feat: add opus-4-7 support + nginx load balancer docker-compose

chore: merge upstream v0.1.114, keep Antigravity customizations
chore: sync VERSION to 0.1.114 [skip ci]
2026-04-17 11:43:38 +08:00 · 2026-04-17 11:42:27 +08:00 · 2026-04-17 02:51:18 +00:00 · 2026-04-17 10:28:12 +08:00 · 2026-04-17 10:19:14 +08:00 · 2026-04-17 10:18:56 +08:00
20 changed files with 540 additions and 264 deletions
--- a/backend/cmd/server/VERSION
+++ b/backend/cmd/server/VERSION
@ -1 +1 @@
-0.1.113
+0.1.114
--- a/backend/internal/domain/constants.go
+++ b/backend/internal/domain/constants.go
@ -71,6 +71,7 @@ const (
 // 与前端 useModelWhitelist.ts 中的 antigravityDefaultMappings 保持一致
 var DefaultAntigravityModelMapping = map[string]string{
 	// Claude 白名单
+	"claude-opus-4-7":            "claude-opus-4-7",          // 官方模型
 	"claude-opus-4-6-thinking":   "claude-opus-4-6-thinking", // 官方模型
 	"claude-opus-4-6":            "claude-opus-4-6-thinking", // 简称映射
 	"claude-opus-4-5-thinking":   "claude-opus-4-6-thinking", // 迁移旧模型
@ -120,6 +121,7 @@ var DefaultAntigravityModelMapping = map[string]string{
 // aws_region 自动调整为匹配的区域前缀（如 eu.、apac.、jp. 等）
 var DefaultBedrockModelMapping = map[string]string{
 	// Claude Opus
+	"claude-opus-4-7":          "us.anthropic.claude-opus-4-7-v1",
 	"claude-opus-4-6-thinking": "us.anthropic.claude-opus-4-6-v1",
 	"claude-opus-4-6":          "us.anthropic.claude-opus-4-6-v1",
 	"claude-opus-4-5-thinking": "us.anthropic.claude-opus-4-5-20251101-v1:0",
--- a/backend/internal/handler/antigravity_http.go
+++ b/backend/internal/handler/antigravity_http.go
@ -213,7 +213,7 @@ func (h *AntigravityHTTPHandler) GetModels(c *gin.Context) {

 	c.JSON(http.StatusOK, gin.H{
 		"models":        models,
-		"default_model": "claude-opus-4-6",
+		"default_model": "claude-opus-4-7",
 	})
 }

--- a/backend/internal/pkg/antigravity/claude_types.go
+++ b/backend/internal/pkg/antigravity/claude_types.go
@ -162,6 +162,7 @@ var claudeModels = []modelDef{
 	{ID: "claude-sonnet-4-5-thinking", DisplayName: "Claude Sonnet 4.5 Thinking", CreatedAt: "2025-09-29T00:00:00Z"},
 	{ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6", CreatedAt: "2026-02-05T00:00:00Z"},
 	{ID: "claude-opus-4-6-thinking", DisplayName: "Claude Opus 4.6 Thinking", CreatedAt: "2026-02-05T00:00:00Z"},
+	{ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7", CreatedAt: "2026-04-17T00:00:00Z"},
 	{ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6", CreatedAt: "2026-02-17T00:00:00Z"},
 }

--- a/backend/internal/pkg/antigravity/request_transformer.go
+++ b/backend/internal/pkg/antigravity/request_transformer.go
@ -1023,8 +1023,12 @@ func maxOutputTokensLimit(model string) int {
 	return maxOutputTokensUpperBound
 }

-func isAntigravityOpus46Model(model string) bool {
-	return strings.HasPrefix(strings.ToLower(model), "claude-opus-4-6")
+// isAntigravityOpusHighTierModel 判断是否为高阶 Opus 模型（4.6+），
+// 用于 adaptive thinking 时覆写为高预算。
+func isAntigravityOpusHighTierModel(model string) bool {
+	lower := strings.ToLower(model)
+	return strings.HasPrefix(lower, "claude-opus-4-6") ||
+		strings.HasPrefix(lower, "claude-opus-4-7")
 }

 func buildGenerationConfig(req *ClaudeRequest) *GeminiGenerationConfig {
@ -1046,12 +1050,12 @@ func buildGenerationConfig(req *ClaudeRequest) *GeminiGenerationConfig {
 		}

 		// - thinking.type=enabled：budget_tokens>0 用显式预算
-		// - thinking.type=adaptive：仅在 Antigravity 的 Opus 4.6 上覆写为 （24576）
+		// - thinking.type=adaptive：在 Antigravity 的高阶 Opus（4.6+）上覆写为 （24576）
 		budget := -1
 		if req.Thinking.BudgetTokens > 0 {
 			budget = req.Thinking.BudgetTokens
 		}
-		if req.Thinking.Type == "adaptive" && isAntigravityOpus46Model(req.Model) {
+		if req.Thinking.Type == "adaptive" && isAntigravityOpusHighTierModel(req.Model) {
 			budget = ClaudeAdaptiveHighThinkingBudgetTokens
 		}

--- a/backend/internal/pkg/claude/constants.go
+++ b/backend/internal/pkg/claude/constants.go
@ -341,6 +341,12 @@ var DefaultModels = []Model{
 		DisplayName: "Claude Opus 4.6",
 		CreatedAt:   "2026-02-06T00:00:00Z",
 	},
+	{
+		ID:          "claude-opus-4-7",
+		Type:        "model",
+		DisplayName: "Claude Opus 4.7",
+		CreatedAt:   "2026-04-17T00:00:00Z",
+	},
 	{
 		ID:          "claude-sonnet-4-6",
 		Type:        "model",
--- a/backend/internal/service/billing_service.go
+++ b/backend/internal/service/billing_service.go
@ -191,6 +191,9 @@ func (s *BillingService) initFallbackPricing() {
 	// Claude 4.6 Opus (与4.5同价)
 	s.fallbackPrices["claude-opus-4.6"] = s.fallbackPrices["claude-opus-4.5"]

