokenhub.tencentmaas.com/v1"
+	case "volcengine": // 火山引擎方舟 OpenAI 兼容端点 // Logan.Li
+		return "https://ark.cn-beijing.volces.com/api/v3"
+	case "baidu": // 百度千帆 v2 OpenAI 兼容端点 // Logan.Li
+		return "https://qianfan.baidubce.com/v2"
 	case "moonshot":
 		return "https://api.moonshot.cn/v1"
 	case "kimi": // Kimi 编码套餐 OpenAI 兼容端点（注意：该端点有客户端白名单，通用调用可能被拒） // Logan.Li
 		return "https://api.kimi.com/coding/v1"
 	case "mimo": // 小米 MiMo OpenAI 兼容端点 // Logan.Li
-		return "https://api.mimo.mi.com/v1"
+		return "https://api.xiaomimimo.com/v1"
 	case "anthropic": // Anthropic 官方 Messages 协议（anthropic.go 客户端拼 /v1/messages） // Logan.Li
 		return "https://api.anthropic.com"
 	case "glm-coding": // 智谱 GLM 编码套餐 Anthropic 协议端点（模型 GLM-5.2） // Logan.Li
@@ -153,20 +173,43 @@ func resolveEndpoint(provider, endpoint string) string {
 		return "https://api.minimaxi.com/anthropic"
 	case "openai": // OpenAI 官方端点 // Logan.Li
 		return "https://api.openai.com/v1"
+	case "google": // Gemini OpenAI 兼容端点 // Logan.Li
+		return "https://generativelanguage.googleapis.com/v1beta/openai"
+	case "xai": // xAI OpenAI 兼容端点 // Logan.Li
+		return "https://api.x.ai/v1"
+	case "mistral": // Mistral OpenAI 兼容端点 // Logan.Li
+		return "https://api.mistral.ai/v1"
+	case "groq": // Groq OpenAI 兼容端点 // Logan.Li
+		return "https://api.groq.com/openai/v1"
+	case "openrouter": // OpenRouter OpenAI 兼容端点 // Logan.Li
+		return "https://openrouter.ai/api/v1"
+	case "together": // Together AI OpenAI 兼容端点 // Logan.Li
+		return "https://api.together.ai/v1"
+	case "cerebras": // Cerebras OpenAI 兼容端点 // Logan.Li
+		return "https://api.cerebras.ai/v1"
 	case "siliconflow": // 硅基流动 OpenAI 兼容端点 // Logan.Li
 		return "https://api.siliconflow.cn/v1"
 	case "ollama":
 		return "http://localhost:11434/v1"
 	case "vllm": // vLLM OpenAI 兼容端点 // Logan.Li
 		return "http://localhost:8000/v1"
-	default: // openai / custom
+	case "lmstudio": // LM Studio OpenAI 兼容端点 // Logan.Li
+		return "http://localhost:1234/v1"
+	case "new-api", "custom": // 网关地址必须由用户明确填写 // Logan.Li
+		return ""
+	default: // 未知的历史供应商保持 OpenAI 默认兼容 // Logan.Li
 		return "https://api.openai.com/v1"
 	}
 }
 
 // isLocalLLMProvider 本地/自建模型，API Key 可选 // Logan.Li
 func isLocalLLMProvider(provider string) bool {
-	return provider == "ollama" || provider == "vllm" || provider == "custom"
+	return provider == "ollama" || provider == "vllm" || provider == "lmstudio" || provider == "custom"
+}
+
+// requiresExplicitEndpoint 判断是否必须手动填写网关地址。 // Logan.Li
+func requiresExplicitEndpoint(provider string) bool {
+	return provider == "new-api" || provider == "custom"
 }
 