+	// Claude 4.7 Opus (暂与4.6同价，待官方定价更新)
+	s.fallbackPrices["claude-opus-4.7"] = s.fallbackPrices["claude-opus-4.6"]
+
 	// Gemini 3.1 Pro
 	s.fallbackPrices["gemini-3.1-pro"] = &ModelPricing{
 		InputPricePerToken:         2e-6,   // $2 per MTok
@ -278,6 +281,9 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing {

 	// 按模型系列匹配
 	if strings.Contains(modelLower, "opus") {
+		if strings.Contains(modelLower, "4.7") || strings.Contains(modelLower, "4-7") {
+			return s.fallbackPrices["claude-opus-4.7"]
+		}
 		if strings.Contains(modelLower, "4.6") || strings.Contains(modelLower, "4-6") {
 			return s.fallbackPrices["claude-opus-4.6"]
 		}
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@ -5180,19 +5180,8 @@ func (s *GatewayService) handleNonStreamingResponseAnthropicAPIKeyPassthrough(
 		s.rateLimitService.UpdateSessionWindow(ctx, account, resp.Header)
 	}

-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	body, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, anthropicTooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"type": "error",
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

@ -5558,19 +5547,8 @@ func (s *GatewayService) handleBedrockNonStreamingResponse(
 	c *gin.Context,
 	account *Account,
 ) (*ClaudeUsage, error) {
-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	body, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, anthropicTooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"type": "error",
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

@ -7277,19 +7255,8 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
 	// 更新5h窗口状态
 	s.rateLimitService.UpdateSessionWindow(ctx, account, resp.Header)

-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	body, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, anthropicTooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"type": "error",
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

@ -8402,16 +8369,15 @@ func (s *GatewayService) ForwardCountTokens(ctx context.Context, c *gin.Context,
 	}

 	// 读取响应体
-	maxReadBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	respBody, err := readUpstreamResponseBodyLimited(resp.Body, maxReadBytes)
+	countTokensTooLarge := func(c *gin.Context) {
+		s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response too large")
+	}
+	respBody, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, countTokensTooLarge)
 	_ = resp.Body.Close()
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response too large")
-			return err
+		if !errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
+			s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 		}
-		s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 		return err
 	}

@ -8425,15 +8391,12 @@ func (s *GatewayService) ForwardCountTokens(ctx context.Context, c *gin.Context,
 			retryResp, retryErr := s.httpUpstream.DoWithTLS(retryReq, proxyURL, account.ID, account.Concurrency, s.tlsFPProfileService.ResolveTLSProfile(account))
 			if retryErr == nil {
 				resp = retryResp
-				respBody, err = readUpstreamResponseBodyLimited(resp.Body, maxReadBytes)
+				respBody, err = ReadUpstreamResponseBody(resp.Body, s.cfg, c, countTokensTooLarge)
 				_ = resp.Body.Close()
 				if err != nil {
-					if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-						setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-						s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response too large")
-						return err
+					if !errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
+						s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 					}
-					s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 					return err
 				}
 			}
@ -8528,16 +8491,15 @@ func (s *GatewayService) forwardCountTokensAnthropicAPIKeyPassthrough(ctx contex
 		return fmt.Errorf("upstream request failed: %w", err)
 	}

-	maxReadBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	respBody, err := readUpstreamResponseBodyLimited(resp.Body, maxReadBytes)
+	countTokensTooLarge := func(c *gin.Context) {
+		s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response too large")
+	}
+	respBody, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, countTokensTooLarge)
 	_ = resp.Body.Close()
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response too large")
-			return err
+		if !errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
+			s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 		}
-		s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
 		return err
 	}

--- a/backend/internal/service/gemini_messages_compat_service.go
+++ b/backend/internal/service/gemini_messages_compat_service.go
@ -2424,18 +2424,8 @@ func (s *GeminiMessagesCompatService) handleNativeNonStreamingResponse(c *gin.Co
 		logger.LegacyPrintf("service.gemini_messages_compat", "[GeminiAPI] ========================================")
 	}

-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	respBody, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	respBody, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, openAITooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

--- a/backend/internal/service/language_server_service.go
+++ b/backend/internal/service/language_server_service.go
@ -310,6 +310,24 @@ type ModelConfig struct {
 // GetAvailableModels 获取可用模型列表
 func (svc *LanguageServerService) GetAvailableModels(ctx context.Context) ([]ModelConfig, error) {
 	models := []ModelConfig{
+		{
+			Name:             "claude-opus-4-7",
+			DisplayName:      "Claude Opus 4.7",
+			MaxTokens:        200000,
+			SupportsThinking: true,
+			ThinkingBudget:   32000,
+			SupportsImages:   true,
+			Provider:         "anthropic",
+		},
+		{
+			Name:             "claude-sonnet-4-7",
+			DisplayName:      "Claude Sonnet 4.7",
+			MaxTokens:        200000,
+			SupportsThinking: true,
+			ThinkingBudget:   16000,
+			SupportsImages:   true,
+			Provider:         "anthropic",
+		},
 		{
 			Name:             "claude-opus-4-6",
 			DisplayName:      "Claude Opus 4.6",
--- a/backend/internal/service/openai_gateway_messages.go
+++ b/backend/internal/service/openai_gateway_messages.go
@ -121,6 +121,28 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic(
 		}
 	}

+	// For API key accounts (including OpenAI-compatible upstream gateways),
+	// ensure promptCacheKey is also propagated via the request body so that
+	// upstreams using the Responses API can derive a stable session identifier
+	// from prompt_cache_key. This makes our Anthropic /v1/messages compatibility
+	// path behave more like a native Responses client.
+	if account.Type == AccountTypeAPIKey {
+		if trimmedKey := strings.TrimSpace(promptCacheKey); trimmedKey != "" {
+			var reqBody map[string]any
+			if err := json.Unmarshal(responsesBody, &reqBody); err != nil {
+				return nil, fmt.Errorf("unmarshal for prompt cache key injection: %w", err)
+			}
+			if existing, ok := reqBody["prompt_cache_key"].(string); !ok || strings.TrimSpace(existing) == "" {
+				reqBody["prompt_cache_key"] = trimmedKey
+				updated, err := json.Marshal(reqBody)
+				if err != nil {
+					return nil, fmt.Errorf("remarshal after prompt cache key injection: %w", err)
+				}
+				responsesBody = updated
+			}
+		}
+	}
+
 	// 5. Get access token
 	token, _, err := s.GetAccessToken(ctx, account)
 	if err != nil {
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@ -3010,18 +3010,8 @@ func (s *OpenAIGatewayService) handleNonStreamingResponsePassthrough(
 	resp *http.Response,
 	c *gin.Context,
 ) (*OpenAIUsage, error) {
-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	body, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, openAITooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

@ -3919,18 +3909,8 @@ func extractOpenAIUsageFromJSONBytes(body []byte) (OpenAIUsage, bool) {
 }

 func (s *OpenAIGatewayService) handleNonStreamingResponse(ctx context.Context, resp *http.Response, c *gin.Context, account *Account, originalModel, mappedModel string) (*OpenAIUsage, error) {
-	maxBytes := resolveUpstreamResponseReadLimit(s.cfg)
-	body, err := readUpstreamResponseBodyLimited(resp.Body, maxBytes)
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, openAITooLargeError)
 	if err != nil {
-		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
-			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
-			c.JSON(http.StatusBadGateway, gin.H{
-				"error": gin.H{
-					"type":    "upstream_error",
-					"message": "Upstream response too large",
-				},
-			})
-		}
 		return nil, err
 	}

--- a/backend/internal/service/pricing_service.go
+++ b/backend/internal/service/pricing_service.go
@ -656,65 +656,95 @@ func (s *PricingService) extractBaseName(model string) string {

 // matchByModelFamily 基于模型系列匹配
 func (s *PricingService) matchByModelFamily(model string) *LiteLLMModelPricing {
-	// Claude模型系列匹配规则
-	familyPatterns := map[string][]string{
-		"opus-4.6":   {"claude-opus-4.6", "claude-opus-4-6"},
-		"opus-4.5":   {"claude-opus-4.5", "claude-opus-4-5"},
-		"opus-4":     {"claude-opus-4", "claude-3-opus"},
-		"sonnet-4.5": {"claude-sonnet-4.5", "claude-sonnet-4-5"},
-		"sonnet-4":   {"claude-sonnet-4", "claude-3-5-sonnet"},
-		"sonnet-3.5": {"claude-3-5-sonnet", "claude-3.5-sonnet"},
-		"sonnet-3":   {"claude-3-sonnet"},
-		"haiku-3.5":  {"claude-3-5-haiku", "claude-3.5-haiku"},
-		"haiku-3":    {"claude-3-haiku"},
+	// modelFamily 定义一个模型系列的匹配和定价查找规则。
+	type modelFamily struct {
+		name    string   // 系列名称
+		match   []string // 用于将模型归类到此系列的模式（strings.Contains 匹配）
+		pricing []string // 用于在定价数据中查找价格的模式（nil 则复用 match；可包含低版本 fallback）
 	}

-	// 确定模型属于哪个系列
-	var matchedFamily string
-	for family, patterns := range familyPatterns {
-		for _, pattern := range patterns {
+	// 按特异性降序排列：高版本号在前，避免 "claude-opus-4"（opus-4 系列）
+	// 因子串关系误匹配 "claude-opus-4-7"（opus-4.7 系列）。
+	// 注意：原 map 实现存在 Go map 迭代随机性导致的同类 bug，此处改为有序切片修复。
+	families := []modelFamily{
+		{name: "opus-4.7", match: []string{"claude-opus-4-7", "claude-opus-4.7"}, pricing: []string{"claude-opus-4-7", "claude-opus-4.7", "claude-opus-4-6"}},
+		{name: "opus-4.6", match: []string{"claude-opus-4-6", "claude-opus-4.6"}},
+		{name: "opus-4.5", match: []string{"claude-opus-4-5", "claude-opus-4.5"}},
+		{name: "opus-4", match: []string{"claude-opus-4", "claude-3-opus"}},
+		{name: "sonnet-4.5", match: []string{"claude-sonnet-4-5", "claude-sonnet-4.5"}},
+		{name: "sonnet-4", match: []string{"claude-sonnet-4", "claude-3-5-sonnet"}},
+		{name: "sonnet-3.5", match: []string{"claude-3-5-sonnet", "claude-3.5-sonnet"}},
+		{name: "sonnet-3", match: []string{"claude-3-sonnet"}},
+		{name: "haiku-3.5", match: []string{"claude-3-5-haiku", "claude-3.5-haiku"}},
+		{name: "haiku-3", match: []string{"claude-3-haiku"}},
+	}
+
+	// Phase 1: 按有序切片归类（最具体的系列优先匹配）
+	var matched *modelFamily
+	for i := range families {
+		for _, pattern := range families[i].match {
 			if strings.Contains(model, pattern) || strings.Contains(model, strings.ReplaceAll(pattern, "-", "")) {
-				matchedFamily = family
+				matched = &families[i]
 				break
 			}
 		}
-		if matchedFamily != "" {
+		if matched != nil {
 			break
 		}
 	}