 // resolveProtocol 根据提供商推断接入协议（决定走 OpenAI 还是 Anthropic 客户端） // Logan.Li
diff --git a/backend/controller/llm_chat.go b/backend/controller/llm_chat.go
index 98612b1..cf05bdb 100644
--- a/backend/controller/llm_chat.go
+++ b/backend/controller/llm_chat.go
@@ -195,10 +195,11 @@ func (h *LLMHandler) llmConfigFromTestRequest(req model.SaveLLMConfigRequest) (m
 	}
 	if strings.Contains(req.APIKey, "****") {
 		var saved model.LLMConfig
-		if err := h.db.First(&saved).Error; err != nil {
+		if err := h.db.Take(&saved).Error; err != nil || !canReuseMaskedAPIKey(req.Provider, req.Endpoint, saved) {
 			if !isLocalLLMProvider(req.Provider) {
-				return cfg, "请填写 API Key"
+				return cfg, "提供商或 Endpoint 已变化，请重新填写 API Key"
 			}
+			cfg.APIKey = ""
 		} else {
 			cfg.APIKey = saved.APIKey
 		}
@@ -209,6 +210,9 @@ func (h *LLMHandler) llmConfigFromTestRequest(req model.SaveLLMConfigRequest) (m
 	if cfg.Model == "" {
 		return cfg, "请填写模型名"
 	}
+	if requiresExplicitEndpoint(cfg.Provider) && strings.TrimSpace(cfg.Endpoint) == "" {
+		return cfg, "请填写 Endpoint"
+	}
 	if cfg.APIKey == "" && !isLocalLLMProvider(cfg.Provider) {
 		return cfg, "请填写 API Key"
 	}
@@ -226,7 +230,7 @@ func (h *LLMHandler) TestLLMConnection(c *gin.Context) {
 	if req.Provider != "" {
 		cfg, errMsg = h.llmConfigFromTestRequest(req)
 	} else {
-		if err := h.db.First(&cfg).Error; err != nil {
+		if err := h.db.Take(&cfg).Error; err != nil {
 			c.JSON(http.StatusOK, model.Response{Code: 400, Message: "请先填写配置并测试连接"})
 			return
 		}
diff --git a/backend/controller/llm_models.go b/backend/controller/llm_models.go
new file mode 100644
index 0000000..387b3d8
--- /dev/null
+++ b/backend/controller/llm_models.go
@@ -0,0 +1,181 @@
+package controller
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"sort"
+	"strings"
+	"time"
+
+	"ezops/model"
+
+	"github.com/gin-gonic/gin"
+)
+
+const maxLLMModelListBytes = 2 << 20
+
+type llmModelListItem struct {
+	ID    string `json:"id"`
+	Name  string `json:"name"`
+	Model string `json:"model"`
+}
+
+// resolveModelListCredential 解析草稿配置，并在前端回传脱敏值时复用已保存密钥。 // Logan.Li
+func (h *LLMHandler) resolveModelListCredential(req model.SaveLLMConfigRequest) (string, string, string) {
+	provider := strings.TrimSpace(req.Provider)
+	if provider == "" {
+		return "", "", "请选择提供商"
+	}
+	endpoint := resolveEndpoint(provider, strings.TrimSpace(req.Endpoint))
+	if endpoint == "" {
+		return "", "", "请填写 Endpoint"
+	}
+	apiKey := strings.TrimSpace(req.APIKey)
+	if strings.Contains(apiKey, "****") {
+		var saved model.LLMConfig
+		if err := h.db.Take(&saved).Error; err != nil || !canReuseMaskedAPIKey(provider, req.Endpoint, saved) {
+			return "", "", "提供商或 Endpoint 已变化，请重新填写 API Key"
+		}
+		apiKey = saved.APIKey
+	}
+	if apiKey == "" && !isLocalLLMProvider(provider) {
+		return "", "", "请填写 API Key"
+	}
+	return endpoint, apiKey, ""
+}
+
+// buildLLMModelListURL 将基础地址转换为兼容协议的模型列表地址。 // Logan.Li
+func buildLLMModelListURL(provider, endpoint string) (string, error) {
+	base := strings.TrimRight(strings.TrimSpace(endpoint), "/")
+	if resolveProtocol(provider) == "anthropic" && !strings.HasSuffix(base, "/v1") {
+		base += "/v1"
+	}
+	parsed, err := url.Parse(base + "/models")
+	if err != nil || parsed.Host == "" || (parsed.Scheme != "http" && parsed.Scheme != "https") {
+		return "", fmt.Errorf("Endpoint 必须是有效的 HTTP 或 HTTPS 地址")
+	}
+	return parsed.String(), nil
+}
+
+// decodeLLMModelItems 兼容 OpenAI/New API 的 data 数组及部分本地服务的 models 数组。 // Logan.Li
+func decodeLLMModelItems(raw json.RawMessage) []string {
+	var items []llmModelListItem
+	if err := json.Unmarshal(raw, &items); err == nil {
+		models := make([]string, 0, len(items))
+		for _, item := range items {
+			name := strings.TrimSpace(item.ID)
+			if name == "" {
+				name = strings.TrimSpace(item.Name)
+			}
+			if name == "" {
+				name = strings.TrimSpace(item.Model)
+			}
+			if name != "" {
+				models = append(models, name)
+			}
+		}
+		return models
+	}
+	var names []string
+	_ = json.Unmarshal(raw, &names)
+	return names
+}
+
+// parseLLMModelList 提取、去重并排序上游返回的模型 ID。 // Logan.Li
+func parseLLMModelList(raw []byte) ([]string, error) {
+	var payload map[string]json.RawMessage
+	if err := json.Unmarshal(raw, &payload); err != nil {
+		return nil, fmt.Errorf("解析模型列表失败: %w", err)
+	}
+	models := decodeLLMModelItems(payload["data"])
+	models = append(models, decodeLLMModelItems(payload["models"])...)
+	unique := make(map[string]struct{}, len(models))
+	result := make([]string, 0, len(models))
+	for _, item := range models {
+		name := strings.TrimSpace(item)
+		if name == "" {
+			continue
+		}
+		if _, exists := unique[name]; exists {
+			continue
+		}
+		unique[name] = struct{}{}
+		result = append(result, name)
+	}
+	sort.Strings(result)
+	return result, nil
+}
+
+// requestLLMModelList 请求上游模型目录，禁止跨域重定向以免凭证被转发。 // Logan.Li
+func requestLLMModelList(ctx context.Context, provider, endpoint, apiKey string) ([]string, error) {
+	requestURL, err := buildLLMModelListURL(provider, endpoint)
+	if err != nil {
+		return nil, err
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, requestURL, nil)
+	if err != nil {
+		return nil, fmt.Errorf("构造模型列表请求失败: %w", err)
+	}
+	req.Header.Set("Accept", "application/json")
+	if apiKey != "" {
+		req.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+	if resolveProtocol(provider) == "anthropic" {
+		req.Header.Set("x-api-key", apiKey)
+		req.Header.Set("anthropic-version", "2023-06-01")
+	}
+	client := &http.Client{
+		Timeout: 20 * time.Second,
+		CheckRedirect: func(_ *http.Request, _ []*http.Request) error {
+			return http.ErrUseLastResponse
+		},
+	}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("读取模型列表失败: %w", err)
+	}
+	defer resp.Body.Close()
+	raw, err := io.ReadAll(io.LimitReader(resp.Body, maxLLMModelListBytes+1))
+	if err != nil {
+		return nil, fmt.Errorf("读取模型列表响应失败: %w", err)
+	}
+	if len(raw) > maxLLMModelListBytes {
+		return nil, fmt.Errorf("模型列表响应超过 2 MiB 限制")
+	}
+	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
+		message := []rune(strings.TrimSpace(string(raw)))
+		if len(message) > 1000 {
+			message = message[:1000]
+		}
+		return nil, fmt.Errorf("模型服务返回 HTTP %d: %s", resp.StatusCode, string(message))
+	}
+	return parseLLMModelList(raw)
+}
+
+// ListLLMModels 从当前草稿 Endpoint 读取可用模型，便于 New API/One API 和本地服务配置。 // Logan.Li
+// POST /api/v1/aiops/llm-config/models // Logan.Li
+func (h *LLMHandler) ListLLMModels(c *gin.Context) {
+	var req model.SaveLLMConfigRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, model.Response{Code: 400, Message: "请求参数错误: " + err.Error()})
+		return
+	}
+	endpoint, apiKey, errMsg := h.resolveModelListCredential(req)
+	if errMsg != "" {
+		c.JSON(http.StatusOK, model.Response{Code: 400, Message: errMsg})
+		return
+	}
+	models, err := requestLLMModelList(c.Request.Context(), req.Provider, endpoint, apiKey)
+	if err != nil {
+		c.JSON(http.StatusOK, model.Response{Code: 502, Message: err.Error()})
+		return
+	}
+	c.JSON(http.StatusOK, model.Response{Code: 200, Data: gin.H{
+		"models": models,
+		"total":  len(models),
+	}})
+}
diff --git a/backend/controller/llm_models_test.go b/backend/controller/llm_models_test.go
new file mode 100644
index 0000000..ec67eaa
--- /dev/null
+++ b/backend/controller/llm_models_test.go
@@ -0,0 +1,96 @@
+package controller
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"reflect"
+	"testing"
+
+	"ezops/model"
+
+	"github.com/glebarez/sqlite"
+	"gorm.io/gorm"
+)
+
+func TestParseLLMModelList(t *testing.T) {
+	raw := []byte(`{"data":[{"id":"model-b"},{"id":"model-a"},{"id":"model-b"}],"models":[{"name":"model-c"}]}`)
+	got, err := parseLLMModelList(raw)
+	if err != nil {
+		t.Fatalf("parseLLMModelList() error = %v", err)
+	}
+	want := []string{"model-a", "model-b", "model-c"}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("parseLLMModelList() = %#v, want %#v", got, want)
+	}
+}
+
+func TestRequestLLMModelListForOpenAICompatibleGateway(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			t.Errorf("request path = %q, want /v1/models", r.URL.Path)
+		}
+		if got := r.Header.Get("Authorization"); got != "Bearer test-key" {
+			t.Errorf("Authorization = %q", got)
+		}
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"data":[{"id":"gpt-test"}]}`))
+	}))
+	defer server.Close()
+
+	models, err := requestLLMModelList(context.Background(), "new-api", server.URL+"/v1", "test-key")
+	if err != nil {
+		t.Fatalf("requestLLMModelList() error = %v", err)
+	}
+	if !reflect.DeepEqual(models, []string{"gpt-test"}) {
+		t.Fatalf("requestLLMModelList() = %#v", models)
+	}
+}
+
+func TestRequestLLMModelListForAnthropic(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			t.Errorf("request path = %q, want /v1/models", r.URL.Path)
+		}
+		if got := r.Header.Get("x-api-key"); got != "anthropic-key" {
+			t.Errorf("x-api-key = %q", got)
+		}
+		if got := r.Header.Get("anthropic-version"); got != "2023-06-01" {
+			t.Errorf("anthropic-version = %q", got)
+		}
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"data":[{"id":"claude-test"}]}`))
+	}))
+	defer server.Close()
+
+	models, err := requestLLMModelList(context.Background(), "anthropic", server.URL, "anthropic-key")
+	if err != nil {
+		t.Fatalf("requestLLMModelList() error = %v", err)
+	}
+	if !reflect.DeepEqual(models, []string{"claude-test"}) {
+		t.Fatalf("requestLLMModelList() = %#v", models)
+	}
+}
+
+func TestResolveModelListCredentialDoesNotReuseKeyAcrossProviders(t *testing.T) {
+	db, err := gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	if err != nil {
+		t.Fatalf("open database: %v", err)
+	}
+	if err := db.AutoMigrate(&model.LLMConfig{}); err != nil {
+		t.Fatalf("migrate database: %v", err)
+	}
+	if err := db.Create(&model.LLMConfig{Provider: "deepseek", APIKey: "deepseek-secret"}).Error; err != nil {
+		t.Fatalf("create config: %v", err)
+	}
+
+	handler := NewLLMHandler(db)
+	_, _, errMsg := handler.resolveModelListCredential(model.SaveLLMConfigRequest{
+		Provider: "new-api",
+		Endpoint: "https://new-api.example.com/v1",
+		APIKey:   "****cret",
+	})
+	if errMsg != "提供商或 Endpoint 已变化，请重新填写 API Key" {
+		t.Fatalf("provider mismatch error = %q", errMsg)
+	}
+}
diff --git a/backend/controller/llm_provider_test.go b/backend/controller/llm_provider_test.go
new file mode 100644
index 0000000..931fa21
--- /dev/null
+++ b/backend/controller/llm_provider_test.go
@@ -0,0 +1,70 @@
+package controller
+
+import (
+	"testing"
+
+	"ezops/model"
+)
+
+func TestResolveEndpoint(t *testing.T) {
+	tests := []struct {
+		provider string
+		endpoint string
+		want     string
+	}{
+		{provider: "google", want: "https://generativelanguage.googleapis.com/v1beta/openai"},
+		{provider: "openrouter", want: "https://openrouter.ai/api/v1"},
+		{provider: "hunyuan", want: "https://tokenhub.tencentmaas.com/v1"},
+		{provider: "mimo", want: "https://api.xiaomimimo.com/v1"},
+		{provider: "together", want: "https://api.together.ai/v1"},
+		{provider: "lmstudio", want: "http://localhost:1234/v1"},
+		{provider: "new-api", want: ""},
+		{provider: "new-api", endpoint: "https://llm.example.com/v1/", want: "https://llm.example.com/v1"},
+	}
+
+	for _, test := range tests {
+		t.Run(test.provider+test.endpoint, func(t *testing.T) {
+			if got := resolveEndpoint(test.provider, test.endpoint); got != test.want {
+				t.Fatalf("resolveEndpoint() = %q, want %q", got, test.want)
+			}
+		})
+	}
+}
+
+func TestResolveProtocol(t *testing.T) {
+	for _, provider := range []string{"anthropic", "glm-coding", "minimax-coding"} {
+		if got := resolveProtocol(provider); got != "anthropic" {
+			t.Fatalf("resolveProtocol(%q) = %q, want anthropic", provider, got)
+		}
+	}
+	for _, provider := range []string{"openai", "google", "openrouter", "new-api"} {
+		if got := resolveProtocol(provider); got != "openai" {
+			t.Fatalf("resolveProtocol(%q) = %q, want openai", provider, got)
+		}
+	}
+}
+
+func TestProviderRequirements(t *testing.T) {
+	if !isLocalLLMProvider("lmstudio") {
+		t.Fatal("LM Studio should allow an empty API key")
+	}
+	if isLocalLLMProvider("new-api") {
+		t.Fatal("New API should require an API key")
+	}
+	if !requiresExplicitEndpoint("new-api") || !requiresExplicitEndpoint("custom") {
+		t.Fatal("API gateway providers should require an explicit endpoint")
+	}
+}
+
+func TestCanReuseMaskedAPIKey(t *testing.T) {
+	saved := model.LLMConfig{ID: 1, Provider: "openai", Endpoint: "https://api.openai.com/v1"}
+	if !canReuseMaskedAPIKey("openai", "https://api.openai.com/v1/", saved) {
+		t.Fatal("same provider and endpoint should reuse the saved key")
+	}
+	if canReuseMaskedAPIKey("openai", "https://gateway.example.com/v1", saved) {
+		t.Fatal("changed endpoint must not reuse the saved key")
+	}
+	if canReuseMaskedAPIKey("new-api", saved.Endpoint, saved) {
+		t.Fatal("changed provider must not reuse the saved key")
+	}
+}
diff --git a/backend/model/llm.go b/backend/model/llm.go
index 7efd4ea..db5e867 100644
--- a/backend/model/llm.go
+++ b/backend/model/llm.go
@@ -3,18 +3,18 @@ package model
 import "time"
 
 // LLMConfig 大语言模型配置（单行配置，类似 LDAPConfig） // Logan.Li
-// 支持 OpenAI 兼容接口（OpenAI / DeepSeek / 智谱GLM / 通义千问 / Moonshot / Kimi / 小米MiMo / Anthropic(需代理) / Ollama / 自定义）
+// 支持 Anthropic Messages 协议与主流 OpenAI 兼容云服务、API 网关及本地模型。 // Logan.Li
 type LLMConfig struct {
-	ID           uint      `gorm:"primarykey" json:"id"`
-	Provider     string    `gorm:"size:50;not null;default:openai" json:"provider"`      // deepseek/glm/glm-coding/kimi/moonshot/minimax/minimax-coding/mimo/siliconflow/openai/anthropic/ollama/vllm/custom
-	Model        string    `gorm:"size:100;not null;default:gpt-4o-mini" json:"model"`    // 模型名，如 gpt-4o-mini / deepseek-v4-flash
-	APIKey       string    `gorm:"size:500" json:"api_key"`                               // API Key（GET 时会被脱敏）
-	Endpoint     string    `gorm:"size:500" json:"endpoint"`                              // 基础地址，如 https://api.openai.com/v1
-	Temperature  float64   `gorm:"default:0.7" json:"temperature"`                        // 采样温度 0-2
-	MaxTokens    int       `gorm:"default:2000" json:"max_tokens"`                        // 最大输出 token
-	SystemPrompt     string    `gorm:"type:text" json:"system_prompt"`                        // 系统提示词（AIOps 运维助手角色）
-	Enabled          bool      `gorm:"default:false" json:"enabled"`                          // 是否启用
-	ReasoningEnabled bool      `gorm:"default:false" json:"reasoning_enabled"`                // 是否启用推理/Plan 模式（OpenAI o-series / DeepSeek R1 / Claude thinking 等） // Logan.Li
+	ID               uint      `gorm:"primarykey" json:"id"`
+	Provider         string    `gorm:"size:50;not null;default:openai" json:"provider"`    // 供应商标识，包含云服务、网关与本地模型 // Logan.Li
+	Model            string    `gorm:"size:100;not null;default:gpt-4o-mini" json:"model"` // 模型名，如 gpt-4o-mini / deepseek-v4-flash
+	APIKey           string    `gorm:"size:500" json:"api_key"`                            // API Key（GET 时会被脱敏）
+	Endpoint         string    `gorm:"size:500" json:"endpoint"`                           // 基础地址，如 https://api.openai.com/v1
+	Temperature      float64   `gorm:"default:0.7" json:"temperature"`                     // 采样温度 0-2
+	MaxTokens        int       `gorm:"default:2000" json:"max_tokens"`                     // 最大输出 token
+	SystemPrompt     string    `gorm:"type:text" json:"system_prompt"`                     // 系统提示词（AIOps 运维助手角色）
+	Enabled          bool      `gorm:"default:false" json:"enabled"`                       // 是否启用
+	ReasoningEnabled bool      `gorm:"default:false" json:"reasoning_enabled"`             // 是否启用推理/Plan 模式（OpenAI o-series / DeepSeek R1 / Claude thinking 等） // Logan.Li
 	CreatedAt        time.Time `json:"created_at"`
 	UpdatedAt        time.Time `json:"updated_at"`
 }
@@ -26,7 +26,7 @@ func (LLMConfig) TableName() string { return "ai_llm_configs" }
 type SaveLLMConfigRequest struct {
 	Provider         string  `json:"provider"`
 	Model            string  `json:"model"`
-	APIKey           string  `json:"api_key"`        // 若为脱敏值（含 ****）则后端保留原值
+	APIKey           string  `json:"api_key"` // 若为脱敏值（含 ****）则后端保留原值
 	Endpoint         string  `json:"endpoint"`
 	Temperature      float64 `json:"temperature"`
 	MaxTokens        int     `json:"max_tokens"`
@@ -37,7 +37,7 @@ type SaveLLMConfigRequest struct {
 
 // ChatMessage 对话消息 // Logan.Li
 type ChatMessage struct {
-	Role    string `json:"role"`    // system / user / assistant
+	Role    string `json:"role"` // system / user / assistant
 	Content string `json:"content"`
 }
 
@@ -58,11 +58,11 @@ type ChatResponse struct {
 // 仅持久化成功完成的对话（用户消息 + 助手回复成对写入），错误气泡为瞬时 UI 不入库
 type ChatHistory struct {
 	ID        uint      `gorm:"primarykey" json:"id"`
-	UserID    uint      `gorm:"index;not null" json:"user_id"`         // 所属用户（JWT 中间件注入）
+	UserID    uint      `gorm:"index;not null" json:"user_id"`                       // 所属用户（JWT 中间件注入）
 	SessionID string    `gorm:"size:64;index;not null;default:''" json:"session_id"` // 对话会话 ID（前端 UUID）
-	Role      string    `gorm:"size:16;not null" json:"role"`          // user / assistant
-	Content   string    `gorm:"type:text;not null" json:"content"`     // 消息内容
-	Model     string    `gorm:"size:100" json:"model"`                 // 生成回复所用模型（仅 assistant 行有意义）
+	Role      string    `gorm:"size:16;not null" json:"role"`                        // user / assistant
+	Content   string    `gorm:"type:text;not null" json:"content"`                   // 消息内容
+	Model     string    `gorm:"size:100" json:"model"`                               // 生成回复所用模型（仅 assistant 行有意义）
 	CreatedAt time.Time `json:"created_at"`
 }
 
diff --git a/backend/pkg/agentcli/llm/anthropic.go b/backend/pkg/agentcli/llm/anthropic.go
index 3bf2e91..72c9bde 100644
--- a/backend/pkg/agentcli/llm/anthropic.go
+++ b/backend/pkg/agentcli/llm/anthropic.go
@@ -1,7 +1,7 @@
 // anthropic.go - Anthropic Messages 协议 LLM 客户端实现 // Logan.Li
 // 支持官方 Claude（api.anthropic.com）及智谱 GLM 编码套餐（open.bigmodel.cn/api/anthropic）等
 // 所有 Anthropic /v1/messages 兼容端点。与 openai.go 实现同一 Client 接口，由 client.go 按 Protocol 路由。
-// 鉴权统一用 Authorization: Bearer（智谱编码套餐实测通过；官方 Claude 较新版本亦支持）+ anthropic-version。
+// 同时发送 x-api-key 与 Bearer，兼容 Anthropic 官方和国内编码套餐网关。 // Logan.Li
 package llm
 
 import (
@@ -167,7 +167,10 @@ func buildTools(tools []Tool) []anthropicTool {
 // isAnthropicThinkingModel 判断是否为支持 thinking 的 Claude 模型 // Logan.Li
 func isAnthropicThinkingModel(model string) bool {
 	m := strings.ToLower(model)
-	return strings.Contains(m, "claude-3-7") || strings.Contains(m, "claude-3.7")
+	return strings.Contains(m, "claude-3-7") || strings.Contains(m, "claude-3.7") ||
+		strings.Contains(m, "claude-sonnet-4") || strings.Contains(m, "claude-opus-4") ||
+		strings.Contains(m, "claude-sonnet-5") || strings.Contains(m, "claude-opus-5") ||
+		strings.Contains(m, "claude-fable-5")
 }
 
 // buildRequest 构造 Anthropic 请求体（Complete 与 Stream 共用） // Logan.Li
@@ -209,6 +212,7 @@ func (c *AnthropicClient) newHTTPRequest(ctx context.Context, body []byte) (*htt
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
 	httpReq.Header.Set("Authorization", "Bearer "+c.cfg.APIKey)
+	httpReq.Header.Set("x-api-key", c.cfg.APIKey)
 	httpReq.Header.Set("anthropic-version", "2023-06-01")
 	return httpReq, nil
 }
@@ -356,8 +360,8 @@ func (c *AnthropicClient) Stream(ctx context.Context, req *CompletionRequest, on
 		}
 	}
 	return &CompletionResponse{
-		Model:  c.cfg.Model,
+		Model:   c.cfg.Model,
 		Message: Message{Role: "assistant", Content: contentBuilder.String()},
-		Usage:  usage,
+		Usage:   usage,
 	}, nil
 }
diff --git a/backend/pkg/agentcli/llm/openai.go b/backend/pkg/agentcli/llm/openai.go
index 1757a22..3f79042 100644
--- a/backend/pkg/agentcli/llm/openai.go
+++ b/backend/pkg/agentcli/llm/openai.go
@@ -49,10 +49,14 @@ func NewOpenAIClient(cfg Config) Client {
 // Name 返回客户端名称 // Logan.Li
 func (c *OpenAIClient) Name() string { return "openai-compatible(" + c.cfg.Model + ")" }
 
-// isOpenAIReasoningModel 判断是否为需要 reasoning_effort 的 OpenAI o-series 模型 // Logan.Li
+// isOpenAIReasoningModel 判断是否为使用新推理参数的 OpenAI 模型。 // Logan.Li
 func isOpenAIReasoningModel(model string) bool {
 	m := strings.ToLower(model)
-	return strings.Contains(m, "o1") || strings.Contains(m, "o3")
+	if index := strings.LastIndex(m, "/"); index >= 0 {
+		m = m[index+1:]
+	}
+	return strings.HasPrefix(m, "o1") || strings.HasPrefix(m, "o3") ||
+		strings.HasPrefix(m, "o4") || strings.HasPrefix(m, "gpt-5")
 }
 
 // isGenericReasoningModel 判断是否为通用推理开关模型（DeepSeek R1 等） // Logan.Li
@@ -61,33 +65,35 @@ func isGenericReasoningModel(model string) bool {
 	return strings.Contains(m, "reasoning") || strings.Contains(m, "r1") || strings.Contains(m, "deepseek-reasoner")
 }
 
-// applyReasoningParams 根据模型类型应用 reasoning 参数 // Logan.Li
+// applyReasoningParams 根据模型类型应用 token 与 reasoning 参数。 // Logan.Li
 func applyReasoningParams(payload *openaiRequest, enabled bool) {
-	if !enabled {
-		return
-	}
 	if isOpenAIReasoningModel(payload.Model) {
-		// OpenAI o-series 使用 reasoning_effort，temperature 必须固定为 1（不传） // Logan.Li
-		payload.ReasoningEffort = "medium"
+		// 新版 OpenAI 推理模型使用 max_completion_tokens 且不接收 temperature。 // Logan.Li
+		payload.MaxCompletionTokens = payload.MaxTokens
+		payload.MaxTokens = 0
 		payload.Temperature = 0
+		if enabled {
+			payload.ReasoningEffort = "medium"
+		}
 		return
 	}
-	if isGenericReasoningModel(payload.Model) {
+	if enabled && isGenericReasoningModel(payload.Model) {
 		payload.Reasoning = true
 	}
 }
 
 // openaiRequest OpenAI chat/completions 请求体 // Logan.Li
 type openaiRequest struct {
-	Model           string    `json:"model"`
-	Messages        []Message `json:"messages"`
-	Tools           []Tool    `json:"tools,omitempty"`
-	ToolChoice      string    `json:"tool_choice,omitempty"` // "auto" 让模型自主决定
-	Temperature     float64   `json:"temperature,omitempty"`
-	MaxTokens       int       `json:"max_tokens,omitempty"`
-	Stream          bool      `json:"stream,omitempty"`          // 流式补全（Stream 方法置 true）
-	ReasoningEffort string    `json:"reasoning_effort,omitempty"` // OpenAI o-series 推理强度 // Logan.Li
-	Reasoning       bool      `json:"reasoning,omitempty"`        // DeepSeek 等部分平台推理开关 // Logan.Li
+	Model               string    `json:"model"`
+	Messages            []Message `json:"messages"`
+	Tools               []Tool    `json:"tools,omitempty"`
+	ToolChoice          string    `json:"tool_choice,omitempty"` // "auto" 让模型自主决定
+	Temperature         float64   `json:"temperature,omitempty"`
+	MaxTokens           int       `json:"max_tokens,omitempty"`
+	MaxCompletionTokens int       `json:"max_completion_tokens,omitempty"` // 新版 OpenAI 推理模型的输出上限 // Logan.Li
+	Stream              bool      `json:"stream,omitempty"`                // 流式补全（Stream 方法置 true）
+	ReasoningEffort     string    `json:"reasoning_effort,omitempty"`      // OpenAI o-series 推理强度 // Logan.Li
+	Reasoning           bool      `json:"reasoning,omitempty"`             // DeepSeek 等部分平台推理开关 // Logan.Li
 }
 