-	if matchedFamily == "" {
-		// 简单的系列匹配
-		if strings.Contains(model, "opus") {
-			if strings.Contains(model, "4.5") || strings.Contains(model, "4-5") {
-				matchedFamily = "opus-4.5"
-			} else {
-				matchedFamily = "opus-4"
+	// Phase 2: 二次兜底——当模型 ID 不含已知模式串时，按关键字粗分
+	if matched == nil {
+		var fallbackName string
+		switch {
+		case strings.Contains(model, "opus"):
+			switch {
+			case strings.Contains(model, "4.7") || strings.Contains(model, "4-7"):
+				fallbackName = "opus-4.7"
+			case strings.Contains(model, "4.6") || strings.Contains(model, "4-6"):
+				fallbackName = "opus-4.6"
+			case strings.Contains(model, "4.5") || strings.Contains(model, "4-5"):
+				fallbackName = "opus-4.5"
+			default:
+				fallbackName = "opus-4"
 			}
-		} else if strings.Contains(model, "sonnet") {
-			if strings.Contains(model, "4.5") || strings.Contains(model, "4-5") {
-				matchedFamily = "sonnet-4.5"
-			} else if strings.Contains(model, "3-5") || strings.Contains(model, "3.5") {
-				matchedFamily = "sonnet-3.5"
-			} else {
-				matchedFamily = "sonnet-4"
+		case strings.Contains(model, "sonnet"):
+			switch {
+			case strings.Contains(model, "4.5") || strings.Contains(model, "4-5"):
+				fallbackName = "sonnet-4.5"
+			case strings.Contains(model, "3-5") || strings.Contains(model, "3.5"):
+				fallbackName = "sonnet-3.5"
+			default:
+				fallbackName = "sonnet-4"
 			}
-		} else if strings.Contains(model, "haiku") {
-			if strings.Contains(model, "3-5") || strings.Contains(model, "3.5") {
-				matchedFamily = "haiku-3.5"
-			} else {
-				matchedFamily = "haiku-3"
+		case strings.Contains(model, "haiku"):
+			switch {
+			case strings.Contains(model, "3-5") || strings.Contains(model, "3.5"):
+				fallbackName = "haiku-3.5"
+			default:
+				fallbackName = "haiku-3"
+			}
+		}
+		if fallbackName != "" {
+			for i := range families {
+				if families[i].name == fallbackName {
+					matched = &families[i]
+					break
+				}
 			}
 		}
 	}

-	if matchedFamily == "" {
+	if matched == nil {
 		return nil
 	}

-	// 在价格数据中查找该系列的模型
-	patterns := familyPatterns[matchedFamily]
-	for _, pattern := range patterns {
+	// Phase 3: 在定价数据中查找该系列的价格
+	lookups := matched.pricing
+	if lookups == nil {
+		lookups = matched.match
+	}
+	for _, pattern := range lookups {
 		for key, pricing := range s.pricingData {
 			keyLower := strings.ToLower(key)
 			if strings.Contains(keyLower, pattern) {
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@ -152,6 +152,11 @@ func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Acc
 			msg := "Credit balance exhausted (400): " + upstreamMsg
 			s.handleAuthError(ctx, account, msg)
 			shouldDisable = true
+		} else if strings.Contains(strings.ToLower(upstreamMsg), "identity verification is required") {
+			// KYC 身份验证要求 → 永久禁用，账号需完成身份验证后才能恢复
+			msg := "Identity verification required (400): " + upstreamMsg
+			s.handleAuthError(ctx, account, msg)
+			shouldDisable = true
 		}
 		// 其他 400 错误（如参数问题）不处理，不禁用账号
 	case 401:
--- a/backend/internal/service/scheduler_snapshot_service.go
+++ b/backend/internal/service/scheduler_snapshot_service.go
@ -20,6 +20,14 @@ var (

 const outboxEventTimeout = 2 * time.Minute

+// batchSeenKey tracks which (groupID, platform) bucket sets have already been
+// rebuilt within a single pollOutbox call, to avoid redundant work when multiple
+// account_changed events share the same groups.
+type batchSeenKey struct {
+	groupID  int64
+	platform string
+}
+
 type SchedulerSnapshotService struct {
 	cache         SchedulerCache
 	outboxRepo    SchedulerOutboxRepository
@ -244,9 +252,10 @@ func (s *SchedulerSnapshotService) pollOutbox() {
 	}

 	watermarkForCheck := watermark
+	seen := make(map[batchSeenKey]struct{})
 	for _, event := range events {
 		eventCtx, cancel := context.WithTimeout(context.Background(), outboxEventTimeout)
-		err := s.handleOutboxEvent(eventCtx, event)
+		err := s.handleOutboxEvent(eventCtx, event, seen)
 		cancel()
 		if err != nil {
 			logger.LegacyPrintf("service.scheduler_snapshot", "[Scheduler] outbox handle failed: id=%d type=%s err=%v", event.ID, event.EventType, err)
@ -255,8 +264,20 @@ func (s *SchedulerSnapshotService) pollOutbox() {
 	}

 	lastID := events[len(events)-1].ID
-	if err := s.cache.SetOutboxWatermark(ctx, lastID); err != nil {
-		logger.LegacyPrintf("service.scheduler_snapshot", "[Scheduler] outbox watermark write failed: %v", err)
+	var wmErr error
+	for i := range 3 {
+		wmCtx, wmCancel := context.WithTimeout(context.Background(), 5*time.Second)
+		wmErr = s.cache.SetOutboxWatermark(wmCtx, lastID)
+		wmCancel()
+		if wmErr == nil {
+			break
+		}
+		if i < 2 {
+			time.Sleep(200 * time.Millisecond)
+		}
+	}
+	if wmErr != nil {
+		logger.LegacyPrintf("service.scheduler_snapshot", "[Scheduler] outbox watermark write failed: %v", wmErr)
 	} else {
 		watermarkForCheck = lastID
 	}
@ -264,18 +285,18 @@ func (s *SchedulerSnapshotService) pollOutbox() {
 	s.checkOutboxLag(ctx, events[0], watermarkForCheck)
 }

-func (s *SchedulerSnapshotService) handleOutboxEvent(ctx context.Context, event SchedulerOutboxEvent) error {
+func (s *SchedulerSnapshotService) handleOutboxEvent(ctx context.Context, event SchedulerOutboxEvent, seen map[batchSeenKey]struct{}) error {
 	switch event.EventType {
 	case SchedulerOutboxEventAccountLastUsed:
 		return s.handleLastUsedEvent(ctx, event.Payload)
 	case SchedulerOutboxEventAccountBulkChanged:
-		return s.handleBulkAccountEvent(ctx, event.Payload)
+		return s.handleBulkAccountEvent(ctx, event.Payload, seen)
 	case SchedulerOutboxEventAccountGroupsChanged:
-		return s.handleAccountEvent(ctx, event.AccountID, event.Payload)
+		return s.handleAccountEvent(ctx, event.AccountID, event.Payload, seen)
 	case SchedulerOutboxEventAccountChanged:
-		return s.handleAccountEvent(ctx, event.AccountID, event.Payload)
+		return s.handleAccountEvent(ctx, event.AccountID, event.Payload, seen)
 	case SchedulerOutboxEventGroupChanged:
-		return s.handleGroupEvent(ctx, event.GroupID)
+		return s.handleGroupEvent(ctx, event.GroupID, seen)
 	case SchedulerOutboxEventFullRebuild:
 		return s.triggerFullRebuild("outbox")
 	default:
@ -309,7 +330,7 @@ func (s *SchedulerSnapshotService) handleLastUsedEvent(ctx context.Context, payl
 	return s.cache.UpdateLastUsed(ctx, updates)
 }