 // openaiResponse OpenAI chat/completions 响应体 // Logan.Li
@@ -245,9 +251,9 @@ func (c *OpenAIClient) Stream(ctx context.Context, req *CompletionRequest, onDel
 		}
 	}
 	return &CompletionResponse{
-		Model:  c.cfg.Model,
+		Model:   c.cfg.Model,
 		Message: Message{Role: "assistant", Content: contentBuilder.String()},
-		Usage:  usage,
+		Usage:   usage,
 	}, nil
 }
 
diff --git a/backend/pkg/agentcli/llm/provider_compat_test.go b/backend/pkg/agentcli/llm/provider_compat_test.go
new file mode 100644
index 0000000..f69dc57
--- /dev/null
+++ b/backend/pkg/agentcli/llm/provider_compat_test.go
@@ -0,0 +1,39 @@
+package llm
+
+import "testing"
+
+func TestApplyReasoningParamsForModernOpenAIModels(t *testing.T) {
+	payload := openaiRequest{
+		Model:       "openai/gpt-5.4",
+		Temperature: 0.7,
+		MaxTokens:   8192,
+	}
+	applyReasoningParams(&payload, true)
+
+	if payload.Temperature != 0 || payload.MaxTokens != 0 {
+		t.Fatalf("unsupported parameters were not cleared: %+v", payload)
+	}
+	if payload.MaxCompletionTokens != 8192 || payload.ReasoningEffort != "medium" {
+		t.Fatalf("modern OpenAI parameters were not applied: %+v", payload)
+	}
+}
+
+func TestApplyReasoningParamsWithoutReasoningEffort(t *testing.T) {
+	payload := openaiRequest{Model: "gpt-5.4-mini", Temperature: 0.7, MaxTokens: 4096}
+	applyReasoningParams(&payload, false)
+
+	if payload.MaxCompletionTokens != 4096 || payload.ReasoningEffort != "" {
+		t.Fatalf("model compatibility parameters are incorrect: %+v", payload)
+	}
+}
+
+func TestAnthropicThinkingModels(t *testing.T) {
+	for _, model := range []string{"claude-3-7-sonnet", "claude-sonnet-4-6", "claude-opus-4-8", "claude-sonnet-5", "claude-fable-5"} {
+		if !isAnthropicThinkingModel(model) {
+			t.Fatalf("%q should support thinking parameters", model)
+		}
+	}
+	if isAnthropicThinkingModel("claude-haiku-4-5") {
+		t.Fatal("Claude Haiku should not receive thinking parameters")
+	}
+}
diff --git a/backend/router/routes.go b/backend/router/routes.go
index 52c9fdc..4f3d7d3 100644
--- a/backend/router/routes.go
+++ b/backend/router/routes.go
@@ -942,6 +942,7 @@ func Register(r *gin.Engine, db *gorm.DB, log *zap.Logger, cfg *config.Config, c
 		aiopsGroup.GET("/llm-config", llmHandler.GetLLMConfig)
 		aiopsGroup.PUT("/llm-config", llmHandler.SaveLLMConfig)
 		aiopsGroup.POST("/llm-config/test", llmHandler.TestLLMConnection)
+		aiopsGroup.POST("/llm-config/models", llmHandler.ListLLMModels) // 读取 OpenAI/New API 兼容模型列表 // Logan.Li
 		aiopsGroup.POST("/chat", llmHandler.Chat)
 		aiopsGroup.POST("/chat/stream", llmHandler.ChatStream)
 		aiopsGroup.GET("/chat-history", llmHandler.GetChatHistory)
diff --git a/frontend/src/api/aiops/assistant.ts b/frontend/src/api/aiops/assistant.ts
index b54f242..f1fc877 100644
--- a/frontend/src/api/aiops/assistant.ts
+++ b/frontend/src/api/aiops/assistant.ts
@@ -33,6 +33,11 @@ export interface LLMTestResult {
   reasoning_enabled: boolean;
 }
 
+export interface LLMModelListResult {
+  models: string[];
+  total: number;
+}
+
 export interface ChatHistoryItem {
   id: number;
   user_id: number;
@@ -192,6 +197,9 @@ export const aiApi = {
   // 测试 LLM 连接（可传草稿配置，保存前即可验证） // Logan.Li
   testConnection: (data?: Partial<LLMConfigData>) =>
     http.post<LLMTestResult>('/aiops/llm-config/test', data ?? {}),
+  // 从 OpenAI/Anthropic 兼容端点读取当前密钥可用的模型。 // Logan.Li
+  listModels: (data: Partial<LLMConfigData>) =>
+    http.post<LLMModelListResult>('/aiops/llm-config/models', data),
   // AI 对话（messages 含历史 + 当前用户消息，session_id 区分多轮对话）
   chat: (messages: { role: 'user' | 'assistant'; content: string }[], sessionId?: string) =>
     http.post<ChatResult>('/aiops/chat', { messages, session_id: sessionId || '' }),
diff --git a/frontend/src/components/aiops/LLMConfig.tsx b/frontend/src/components/aiops/LLMConfig.tsx
index de8452f..62335ee 100644
--- a/frontend/src/components/aiops/LLMConfig.tsx
+++ b/frontend/src/components/aiops/LLMConfig.tsx
@@ -12,153 +12,25 @@ import {
 import { toast } from 'sonner';
 import {
   Loader2, Save, Wifi, KeyRound, Brain, Sparkles, MessageSquare, AlertCircle,
-  ExternalLink, CheckCircle2, ShoppingBag,
+  ExternalLink, CheckCircle2, ShoppingBag, RefreshCw,
 } from 'lucide-react';
 import { aiApi, type LLMConfigData, type LLMTestResult } from '@/api';
 import { PageShell } from '@/components/layout/PageShell';
 import { useTranslation } from 'react-i18next';
-import i18n from '@/i18n';
-
-// 单个提供商配置：默认 endpoint、候选模型、套餐/充值引导链接 // Logan.Li
-interface ProviderMeta {
-  id: string;
-  name: string;
-  endpoint: string;
-  models: string[];
-  apiKeyOptional?: boolean; // 本地/自建模型 API Key 可留空 // Logan.Li
-  planUrl?: string;
-  planLabel?: string;
-  planHint?: string;
-}
+import {
+  CLOUD_CODING_PROVIDERS,
+  CLOUD_GENERAL_PROVIDERS,
+  endpointPlaceholder,
+  findProvider,
+  GATEWAY_PROVIDERS,
+  LOCAL_PROVIDERS,
+  needsManualModel,
+} from './llmProviders';
 
 // 连接凭证快照（仅含影响连通的字段，切换后需重新验证） // Logan.Li
 const connectionSnapshot = (c: LLMConfigData) =>
   `${c.provider}|${c.model}|${c.endpoint}|${c.api_key}`;
 
-// 云服务（按量 / 通用 API） // Logan.Li
-const CLOUD_GENERAL_PROVIDERS: ProviderMeta[] = [
-  {
-    id: 'deepseek', name: 'DeepSeek', endpoint: 'https://api.deepseek.com/v1',
-    models: ['deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-reasoner', 'deepseek-chat'],
-    planUrl: 'https://platform.deepseek.com',
-    planLabel: i18n.t('aiops.getApiKey'),
-    planHint: i18n.t('aiops.deepseekHint'),
-  },
-  {
-    id: 'glm', name: i18n.t('aiops.zhipuGeneral'), endpoint: 'https://open.bigmodel.cn/api/paas/v4',
-    models: ['glm-4.5', 'glm-4.6', 'glm-4-flash', 'glm-4-air', 'glm-4-plus'],
-    planUrl: 'https://open.bigmodel.cn/',
-    planLabel: i18n.t('aiops.zhipuOpenPlatform'),
-    planHint: i18n.t('aiops.zhipuGetKey'),
-  },
-  {
-    id: 'moonshot', name: 'Kimi (Moonshot)', endpoint: 'https://api.moonshot.cn/v1',
-    models: ['kimi-latest', 'moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k'],
-    planUrl: 'https://www.kimi.com/membership/pricing',
-    planLabel: i18n.t('aiops.buyKimiPlan'),
-    planHint: i18n.t('aiops.kimiGeneralHint'),
-  },
-  {
-    id: 'minimax', name: i18n.t('aiops.minimaxPayg'), endpoint: 'https://api.minimaxi.com/v1',
-    models: ['MiniMax-M3', 'MiniMax-M2.7', 'MiniMax-M2.7-highspeed', 'MiniMax-M2.5', 'MiniMax-M2.5-highspeed'],
-    planUrl: 'https://platform.minimaxi.com/user-center/basic-information/interface-key',
-    planLabel: i18n.t('aiops.getPaygApiKey'),
-    planHint: i18n.t('aiops.byTokenBilling'),
-  },
-  {
-    id: 'mimo', name: i18n.t('aiops.mimoXiaomi'), endpoint: 'https://api.mimo.mi.com/v1',
-    models: ['mimo-v2', 'mimo-v2-flash'],
-    planUrl: 'https://mimo.mi.com/',
-    planLabel: i18n.t('aiops.mimoPlatform'),
-  },
-  {
-    id: 'siliconflow', name: i18n.t('aiops.providerSiliconflow'), endpoint: 'https://api.siliconflow.cn/v1',
-    models: ['deepseek-ai/DeepSeek-V3', 'Qwen/Qwen2.5-72B-Instruct', 'Pro/deepseek-ai/DeepSeek-R1'],
-    planUrl: 'https://cloud.siliconflow.cn/',
-    planLabel: i18n.t('aiops.siliconflowConsole'),
-    planHint: i18n.t('aiops.siliconflowHint'),
-  },
-  {
-    id: 'openai', name: 'OpenAI', endpoint: 'https://api.openai.com/v1',
-    models: ['gpt-4o', 'gpt-4o-mini', 'gpt-4.1', 'gpt-4.1-mini', 'o3-mini', 'o4-mini', 'o1-mini'],
-    planUrl: 'https://platform.openai.com/api-keys',
-    planLabel: i18n.t('system.createApiKey'),
-    planHint: i18n.t('aiops.openaiHint'),
-  },
-  {
-    id: 'anthropic', name: 'Anthropic Claude', endpoint: 'https://api.anthropic.com',
-    models: ['claude-sonnet-4-6', 'claude-opus-4-8', 'claude-haiku-4-5'],
-    planUrl: 'https://console.anthropic.com/settings/billing',
-    planLabel: i18n.t('aiops.anthropicBill'),
-    planHint: i18n.t('aiops.customProxyHint'),
-  },
-];
-
-// 编码套餐 Coding Plan（固定月费，适合 Agent CLI 高频调用） // Logan.Li
-const CLOUD_CODING_PROVIDERS: ProviderMeta[] = [
-  {
-    id: 'glm-coding', name: i18n.t('aiops.zhipuCoding'), endpoint: 'https://open.bigmodel.cn/api/anthropic',
-    models: ['GLM-5.2', 'GLM-5.1'],
-    planUrl: 'https://bigmodel.cn/glm-coding',
-    planLabel: i18n.t('aiops.buyGlmCoding'),
-    planHint: i18n.t('aiops.zhipuCodingHint'),
-  },
-  {
-    id: 'kimi', name: i18n.t('aiops.kimiCoding'), endpoint: 'https://api.kimi.com/coding/v1',
-    models: ['kimi-for-coding', 'kimi-coding-latest'],
-    planUrl: 'https://www.kimi.com/membership/pricing',
-    planLabel: i18n.t('aiops.buyKimiPlan'),
-    planHint: i18n.t('aiops.kimiCodingEndpoint'),
-  },
-  {
-    id: 'minimax-coding', name: i18n.t('aiops.minimaxCoding'), endpoint: 'https://api.minimaxi.com/anthropic',
-    models: ['MiniMax-M3', 'MiniMax-M3[1m]', 'MiniMax-M2.7'],
-    planUrl: 'https://platform.minimaxi.com/user-center/payment/token-plan',
-    planLabel: i18n.t('aiops.buyTokenPack'),
-    planHint: i18n.t('aiops.anthropicSubHint'),
-  },
-];
-
-const CLOUD_PROVIDERS: ProviderMeta[] = [...CLOUD_GENERAL_PROVIDERS, ...CLOUD_CODING_PROVIDERS];
-
-// 自定义接入：手动填请求地址 + API Key + 模型名（OpenAI 兼容） // Logan.Li
-const CUSTOM_PROVIDERS: ProviderMeta[] = [
-  {
-    id: 'custom', name: i18n.t('aiops.customProvider'), endpoint: '', models: [],
-    apiKeyOptional: true,
-    planHint: i18n.t('aiops.customProviderDesc'),
-  },
-];
-
-// 本地/自建模型（API Key 可选，置于列表底部） // Logan.Li
-const LOCAL_PROVIDERS: ProviderMeta[] = [
-  {
-    id: 'ollama', name: i18n.t('aiops.ollamaLocal'), endpoint: 'http://localhost:11434/v1',
-    models: ['qwen2.5:7b', 'llama3.2', 'deepseek-r1:7b'],
-    apiKeyOptional: true,
-  },
-  {
-    id: 'vllm', name: i18n.t('aiops.vllmSelfHosted'), endpoint: 'http://localhost:8000/v1',
-    models: [],
-    apiKeyOptional: true,
-  },
-];
-
-const PROVIDERS: ProviderMeta[] = [...CLOUD_PROVIDERS, ...CUSTOM_PROVIDERS, ...LOCAL_PROVIDERS];
-
-// 模型名需手动输入的提供商（空 models 或需填接入点 ID） // Logan.Li
-const needsManualModel = (provider: string, prov?: ProviderMeta) =>
-  provider === 'custom' || provider === 'vllm' || !prov?.models.length;
-
-const endpointPlaceholder = (provider: string) => {
-  switch (provider) {
-    case 'custom': return 'https://openrouter.ai/api/v1';
-    case 'ollama': return 'http://localhost:11434/v1';
-    case 'vllm': return 'http://localhost:8000/v1';
-    default: return 'https://api.xxx.com/v1';
-  }
-};
-
 const defaultCfg: LLMConfigData = {
   provider: 'deepseek', model: 'deepseek-v4-pro', api_key: '',
   endpoint: 'https://api.deepseek.com/v1', temperature: 0.7, max_tokens: 8192,
@@ -171,20 +43,36 @@ export default function LLMConfigPage() {
   const [loading, setLoading] = useState(true);
   const [saving, setSaving] = useState(false);
   const [testing, setTesting] = useState(false);
+  const [loadingModels, setLoadingModels] = useState(false);
+  const [discoveredModels, setDiscoveredModels] = useState<string[]>([]);
   const [testResult, setTestResult] = useState<LLMTestResult | null>(null);
   const [modelMode, setModelMode] = useState<'select' | 'input'>('select');
   const [verifiedSnapshot, setVerifiedSnapshot] = useState<string | null>(null);
 
+  const prov = findProvider(cfg.provider);
   const currentSnapshot = useMemo(() => connectionSnapshot(cfg), [cfg]);
   const isVerified = verifiedSnapshot === currentSnapshot;
+  const models = useMemo(() => Array.from(new Set([
+    ...(prov?.models ?? []),
+    ...discoveredModels,
+    ...(cfg.model?.trim() ? [cfg.model.trim()] : []),
+  ])), [cfg.model, discoveredModels, prov]);
+  const manualModel = needsManualModel(cfg.provider, prov);
+  const isCustom = cfg.provider === 'custom';
+  const apiKeyOptional = prov?.apiKeyOptional ?? false;
+  const endpointRequired = !prov?.endpoint;
+  const canTest = Boolean(cfg.model?.trim())
+    && (endpointRequired ? Boolean(cfg.endpoint?.trim()) : true)
+    && (apiKeyOptional || Boolean(cfg.api_key?.trim()));
 
   const load = useCallback(async () => {
     setLoading(true);
     try {
       const saved = await aiApi.getConfig();
-      const merged = { ...defaultCfg, ...saved };
+      const merged = saved.configured ? { ...defaultCfg, ...saved } : defaultCfg;
       setCfg(merged);
-      const prov = PROVIDERS.find((x) => x.id === merged.provider);
+      setDiscoveredModels([]);
+      const prov = findProvider(merged.provider);
       const inList = prov ? prov.models.includes(merged.model) : false;
       setModelMode(needsManualModel(merged.provider, prov) || !inList ? 'input' : 'select');
       if (saved.configured) {
@@ -244,14 +132,41 @@ export default function LLMConfigPage() {
     finally { setTesting(false); }
   };
 
+  const discoverModels = async () => {
+    if (!cfg.endpoint?.trim()) {
+      toast.error(t('system.fillEndpointFirst'));
+      return;
+    }
+    if (!apiKeyOptional && !cfg.api_key?.trim()) {
+      toast.error(t('system.fillApiKeyFirst'));
+      return;
+    }
+    setLoadingModels(true);
+    try {
+      const result = await aiApi.listModels(cfg);
+      if (!result.models.length) {
+        toast.error(t('aiops.noModelsDiscovered'));
+        return;
+      }
+      setDiscoveredModels(result.models);
+      setModelMode('select');
+      if (!cfg.model?.trim()) {
+        set('model', result.models[0]);
+      }
+      toast.success(t('aiops.modelsDiscovered', { count: result.total }));
+    }
+    catch { toast.error(t('aiops.modelListFailed')); }
+    finally { setLoadingModels(false); }
+  };
+
   const onProvider = (p: string) => {
-    const next = PROVIDERS.find((x) => x.id === p);
+    const next = findProvider(p);
     if (!next) return;
     const model = next.models[0] || '';
     setModelMode(needsManualModel(p, next) ? 'input' : 'select');
+    setDiscoveredModels([]);
     invalidateVerify();
-    const endpoint = p === 'custom' ? '' : (next.endpoint || cfg.endpoint);
-    setCfg({ ...cfg, provider: p, endpoint, model });
+    setCfg({ ...cfg, provider: p, endpoint: next.endpoint, model, api_key: '' });
   };
 
   const onModel = (m: string) => {
@@ -270,16 +185,8 @@ export default function LLMConfigPage() {
     setCfg({ ...cfg, [k]: v });
   };
 
-  const prov = PROVIDERS.find((x) => x.id === cfg.provider);
   const unknownProvider = Boolean(cfg.provider && !prov);
-  const models = prov?.models ?? [];
-  const manualModel = needsManualModel(cfg.provider, prov);
-  const isCustom = cfg.provider === 'custom';
-  const apiKeyOptional = prov?.apiKeyOptional ?? false;
-  const endpointRequired = apiKeyOptional || cfg.provider === 'custom';
-  const canTest = Boolean(cfg.model?.trim())
-    && (endpointRequired ? Boolean(cfg.endpoint?.trim()) : true)
-    && (apiKeyOptional || Boolean(cfg.api_key?.trim()));
+  const showModelInput = modelMode === 'input' || (manualModel && discoveredModels.length === 0);
 