-func (s *SchedulerSnapshotService) handleBulkAccountEvent(ctx context.Context, payload map[string]any) error {
+func (s *SchedulerSnapshotService) handleBulkAccountEvent(ctx context.Context, payload map[string]any, seen map[batchSeenKey]struct{}) error {
 	if payload == nil {
 		return nil
 	}
@ -323,15 +344,15 @@ func (s *SchedulerSnapshotService) handleBulkAccountEvent(ctx context.Context, p
 	}

 	ids := make([]int64, 0, len(rawIDs))
-	seen := make(map[int64]struct{}, len(rawIDs))
+	seenIDs := make(map[int64]struct{}, len(rawIDs))
 	for _, id := range rawIDs {
 		if id <= 0 {
 			continue
 		}
-		if _, exists := seen[id]; exists {
+		if _, exists := seenIDs[id]; exists {
 			continue
 		}
-		seen[id] = struct{}{}
+		seenIDs[id] = struct{}{}
 		ids = append(ids, id)
 	}
 	if len(ids) == 0 {
@ -384,10 +405,10 @@ func (s *SchedulerSnapshotService) handleBulkAccountEvent(ctx context.Context, p
 	for gid := range rebuildGroupSet {
 		rebuildGroupIDs = append(rebuildGroupIDs, gid)
 	}
-	return s.rebuildByGroupIDs(ctx, rebuildGroupIDs, "account_bulk_change")
+	return s.rebuildByGroupIDs(ctx, rebuildGroupIDs, "account_bulk_change", seen)
 }

-func (s *SchedulerSnapshotService) handleAccountEvent(ctx context.Context, accountID *int64, payload map[string]any) error {
+func (s *SchedulerSnapshotService) handleAccountEvent(ctx context.Context, accountID *int64, payload map[string]any, seen map[batchSeenKey]struct{}) error {
 	if accountID == nil || *accountID <= 0 {
 		return nil
 	}
@ -408,7 +429,7 @@ func (s *SchedulerSnapshotService) handleAccountEvent(ctx context.Context, accou
 					return err
 				}
 			}
-			return s.rebuildByGroupIDs(ctx, groupIDs, "account_miss")
+			return s.rebuildByGroupIDs(ctx, groupIDs, "account_miss", seen)
 		}
 		return err
 	}
@ -420,18 +441,18 @@ func (s *SchedulerSnapshotService) handleAccountEvent(ctx context.Context, accou
 	if len(groupIDs) == 0 {
 		groupIDs = account.GroupIDs
 	}
-	return s.rebuildByAccount(ctx, account, groupIDs, "account_change")
+	return s.rebuildByAccount(ctx, account, groupIDs, "account_change", seen)
 }

-func (s *SchedulerSnapshotService) handleGroupEvent(ctx context.Context, groupID *int64) error {
+func (s *SchedulerSnapshotService) handleGroupEvent(ctx context.Context, groupID *int64, seen map[batchSeenKey]struct{}) error {
 	if groupID == nil || *groupID <= 0 {
 		return nil
 	}
 	groupIDs := []int64{*groupID}
-	return s.rebuildByGroupIDs(ctx, groupIDs, "group_change")
+	return s.rebuildByGroupIDs(ctx, groupIDs, "group_change", seen)
 }

-func (s *SchedulerSnapshotService) rebuildByAccount(ctx context.Context, account *Account, groupIDs []int64, reason string) error {
+func (s *SchedulerSnapshotService) rebuildByAccount(ctx context.Context, account *Account, groupIDs []int64, reason string, seen map[batchSeenKey]struct{}) error {
 	if account == nil {
 		return nil
 	}
@ -441,21 +462,21 @@ func (s *SchedulerSnapshotService) rebuildByAccount(ctx context.Context, account
 	}

 	var firstErr error
-	if err := s.rebuildBucketsForPlatform(ctx, account.Platform, groupIDs, reason); err != nil && firstErr == nil {
+	if err := s.rebuildBucketsForPlatform(ctx, account.Platform, groupIDs, reason, seen); err != nil && firstErr == nil {
 		firstErr = err
 	}
 	if account.Platform == PlatformAntigravity && account.IsMixedSchedulingEnabled() {
-		if err := s.rebuildBucketsForPlatform(ctx, PlatformAnthropic, groupIDs, reason); err != nil && firstErr == nil {
+		if err := s.rebuildBucketsForPlatform(ctx, PlatformAnthropic, groupIDs, reason, seen); err != nil && firstErr == nil {
 			firstErr = err
 		}
-		if err := s.rebuildBucketsForPlatform(ctx, PlatformGemini, groupIDs, reason); err != nil && firstErr == nil {
+		if err := s.rebuildBucketsForPlatform(ctx, PlatformGemini, groupIDs, reason, seen); err != nil && firstErr == nil {
 			firstErr = err
 		}
 	}
 	return firstErr
 }

-func (s *SchedulerSnapshotService) rebuildByGroupIDs(ctx context.Context, groupIDs []int64, reason string) error {
+func (s *SchedulerSnapshotService) rebuildByGroupIDs(ctx context.Context, groupIDs []int64, reason string, seen map[batchSeenKey]struct{}) error {
 	groupIDs = s.normalizeGroupIDs(groupIDs)
 	if len(groupIDs) == 0 {
 		return nil
@ -463,19 +484,30 @@ func (s *SchedulerSnapshotService) rebuildByGroupIDs(ctx context.Context, groupI
 	platforms := []string{PlatformAnthropic, PlatformGemini, PlatformOpenAI, PlatformAntigravity}
 	var firstErr error
 	for _, platform := range platforms {
-		if err := s.rebuildBucketsForPlatform(ctx, platform, groupIDs, reason); err != nil && firstErr == nil {
+		if err := s.rebuildBucketsForPlatform(ctx, platform, groupIDs, reason, seen); err != nil && firstErr == nil {
 			firstErr = err
 		}
 	}
 	return firstErr
 }