   return (
     <PageShell
@@ -294,7 +201,7 @@ export default function LLMConfigPage() {
       <Card className="dg-card">
         <CardHeader><CardTitle className="text-[var(--dg-text-primary)] text-base">{t('aiops.providerAndCredential')}</CardTitle></CardHeader>
         <CardContent className="space-y-4">
-          <div className="grid grid-cols-2 gap-4">
+          <div className="grid grid-cols-1 gap-4 sm:grid-cols-2">
             <div className="space-y-1.5">
               <Label className="text-[var(--dg-text-secondary)]">{t('aiops.providerLabel')}</Label>
               <Select value={cfg.provider} onValueChange={onProvider}>
@@ -309,33 +216,44 @@ export default function LLMConfigPage() {
                   <SelectGroup>
                     <SelectLabel>{t('aiops.cloudService')}</SelectLabel>
                     {CLOUD_GENERAL_PROVIDERS.map((p) => (
-                      <SelectItem key={p.id} value={p.id}>{p.name}</SelectItem>
+                      <SelectItem key={p.id} value={p.id}>{t(p.nameKey)}</SelectItem>
                     ))}
                   </SelectGroup>
                   <SelectGroup>
                     <SelectLabel>{t('aiops.codingPlan')}</SelectLabel>
                     {CLOUD_CODING_PROVIDERS.map((p) => (
-                      <SelectItem key={p.id} value={p.id}>{p.name}</SelectItem>
+                      <SelectItem key={p.id} value={p.id}>{t(p.nameKey)}</SelectItem>
                     ))}
                   </SelectGroup>
                   <SelectGroup>
-                    <SelectLabel>{t('aiops.customAccess')}</SelectLabel>
-                    {CUSTOM_PROVIDERS.map((p) => (
-                      <SelectItem key={p.id} value={p.id}>{p.name}</SelectItem>
+                    <SelectLabel>{t('aiops.apiGateway')}</SelectLabel>
+                    {GATEWAY_PROVIDERS.map((p) => (
+                      <SelectItem key={p.id} value={p.id}>{t(p.nameKey)}</SelectItem>
                     ))}
                   </SelectGroup>
                   <SelectGroup>
                     <SelectLabel>{t('aiops.localSelfHosted')}</SelectLabel>
                     {LOCAL_PROVIDERS.map((p) => (
-                      <SelectItem key={p.id} value={p.id}>{p.name}</SelectItem>
+                      <SelectItem key={p.id} value={p.id}>{t(p.nameKey)}</SelectItem>
                     ))}
                   </SelectGroup>
                 </SelectContent>
               </Select>
             </div>
             <div className="space-y-1.5">
-              <Label className="text-[var(--dg-text-secondary)]">{t('aiops.modelLabel')}</Label>
-              {manualModel || modelMode === 'input' ? (
+              <div className="flex min-h-5 items-center justify-between gap-2">
+                <Label className="text-[var(--dg-text-secondary)]">{t('aiops.modelLabel')}</Label>
+                <button
+                  type="button"
+                  onClick={discoverModels}
+                  disabled={loadingModels}
+                  className="inline-flex items-center gap-1 text-xs dg-text-blue hover:underline disabled:cursor-not-allowed disabled:opacity-50"
+                >
+                  <RefreshCw className={`h-3 w-3 ${loadingModels ? 'animate-spin' : ''}`} />
+                  {t('aiops.loadModels')}
+                </button>
+              </div>
+              {showModelInput ? (
                 <Input value={cfg.model} onChange={(e) => set('model', e.target.value)}
                   placeholder={isCustom ? t('aiops.modelPlaceholder') : t('aiops.manualInputModel')}
                   className="dg-input" />
@@ -365,7 +283,7 @@ export default function LLMConfigPage() {
               <ShoppingBag className="w-4 h-4 shrink-0 mt-0.5 text-[var(--dg-orange)]" />
               <div className="flex-1 min-w-0 space-y-1">
                 <div className="text-sm text-[var(--dg-text-secondary)]">
-                  {prov.planHint || t('aiops.providerNeedsKey', { name: prov.name })}
+                  {prov.planHintKey ? t(prov.planHintKey) : t('aiops.providerNeedsKey', { name: t(prov.nameKey) })}
                 </div>
                 <a
                   href={prov.planUrl}
@@ -373,17 +291,17 @@ export default function LLMConfigPage() {
                   rel="noreferrer"
                   className="inline-flex items-center gap-1 text-xs dg-text-blue hover:underline"
                 >
-                  {prov.planLabel || t('aiops.goEnable')}
+                  {prov.planLabelKey ? t(prov.planLabelKey) : t('aiops.goEnable')}
                   <ExternalLink className="w-3 h-3" />
                 </a>
               </div>
             </div>
           )}
 
-          {isCustom && prov?.planHint && (
+          {isCustom && prov?.planHintKey && (
             <div className="flex items-start gap-3 rounded-lg bg-[var(--dg-bg-tertiary)] px-3 py-2.5">
               <KeyRound className="w-4 h-4 shrink-0 mt-0.5 text-[var(--dg-blue)]" />
-              <div className="text-sm text-[var(--dg-text-secondary)]">{prov.planHint}</div>
+              <div className="text-sm text-[var(--dg-text-secondary)]">{t(prov.planHintKey)}</div>
             </div>
           )}
 
@@ -402,8 +320,8 @@ export default function LLMConfigPage() {
             <Label className="text-[var(--dg-text-secondary)]">{t('aiops.endpointLabel')}</Label>
             <Input value={cfg.endpoint} onChange={(e) => set('endpoint', e.target.value)}
               placeholder={endpointPlaceholder(cfg.provider)} className="dg-input" />
-            {prov?.planHint && !prov.planUrl && (
-              <p className="text-xs dg-text-muted">{prov.planHint}</p>
+            {prov?.planHintKey && !prov.planUrl && !isCustom && (
+              <p className="text-xs dg-text-muted">{t(prov.planHintKey)}</p>
             )}
           </div>
         </CardContent>
@@ -486,4 +404,4 @@ export default function LLMConfigPage() {
       </div>
     </PageShell>
   );
-}
\ No newline at end of file
+}
diff --git a/frontend/src/components/aiops/llmProviders.ts b/frontend/src/components/aiops/llmProviders.ts
new file mode 100644
index 0000000..0c195d5
--- /dev/null
+++ b/frontend/src/components/aiops/llmProviders.ts
@@ -0,0 +1,193 @@
+// LLM 供应商目录参考 models.dev，兼容通道参考 New API；仅收录当前客户端可调用的 Chat/Messages 服务。 // Logan.Li
+export interface ProviderMeta {
+  id: string;
+  nameKey: string;
+  endpoint: string;
+  models: string[];
+  apiKeyOptional?: boolean;
+  planUrl?: string;
+  planLabelKey?: string;
+  planHintKey?: string;
+}
+
+// 主流公有云模型服务，均使用 OpenAI 兼容协议，Anthropic 除外。 // Logan.Li
+export const CLOUD_GENERAL_PROVIDERS: ProviderMeta[] = [
+  {
+    id: 'deepseek', nameKey: 'aiops.providerDeepseek', endpoint: 'https://api.deepseek.com/v1',
+    models: ['deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-reasoner', 'deepseek-chat'],
+    planUrl: 'https://platform.deepseek.com', planLabelKey: 'aiops.getApiKey',
+    planHintKey: 'aiops.deepseekHint',
+  },
+  {
+    id: 'qwen', nameKey: 'aiops.providerQwen', endpoint: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+    models: ['qwen3.7-plus', 'qwen3.7-max', 'qwen3.6-plus', 'qwen3-coder-flash', 'qwq-plus'],
+    planUrl: 'https://bailian.console.aliyun.com/', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'glm', nameKey: 'aiops.providerZhipu', endpoint: 'https://open.bigmodel.cn/api/paas/v4',
+    models: ['glm-5.2', 'glm-5.1', 'glm-5'],
+    planUrl: 'https://open.bigmodel.cn/', planLabelKey: 'aiops.zhipuOpenPlatform',
+    planHintKey: 'aiops.zhipuGetKey',
+  },
+  {
+    id: 'moonshot', nameKey: 'aiops.providerMoonshot', endpoint: 'https://api.moonshot.cn/v1',
+    models: ['kimi-k3', 'kimi-k2.7-code', 'kimi-k2.6', 'kimi-k2.5', 'kimi-k2-thinking-turbo'],
+    planUrl: 'https://platform.moonshot.cn/console/api-keys', planLabelKey: 'aiops.getApiKey',
+    planHintKey: 'aiops.kimiGeneralHint',
+  },
+  {
+    id: 'minimax', nameKey: 'aiops.providerMinimax', endpoint: 'https://api.minimaxi.com/v1',
+    models: ['MiniMax-M3', 'MiniMax-M2.7', 'MiniMax-M2.7-highspeed', 'MiniMax-M2.5'],
+    planUrl: 'https://platform.minimaxi.com/user-center/basic-information/interface-key',
+    planLabelKey: 'aiops.getPaygApiKey', planHintKey: 'aiops.byTokenBilling',
+  },
+  {
+    id: 'mimo', nameKey: 'aiops.mimoXiaomi', endpoint: 'https://api.xiaomimimo.com/v1',
+    models: ['mimo-v2.5-pro', 'mimo-v2.5', 'mimo-v2-pro', 'mimo-v2-flash'],
+    planUrl: 'https://platform.xiaomimimo.com/', planLabelKey: 'aiops.mimoPlatform',
+  },
+  {
+    id: 'hunyuan', nameKey: 'aiops.providerTencentTokenHub', endpoint: 'https://tokenhub.tencentmaas.com/v1',
+    models: ['hy3', 'hy3-preview'],
+    planUrl: 'https://console.cloud.tencent.com/lkeap', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'volcengine', nameKey: 'aiops.providerVolcengine', endpoint: 'https://ark.cn-beijing.volces.com/api/v3',
+    models: [], planUrl: 'https://console.volcengine.com/ark', planLabelKey: 'aiops.goEnable',
+    planHintKey: 'aiops.endpointModelHint',
+  },
+  {
+    id: 'baidu', nameKey: 'aiops.providerBaidu', endpoint: 'https://qianfan.baidubce.com/v2',
+    models: [], planUrl: 'https://console.bce.baidu.com/qianfan/ais/console/applicationConsole/application',
+    planLabelKey: 'aiops.getApiKey', planHintKey: 'aiops.manualModelHint',
+  },
+  {
+    id: 'siliconflow', nameKey: 'aiops.providerSiliconflow', endpoint: 'https://api.siliconflow.cn/v1',
+    models: ['deepseek-ai/DeepSeek-V4-Pro', 'deepseek-ai/DeepSeek-V4-Flash', 'Pro/deepseek-ai/DeepSeek-R1', 'Qwen/Qwen3.5-397B-A17B'],
+    planUrl: 'https://cloud.siliconflow.cn/', planLabelKey: 'aiops.siliconflowConsole',
+    planHintKey: 'aiops.siliconflowHint',
+  },
+  {
+    id: 'openai', nameKey: 'aiops.providerOpenai', endpoint: 'https://api.openai.com/v1',
+    models: ['gpt-5.5', 'gpt-5.4', 'gpt-5.4-mini', 'gpt-5.3-chat-latest', 'gpt-4.1', 'o4-mini'],
+    planUrl: 'https://platform.openai.com/api-keys', planLabelKey: 'system.createApiKey',
+    planHintKey: 'aiops.openaiHint',
+  },
+  {
+    id: 'anthropic', nameKey: 'aiops.providerAnthropic', endpoint: 'https://api.anthropic.com',
+    models: ['claude-sonnet-5', 'claude-fable-5', 'claude-opus-4-8', 'claude-sonnet-4-6', 'claude-haiku-4-5'],
+    planUrl: 'https://console.anthropic.com/settings/keys', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'google', nameKey: 'aiops.providerGemini', endpoint: 'https://generativelanguage.googleapis.com/v1beta/openai',
+    models: ['gemini-3.5-flash', 'gemini-3.1-pro-preview', 'gemini-3-flash-preview', 'gemini-2.5-pro', 'gemini-2.5-flash'],
+    planUrl: 'https://aistudio.google.com/app/apikey', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'xai', nameKey: 'aiops.providerXai', endpoint: 'https://api.x.ai/v1',
+    models: ['grok-4.5', 'grok-4.3', 'grok-4.20-0309-reasoning', 'grok-4.20-0309-non-reasoning'],
+    planUrl: 'https://console.x.ai/', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'mistral', nameKey: 'aiops.providerMistral', endpoint: 'https://api.mistral.ai/v1',
+    models: ['mistral-large-latest', 'mistral-medium-latest', 'mistral-small-latest', 'codestral-latest'],
+    planUrl: 'https://console.mistral.ai/api-keys', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'groq', nameKey: 'aiops.providerGroq', endpoint: 'https://api.groq.com/openai/v1',
+    models: ['openai/gpt-oss-120b', 'llama-3.3-70b-versatile', 'qwen/qwen3-32b'],
+    planUrl: 'https://console.groq.com/keys', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'openrouter', nameKey: 'aiops.providerOpenrouter', endpoint: 'https://openrouter.ai/api/v1',
+    models: ['openai/gpt-5.4', 'anthropic/claude-sonnet-4.6', 'google/gemini-3.1-pro-preview', 'deepseek/deepseek-v4-pro'],
+    planUrl: 'https://openrouter.ai/settings/keys', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'together', nameKey: 'aiops.providerTogether', endpoint: 'https://api.together.ai/v1',
+    models: ['deepseek-ai/DeepSeek-V4-Pro', 'moonshotai/Kimi-K3', 'Qwen/Qwen3.7-Max', 'openai/gpt-oss-120b'],
+    planUrl: 'https://api.together.ai/settings/api-keys', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'cerebras', nameKey: 'aiops.providerCerebras', endpoint: 'https://api.cerebras.ai/v1',
+    models: ['gpt-oss-120b', 'zai-glm-4.7', 'gemma-4-31b'],
+    planUrl: 'https://cloud.cerebras.ai/platform/', planLabelKey: 'aiops.getApiKey',
+  },
+  {
+    id: 'perplexity', nameKey: 'aiops.providerPerplexity', endpoint: 'https://api.perplexity.ai',
+    models: ['sonar-pro', 'sonar-reasoning-pro', 'sonar'],
+    planUrl: 'https://www.perplexity.ai/account/api/keys', planLabelKey: 'aiops.getApiKey',
+  },
+];
+
+// 编码套餐使用独立密钥或专用端点，不与通用 API 配置混用。 // Logan.Li
+export const CLOUD_CODING_PROVIDERS: ProviderMeta[] = [
+  {
+    id: 'glm-coding', nameKey: 'aiops.zhipuCoding', endpoint: 'https://open.bigmodel.cn/api/anthropic',
+    models: ['GLM-5.2', 'GLM-5.1'], planUrl: 'https://bigmodel.cn/glm-coding',
+    planLabelKey: 'aiops.buyGlmCoding', planHintKey: 'aiops.zhipuCodingHint',
+  },
+  {
+    id: 'kimi', nameKey: 'aiops.kimiCoding', endpoint: 'https://api.kimi.com/coding/v1',
+    models: ['kimi-for-coding', 'kimi-coding-latest'], planUrl: 'https://www.kimi.com/membership/pricing',
+    planLabelKey: 'aiops.buyKimiPlan', planHintKey: 'aiops.kimiCodingEndpoint',
+  },
+  {
+    id: 'minimax-coding', nameKey: 'aiops.minimaxCoding', endpoint: 'https://api.minimaxi.com/anthropic',
+    models: ['MiniMax-M3', 'MiniMax-M3[1m]', 'MiniMax-M2.7'],
+    planUrl: 'https://platform.minimaxi.com/user-center/payment/token-plan',
+    planLabelKey: 'aiops.buyTokenPack', planHintKey: 'aiops.anthropicSubHint',
+  },
+];
+
+// New API/One API 是需要密钥的 OpenAI 兼容网关，自定义接入则允许无鉴权。 // Logan.Li
+export const GATEWAY_PROVIDERS: ProviderMeta[] = [
+  {
+    id: 'new-api', nameKey: 'aiops.providerNewApi', endpoint: '', models: [],
+    planHintKey: 'aiops.newApiHint',
+  },
+  {
+    id: 'custom', nameKey: 'aiops.customProvider', endpoint: '', models: [],
+    apiKeyOptional: true, planHintKey: 'aiops.customProviderDesc',
+  },
+];
+
+// 本地与自建服务默认不要求 API Key。 // Logan.Li
+export const LOCAL_PROVIDERS: ProviderMeta[] = [
+  {
+    id: 'ollama', nameKey: 'aiops.ollamaLocal', endpoint: 'http://localhost:11434/v1',
+    models: ['qwen3:8b', 'llama3.3', 'deepseek-r1:8b'], apiKeyOptional: true,
+  },
+  {
+    id: 'vllm', nameKey: 'aiops.vllmSelfHosted', endpoint: 'http://localhost:8000/v1',
+    models: [], apiKeyOptional: true,
+  },
+  {
+    id: 'lmstudio', nameKey: 'aiops.providerLmStudio', endpoint: 'http://localhost:1234/v1',
+    models: [], apiKeyOptional: true,
+  },
+];
+
+export const PROVIDERS: ProviderMeta[] = [
+  ...CLOUD_GENERAL_PROVIDERS,
+  ...CLOUD_CODING_PROVIDERS,
+  ...GATEWAY_PROVIDERS,
+  ...LOCAL_PROVIDERS,
+];
+
+export const findProvider = (provider: string) => PROVIDERS.find((item) => item.id === provider);
+
+export const needsManualModel = (provider: string, meta?: ProviderMeta) =>
+  provider === 'custom' || provider === 'new-api' || provider === 'vllm'
+  || provider === 'lmstudio' || !meta?.models.length;
+
+export const endpointPlaceholder = (provider: string) => {
+  switch (provider) {
+    case 'new-api': return 'https://new-api.example.com/v1';
+    case 'custom': return 'https://gateway.example.com/v1';
+    case 'ollama': return 'http://localhost:11434/v1';
+    case 'vllm': return 'http://localhost:8000/v1';
+    case 'lmstudio': return 'http://localhost:1234/v1';
+    default: return 'https://api.example.com/v1';
+  }
+};
diff --git a/frontend/src/locales/en-US.json b/frontend/src/locales/en-US.json
index fd32787..4da9073 100644
--- a/frontend/src/locales/en-US.json
+++ b/frontend/src/locales/en-US.json
@@ -8380,12 +8380,14 @@
       "eventAborted": "Task aborted",
       "eventFailed": "Task execution failed"
     },
-    "cloudService": "Cloud Services",
+    "agentCliSharedConfig": "This connection is shared by AI Assistant and Agent CLI",
+    "apiGateway": "API Gateways",
+    "cloudService": "Official Cloud Services",
     "codingPlan": "Coding Plan",
     "collapseHistory": "Collapse history",
     "confidenceLabel": "Confidence",
     "confidencePercent": "Confidence {{n}}%",