-func (s *SchedulerSnapshotService) rebuildBucketsForPlatform(ctx context.Context, platform string, groupIDs []int64, reason string) error {
+func (s *SchedulerSnapshotService) rebuildBucketsForPlatform(ctx context.Context, platform string, groupIDs []int64, reason string, seen map[batchSeenKey]struct{}) error {
 	if platform == "" {
 		return nil
 	}
 	var firstErr error
 	for _, gid := range groupIDs {
+		// Within a single poll batch, skip (groupID, platform) pairs that were
+		// already rebuilt. The first rebuild loads fresh DB data for all accounts
+		// in the group, so subsequent rebuilds for the same group+platform within
+		// the same batch are redundant.
+		if seen != nil {
+			key := batchSeenKey{gid, platform}
+			if _, exists := seen[key]; exists {
+				continue
+			}
+			seen[key] = struct{}{}
+		}
 		if err := s.rebuildBucket(ctx, SchedulerBucket{GroupID: gid, Platform: platform, Mode: SchedulerModeSingle}, reason); err != nil && firstErr == nil {
 			firstErr = err
 		}
--- a/backend/internal/service/upstream_response_limit.go
+++ b/backend/internal/service/upstream_response_limit.go
@ -4,8 +4,10 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"net/http"

 	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
 )

 var ErrUpstreamResponseBodyTooLarge = errors.New("upstream response body too large")
@ -36,3 +38,44 @@ func readUpstreamResponseBodyLimited(reader io.Reader, maxBytes int64) ([]byte,
 	}
 	return body, nil
 }
+
+// TooLargeWriter 在响应超限时向客户端写格式化的错误响应。
+type TooLargeWriter func(c *gin.Context)
+
+// ReadUpstreamResponseBody 读取上游非流式响应体。
+// 超限时自动记录 ops error 并调用 onTooLarge 向客户端写错误。
+func ReadUpstreamResponseBody(reader io.Reader, cfg *config.Config, c *gin.Context, onTooLarge TooLargeWriter) ([]byte, error) {
+	maxBytes := resolveUpstreamResponseReadLimit(cfg)
+	body, err := readUpstreamResponseBodyLimited(reader, maxBytes)
+	if err != nil {
+		if errors.Is(err, ErrUpstreamResponseBodyTooLarge) {
+			setOpsUpstreamError(c, http.StatusBadGateway, "upstream response too large", "")
+			if onTooLarge != nil {
+				onTooLarge(c)
+			}
+		}
+		return nil, err
+	}
+	return body, nil
+}
+
+// anthropicTooLargeError 以 Anthropic Messages API 格式写入超限错误。
+func anthropicTooLargeError(c *gin.Context) {
+	c.JSON(http.StatusBadGateway, gin.H{
+		"type": "error",
+		"error": gin.H{
+			"type":    "upstream_error",
+			"message": "Upstream response too large",
+		},
+	})
+}
+
+// openAITooLargeError 以 OpenAI / Gemini 格式写入超限错误。
+func openAITooLargeError(c *gin.Context) {
+	c.JSON(http.StatusBadGateway, gin.H{
+		"error": gin.H{
+			"type":    "upstream_error",
+			"message": "Upstream response too large",
+		},
+	})
+}
--- a/backend/internal/service/upstream_response_limit_test.go
+++ b/backend/internal/service/upstream_response_limit_test.go
@ -4,8 +4,10 @@ import (
 	"bytes"
 	"errors"
 	"testing"
+	"testing/iotest"

 	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
 )

@ -35,3 +37,44 @@ func TestReadUpstreamResponseBodyLimited(t *testing.T) {
 		require.True(t, errors.Is(err, ErrUpstreamResponseBodyTooLarge))
 	})
 }
+
+func TestReadUpstreamResponseBody(t *testing.T) {
+	t.Run("within limit", func(t *testing.T) {
+		body, err := ReadUpstreamResponseBody(bytes.NewReader([]byte("ok")), nil, nil, nil)
+		require.NoError(t, err)
+		require.Equal(t, []byte("ok"), body)
+	})
+
+	t.Run("exceeds limit calls onTooLarge", func(t *testing.T) {
+		cfg := &config.Config{}
+		cfg.Gateway.UpstreamResponseReadMaxBytes = 3
+
+		called := false
+		onTooLarge := func(_ *gin.Context) { called = true }
+
+		body, err := ReadUpstreamResponseBody(bytes.NewReader([]byte("toolong")), cfg, nil, onTooLarge)
+		require.Nil(t, body)
+		require.True(t, errors.Is(err, ErrUpstreamResponseBodyTooLarge))
+		require.True(t, called)
+	})
+
+	t.Run("nil onTooLarge does not panic", func(t *testing.T) {
+		cfg := &config.Config{}
+		cfg.Gateway.UpstreamResponseReadMaxBytes = 3
+
+		body, err := ReadUpstreamResponseBody(bytes.NewReader([]byte("toolong")), cfg, nil, nil)
+		require.Nil(t, body)
+		require.True(t, errors.Is(err, ErrUpstreamResponseBodyTooLarge))
+	})
+
+	t.Run("io error does not call onTooLarge", func(t *testing.T) {
+		called := false
+		onTooLarge := func(_ *gin.Context) { called = true }
+
+		body, err := ReadUpstreamResponseBody(iotest.ErrReader(errors.New("disk failure")), nil, nil, onTooLarge)
+		require.Nil(t, body)
+		require.Error(t, err)
+		require.False(t, errors.Is(err, ErrUpstreamResponseBodyTooLarge))
+		require.False(t, called)
+	})
+}
--- a/deploy/docker-compose.yml
+++ b/deploy/docker-compose.yml
@ -1,57 +1,82 @@
 # =============================================================================
-# Sub2API Docker Compose Configuration
+# Sub2API Docker Compose Configuration (负载均衡版)
 # =============================================================================
 # Quick Start:
 #   1. Copy .env.example to .env and configure
-#   2. docker-compose up -d
-#   3. Check logs: docker-compose logs -f sub2api
-#   4. Access: http://localhost:8080
+#   2. docker compose up -d
+#   3. Check logs: docker compose logs -f
+#   4. Access: http://localhost (via nginx)
 #
-# All configuration is done via environment variables.
-# No Setup Wizard needed - the system auto-initializes on first run.
+# 扩缩容:
+#   docker compose up -d --scale sub2api=5   # 扩到 5 个实例
+#   docker compose up -d --scale sub2api=2   # 缩回 2 个实例
+#
+# 注意事项:
+#   - JWT_SECRET / TOTP_ENCRYPTION_KEY 必须固定，多实例共享同一个值
+#   - PostgreSQL / Redis 单实例，不参与水平扩展
+#   - postgres 端口默认不对外暴露，如需调试取消注释 127.0.0.1:5433:5432
+#   - redis 端口默认不对外暴露，如需调试取消注释 127.0.0.1:6380:6379
 # =============================================================================