-    "configHelpText": "Cloud Services: Select a provider → Activate and enter API Key → Test Connection → Save.\n          Coding Plans: GLM / Kimi / MiniMax fixed monthly plans, ideal for frequent Agent CLI usage (note: keys are not shared with pay-as-you-go versions).\n          Custom LLM: Enter endpoint + API Key + model name, compatible with OpenRouter and other OpenAI-format services.\n          Local/Self-hosted (Ollama, vLLM): Enter endpoint and model name; API Key can be left blank.\n          Testing sends \"What model are you? Please answer briefly.\" to confirm the target model.",
+    "configHelpText": "Official cloud service: Select a provider and enter its API Key; you can load the models available to that key.\n          API gateway: For New API / One API, enter the site's /v1 URL, token, and model name.\n          Local/self-hosted (Ollama, vLLM, LM Studio): Enter the Endpoint and model name; API Key is optional.\n          Test the connection before saving; the test sends one short message to confirm the actual model.",
     "configLlmFirst": "Please configure an LLM first...",
     "confirmDeleteSession": "Delete \"{{title}}\"? This action cannot be undone.",
     "confirmExecute": "Confirm Execute",
@@ -8406,7 +8408,7 @@
     "currentValue": "Current",
     "customAccess": "Custom Access",
     "customProvider": "Custom LLM",
-    "customProviderDesc": "Fill in an OpenAI-compatible endpoint, API Key (blank if no auth) and model name; works with OpenRouter, intranet proxies and other unlisted services",
+    "customProviderDesc": "Enter an OpenAI-compatible endpoint, API Key (blank if no auth), and model name; suitable for internal proxies and other unlisted services",
     "customProxyHint": "Direct access from China may be restricted; use your own network or a compatible proxy via \"Custom LLM\"",
     "dailyOpsReport": "Generate ops report",
     "daysToThreshold": "reaches threshold in {{days}} days",
@@ -8420,6 +8422,7 @@
     "enableLlm": "Enable LLM Service",
     "enableReasoning": "Enable reasoning mode",
     "endpointLabel": "Endpoint",
+    "endpointModelHint": "Enter the endpoint ID created in the provider console",
     "errorOccurred": "An error occurred",
     "examplePrefix": "Ex: ",
     "executeAutoFix": "Execute Auto-Fix",
@@ -8479,12 +8482,14 @@
     "llmTopP": "Top P",
     "llmVisionModel": "Vision Model",
     "loadAlertsDot": "Loading alert data...",
+    "loadModels": "Load Models",
     "loadAnomalyDot": "Loading anomaly forecast...",
     "loadingCapacityData": "Loading capacity data...",
     "loadingMetricData": "Loading metric data...",
     "localNoKeyHint": "Local or intranet self-hosted services usually need no key; fill it in only if the gateway requires auth.",
     "localSelfHosted": "Local / Self-hosted",
     "manualInputModel": "Enter model name manually",
+    "manualModelHint": "Enter the model ID shown in the provider console",
     "mayRecur": "{{name}} may recur",
     "mayTrigger": "{{name}} may trigger",
     "memoryKeyword": "Memory",
@@ -8495,11 +8500,14 @@
     "minimaxPayg": "MiniMax (pay-as-you-go)",
     "modelColon": "Model: ",
     "modelLabel": "Model",
+    "modelListFailed": "Could not load models; check the Endpoint, API Key, and upstream service",
     "modelPlaceholder": "e.g. deepseek/deepseek-chat or gpt-4o",
     "modelUnknown": "Unknown",
+    "modelsDiscovered": "Loaded {{count}} models",
     "networkKeyword": "Network",
     "newConversation": "New Chat",
     "newConversationStarted": "New conversation started",
+    "newApiHint": "Enter the OpenAI-compatible URL of your New API or One API instance (usually ending in /v1), its token, and the model name",
     "newSessionBtn": "New",
     "newTask": "New Task",
     "next12Hours": "Next 12 hours",
@@ -8515,6 +8523,7 @@
     "noHostsHint": "No hosts yet; add servers in \"Asset Management\" first",
     "noMetricData": "No metric data",
     "noModelContent": "(The model returned no content)",
+    "noModelsDiscovered": "The upstream returned no models; enter the model name manually",
     "noneFallback": "(none)",
     "notConnectedHint": "Connection not ready; check the LLM config or refresh the page",
     "notConnectedRetry": "Connection not ready; please retry later",
@@ -8620,23 +8629,33 @@
     "providerAndCredential": "Model & Credential",
     "providerAnthropic": "Anthropic",
     "providerAzure": "Azure OpenAI",
+    "providerBaidu": "Baidu Qianfan",
     "providerBaichuan": "Baichuan",
+    "providerCerebras": "Cerebras",
     "providerCohere": "Cohere",
     "providerColon": "Provider: ",
     "providerCustom": "Custom",
     "providerDeepseek": "DeepSeek",
     "providerGemini": "Google Gemini",
+    "providerGroq": "Groq",
     "providerLabel": "Provider",
+    "providerLmStudio": "LM Studio",
     "providerMinimax": "MiniMax",
     "providerMistral": "Mistral",
     "providerMoonshot": "Moonshot",
+    "providerNewApi": "New API / One API",
     "providerNeedsKey": "To use {{name}}, you need to activate the service and obtain an API Key first",
     "providerOllama": "Ollama",
     "providerOpenai": "OpenAI",
     "providerOpenrouter": "OpenRouter",
+    "providerPerplexity": "Perplexity",
     "providerQwen": "Qwen",
     "providerSiliconflow": "SiliconFlow",
     "providerSpark": "iFlytek Spark",
+    "providerTencentTokenHub": "Tencent Cloud TokenHub",
+    "providerTogether": "Together AI",
+    "providerVolcengine": "Volcengine Ark",
+    "providerXai": "xAI Grok",
     "providerYi": "01.AI",
     "providerZhipu": "Zhipu GLM",
     "quickActions": "Quick Actions",
@@ -8646,7 +8665,7 @@
     "reasoningHigh": "High",
     "reasoningLow": "Low",
     "reasoningMedium": "Medium",
-    "reasoningModelHint": "Suitable for reasoning models like OpenAI o-series / DeepSeek R1 / Claude 3.7 Sonnet. May cause errors with standard models.",
+    "reasoningModelHint": "For reasoning models such as OpenAI GPT-5 / o-series, DeepSeek R1, and Claude Sonnet / Opus 4.",
     "reasoningOn": "Reasoning mode is on",
     "reasoningPlanMode": "Reasoning / Plan Mode",
     "recommended": "Recommended",
diff --git a/frontend/src/locales/zh-CN.json b/frontend/src/locales/zh-CN.json
index 3e3a3cb..6833db5 100644
--- a/frontend/src/locales/zh-CN.json
+++ b/frontend/src/locales/zh-CN.json
@@ -8380,12 +8380,14 @@
       "eventAborted": "任务已取消",
       "eventFailed": "任务执行失败"
     },
-    "cloudService": "云服务",
+    "agentCliSharedConfig": "智能助手与 Agent CLI 共用此连接配置",
+    "apiGateway": "API 网关",
+    "cloudService": "官方云服务",
     "codingPlan": "编码套餐 Coding Plan",
     "collapseHistory": "收起历史任务",
     "confidenceLabel": "置信度",
     "confidencePercent": "置信度 {{n}}%",
-    "configHelpText": "云服务：选择提供商 → 开通并填写 API Key → 测试连接 → 保存。\n          编码套餐：GLM / Kimi / MiniMax 固定月费套餐，适合 Agent CLI 高频调用（注意 Key 与按量版不互通）。\n          大模型自定义：填请求地址 + API Key + 模型名，兼容 OpenRouter 等 OpenAI 格式服务。\n          本地/自建（Ollama、vLLM）：填写 Endpoint 与模型名即可，API Key 可留空。\n          测试时会发送「你是什么模型？请简短回答。」以确认接入目标模型。",
+    "configHelpText": "官方云服务：选择提供商并填写 API Key，可读取当前密钥可用的模型。\n          API 网关：New API / One API 需填写站点的 /v1 地址、令牌和模型名。\n          本地/自建（Ollama、vLLM、LM Studio）：填写 Endpoint 与模型名即可，API Key 可留空。\n          保存前必须先测试连接；测试会发送一条简短消息确认实际接入模型。",
     "configLlmFirst": "请先配置大模型...",
     "confirmDeleteSession": "确定删除「{{title}}」？此操作不可恢复。",
     "confirmExecute": "确认执行",
@@ -8406,7 +8408,7 @@
     "currentValue": "当前值",
     "customAccess": "自定义接入",
     "customProvider": "大模型自定义",
-    "customProviderDesc": "填写 OpenAI 兼容的请求地址、API Key（无鉴权可留空）和模型名；适用于 OpenRouter、内网代理、其他未列出的服务",
+    "customProviderDesc": "填写 OpenAI 兼容的请求地址、API Key（无鉴权可留空）和模型名；适用于内网代理或其他未列出的服务",
     "customProxyHint": "国内直连可能受限，需自备网络或使用「大模型自定义」接入兼容代理",
     "dailyOpsReport": "生成运维报告",
     "daysToThreshold": "{{days}} 天后触及阈值",
@@ -8420,6 +8422,7 @@
     "enableLlm": "启用 LLM 服务",
     "enableReasoning": "启用推理模式",
     "endpointLabel": "接口地址（Endpoint）",
+    "endpointModelHint": "请填写在提供商控制台创建的接入点 ID",
     "errorOccurred": "发生错误",
     "examplePrefix": "例：",
     "executeAutoFix": "执行自动修复",
@@ -8479,12 +8482,14 @@
     "llmTopP": "Top P",
     "llmVisionModel": "视觉模型",
     "loadAlertsDot": "加载告警数据...",
+    "loadModels": "获取模型",
     "loadAnomalyDot": "加载异常预测...",
     "loadingCapacityData": "加载容量数据...",
     "loadingMetricData": "加载指标数据...",
     "localNoKeyHint": "本地或内网自建通常无需密钥；若网关启用了鉴权再填写。",
     "localSelfHosted": "本地 / 自建",
     "manualInputModel": "手动输入模型名",
+    "manualModelHint": "请按提供商控制台显示的模型 ID 手动填写",
     "mayRecur": "{{name}}可能再次发生",
     "mayTrigger": "{{name}}可能触发",
     "memoryKeyword": "内存",
@@ -8495,11 +8500,14 @@
     "minimaxPayg": "MiniMax (按量)",
     "modelColon": "模型:",
     "modelLabel": "模型",
+    "modelListFailed": "读取模型列表失败，请检查 Endpoint、API Key 和上游服务",
     "modelPlaceholder": "如 deepseek/deepseek-chat 或 gpt-4o",
     "modelUnknown": "未知",
+    "modelsDiscovered": "已加载 {{count}} 个模型",
     "networkKeyword": "网络",
     "newConversation": "新对话",
     "newConversationStarted": "已开始新对话",
+    "newApiHint": "填写 New API 或 One API 站点的 OpenAI 兼容地址（通常以 /v1 结尾）、令牌和模型名",
     "newSessionBtn": "新建",
     "newTask": "新建任务",
     "next12Hours": "未来12小时",
@@ -8515,6 +8523,7 @@
     "noHostsHint": "暂无主机，请先前往「资产管理」添加服务器",
     "noMetricData": "暂无指标数据",
     "noModelContent": "（模型未返回内容）",
+    "noModelsDiscovered": "上游没有返回可用模型，请手动填写模型名",
     "noneFallback": "（无）",
     "notConnectedHint": "连接未就绪，请检查大模型配置或刷新页面重试",
     "notConnectedRetry": "连接未就绪，请稍后重试",
@@ -8620,23 +8629,33 @@
     "providerAndCredential": "模型与凭证",
     "providerAnthropic": "Anthropic",
     "providerAzure": "Azure OpenAI",
+    "providerBaidu": "百度千帆",
     "providerBaichuan": "百川",
+    "providerCerebras": "Cerebras",
     "providerCohere": "Cohere",
     "providerColon": "提供商:",
     "providerCustom": "自定义",
     "providerDeepseek": "DeepSeek",
     "providerGemini": "Google Gemini",
+    "providerGroq": "Groq",
     "providerLabel": "提供商",
+    "providerLmStudio": "LM Studio",
     "providerMinimax": "MiniMax",
     "providerMistral": "Mistral",
     "providerMoonshot": "月之暗面",
+    "providerNewApi": "New API / One API",
     "providerNeedsKey": "使用 {{name}} 需先开通服务并获取 API Key",
     "providerOllama": "Ollama",
     "providerOpenai": "OpenAI",
     "providerOpenrouter": "OpenRouter",
+    "providerPerplexity": "Perplexity",
     "providerQwen": "通义千问",
     "providerSiliconflow": "硅基流动",
     "providerSpark": "讯飞星火",
+    "providerTencentTokenHub": "腾讯云 TokenHub",
+    "providerTogether": "Together AI",
+    "providerVolcengine": "火山引擎方舟",
+    "providerXai": "xAI Grok",
     "providerYi": "零一万物",
     "providerZhipu": "智谱 GLM",
     "quickActions": "快速操作",
@@ -8646,7 +8665,7 @@
     "reasoningHigh": "高",
     "reasoningLow": "低",
     "reasoningMedium": "中",
-    "reasoningModelHint": "适用于 OpenAI o-series / DeepSeek R1 / Claude 3.7 Sonnet 等推理模型；普通模型开启后可能报错。",
+    "reasoningModelHint": "适用于 OpenAI GPT-5 / o-series、DeepSeek R1、Claude Sonnet / Opus 4 等推理模型。",
     "reasoningOn": "推理模式已开启",
     "reasoningPlanMode": "推理 / Plan 模式",
     "recommended": "推荐",