 services:
  # ===========================================================================
-  # Sub2API Application
+  # Nginx 负载均衡（入口）
+  # ===========================================================================
+  nginx:
+    image: nginx:alpine
+    container_name: sub2api-nginx
+    restart: unless-stopped
+    ulimits:
+      nofile:
+        soft: 65535
+        hard: 65535
+    ports:
+      - "0.0.0.0:80:80"
+      - "0.0.0.0:443:443"
+    volumes:
+      - ./nginx/nginx.conf:/etc/nginx/nginx.conf:ro
+      - ./nginx/certs:/etc/nginx/certs:ro
+    depends_on:
+      sub2api:
+        condition: service_healthy
+    networks:
+      - sub2api-network
+    healthcheck:
+      test: [ "CMD", "wget", "-q", "-T", "3", "-O", "/dev/null", "http://localhost/health" ]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 10s
+
+  # ===========================================================================
+  # Sub2API Application（多实例，通过 --scale 控制数量）
  # ===========================================================================
  sub2api:
-    image: weishaw/sub2api:latest
-    container_name: sub2api
+    image: docker.io/zfc931912343/sub2api:latest
    restart: unless-stopped
    ulimits:
      nofile:
        soft: 100000
        hard: 100000
-    ports:
-      - "${BIND_HOST:-0.0.0.0}:${SERVER_PORT:-8080}:8080"
+    # 不直接暴露端口，由 nginx 代理
+    expose:
+      - "8080"
    volumes:
-      # Data persistence (config.yaml will be auto-generated here)
      - sub2api_data:/app/data
-      # Optional: Mount custom config.yaml (uncomment and create the file first)
-      # Copy config.example.yaml to config.yaml, modify it, then uncomment:
+      # Optional: 挂载自定义 config.yaml（先从 config.example.yaml 复制并修改）
      # - ./config.yaml:/app/data/config.yaml
-      # Optional: Mount a custom Codex instructions template file, then point
-      # gateway.forced_codex_instructions_template_file at /app/data/codex-instructions.md.tmpl
-      # in config.yaml.
+      # Optional: 自定义 Codex instructions 模板
      # - ./codex-instructions.md.tmpl:/app/data/codex-instructions.md.tmpl:ro
    environment:
-      # =======================================================================
-      # Auto Setup (REQUIRED for Docker deployment)
-      # =======================================================================
      - AUTO_SETUP=true

-      # =======================================================================
-      # Server Configuration
-      # =======================================================================
+      # --- Server ---
      - SERVER_HOST=0.0.0.0
      - SERVER_PORT=8080
      - SERVER_MODE=${SERVER_MODE:-release}
      - RUN_MODE=${RUN_MODE:-standard}

-      # =======================================================================
-      # Database Configuration (PostgreSQL)
-      # =======================================================================
+      # --- Database (PostgreSQL) ---
      - DATABASE_HOST=postgres
      - DATABASE_PORT=5432
      - DATABASE_USER=${POSTGRES_USER:-sub2api}
@ -63,9 +88,7 @@ services:
      - DATABASE_CONN_MAX_LIFETIME_MINUTES=${DATABASE_CONN_MAX_LIFETIME_MINUTES:-30}
      - DATABASE_CONN_MAX_IDLE_TIME_MINUTES=${DATABASE_CONN_MAX_IDLE_TIME_MINUTES:-5}

-      # =======================================================================
-      # Redis Configuration
-      # =======================================================================
+      # --- Redis ---
      - REDIS_HOST=redis
      - REDIS_PORT=6379
      - REDIS_PASSWORD=${REDIS_PASSWORD:-}
@ -74,74 +97,39 @@ services:
      - REDIS_MIN_IDLE_CONNS=${REDIS_MIN_IDLE_CONNS:-10}
      - REDIS_ENABLE_TLS=${REDIS_ENABLE_TLS:-false}

-      # =======================================================================
-      # Admin Account (auto-created on first run)
-      # =======================================================================
+      # --- Admin（仅首次启动生效）---
      - ADMIN_EMAIL=${ADMIN_EMAIL:-admin@sub2api.local}
      - ADMIN_PASSWORD=${ADMIN_PASSWORD:-}

-      # =======================================================================
-      # JWT Configuration
-      # =======================================================================
-      # IMPORTANT: Set a fixed JWT_SECRET to prevent login sessions from being
-      # invalidated after container restarts. If left empty, a random secret
-      # will be generated on each startup.
-      # Generate a secure secret: openssl rand -hex 32
-      - JWT_SECRET=${JWT_SECRET:-}
+      # --- JWT（多实例必须固定，否则重启后 session 失效）---
+      # 生成: openssl rand -hex 32
+      - JWT_SECRET=${JWT_SECRET:?JWT_SECRET is required for multi-instance}
      - JWT_EXPIRE_HOUR=${JWT_EXPIRE_HOUR:-24}

-      # =======================================================================
-      # TOTP (2FA) Configuration
-      # =======================================================================
-      # IMPORTANT: Set a fixed encryption key for TOTP secrets. If left empty,
-      # a random key will be generated on each startup, causing all existing
-      # TOTP configurations to become invalid (users won't be able to login
-      # with 2FA).
-      # Generate a secure key: openssl rand -hex 32
-      - TOTP_ENCRYPTION_KEY=${TOTP_ENCRYPTION_KEY:-}
+      # --- TOTP 2FA（多实例必须固定，否则 2FA 失效）---
+      # 生成: openssl rand -hex 32
+      - TOTP_ENCRYPTION_KEY=${TOTP_ENCRYPTION_KEY:?TOTP_ENCRYPTION_KEY is required for multi-instance}

-      # =======================================================================
-      # Timezone Configuration
-      # This affects ALL time operations in the application:
-      # - Database timestamps
-      # - Usage statistics "today" boundary
-      # - Subscription expiry times
-      # - Log timestamps
-      # Common values: Asia/Shanghai, America/New_York, Europe/London, UTC
-      # =======================================================================
+      # --- Timezone ---
      - TZ=${TZ:-Asia/Shanghai}

-      # =======================================================================
-      # Gemini OAuth Configuration (for Gemini accounts)
-      # =======================================================================
+      # --- Gemini OAuth ---
      - GEMINI_OAUTH_CLIENT_ID=${GEMINI_OAUTH_CLIENT_ID:-}
      - GEMINI_OAUTH_CLIENT_SECRET=${GEMINI_OAUTH_CLIENT_SECRET:-}
      - GEMINI_OAUTH_SCOPES=${GEMINI_OAUTH_SCOPES:-}
      - GEMINI_QUOTA_POLICY=${GEMINI_QUOTA_POLICY:-}
-
-      # Built-in OAuth client secrets (optional)
-      # SECURITY: This repo does not embed third-party client_secret.
      - GEMINI_CLI_OAUTH_CLIENT_SECRET=${GEMINI_CLI_OAUTH_CLIENT_SECRET:-}
      - ANTIGRAVITY_OAUTH_CLIENT_SECRET=${ANTIGRAVITY_OAUTH_CLIENT_SECRET:-}

-      # =======================================================================
-      # Security Configuration (URL Allowlist)
-      # =======================================================================
-      # Enable URL allowlist validation (false to skip allowlist checks)
+      # --- Security ---
      - SECURITY_URL_ALLOWLIST_ENABLED=${SECURITY_URL_ALLOWLIST_ENABLED:-false}
-      # Allow insecure HTTP URLs when allowlist is disabled (default: false, requires https)
      - SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP=${SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP:-false}
-      # Allow private IP addresses for upstream/pricing/CRS (for internal deployments)
      - SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS=${SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS:-false}
-      # Upstream hosts whitelist (comma-separated, only used when enabled=true)
      - SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS=${SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS:-}

-      # =======================================================================
-      # Update Configuration (在线更新配置)
-      # =======================================================================
-      # Proxy for accessing GitHub (online updates + pricing data)
-      # Examples: http://host:port, socks5://host:port
+      # --- Update Proxy（国内机器可配置代理访问 GitHub）---
      - UPDATE_PROXY_URL=${UPDATE_PROXY_URL:-}
+
    depends_on:
      postgres:
        condition: service_healthy
@ -150,17 +138,7 @@ services:
    networks:
      - sub2api-network
    healthcheck:
-      test:
-        [
-          "CMD",
-          "wget",
-          "-q",
-          "-T",
-          "5",
-          "-O",
-          "/dev/null",
-          "http://localhost:8080/health",
-        ]
+      test: [ "CMD", "wget", "-q", "-T", "5", "-O", "/dev/null", "http://localhost:8080/health" ]
      interval: 30s
      timeout: 10s
      retries: 3
@ -180,9 +158,7 @@ services:
    volumes:
      - postgres_data:/var/lib/postgresql/data
    environment:
-      # postgres:18-alpine 默认 PGDATA=/var/lib/postgresql/18/docker（位于镜像声明的匿名卷 /var/lib/postgresql 内）。
-      # 若不显式设置 PGDATA，则即使挂载了 postgres_data 到 /var/lib/postgresql/data，数据也不会落盘到该命名卷，
-      # docker compose down/up 后会触发 initdb 重新初始化，导致用户/密码等数据丢失。
+      # postgres:18-alpine 默认 PGDATA 在镜像内部匿名卷，必须显式指定才能持久化到命名卷
      - PGDATA=/var/lib/postgresql/data
      - POSTGRES_USER=${POSTGRES_USER:-sub2api}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
@ -191,19 +167,14 @@ services:
    networks:
      - sub2api-network
    healthcheck:
-      test:
-        [
-          "CMD-SHELL",
-          "pg_isready -U ${POSTGRES_USER:-sub2api} -d ${POSTGRES_DB:-sub2api}",
-        ]
+      test: [ "CMD-SHELL", "pg_isready -U ${POSTGRES_USER:-sub2api} -d ${POSTGRES_DB:-sub2api}" ]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s
-    ports:
-      - 5432:5432
-    # 注意：不暴露端口到宿主机，应用通过内部网络连接
-    # 如需调试，可临时添加：ports: ["127.0.0.1:5433:5432"]
+    # 默认不对外暴露，如需本地调试取消注释
+    # ports:
+    #   - "127.0.0.1:5433:5432"

  # ===========================================================================
  # Redis Cache
@ -227,18 +198,19 @@ services:
        ${REDIS_PASSWORD:+--requirepass "$REDIS_PASSWORD"}'
    environment:
      - TZ=${TZ:-Asia/Shanghai}
-      # REDISCLI_AUTH is used by redis-cli for authentication (safer than -a flag)
      - REDISCLI_AUTH=${REDIS_PASSWORD:-}
    networks:
      - sub2api-network
    healthcheck:
-      test: ["CMD", "redis-cli", "ping"]
+      test: [ "CMD", "redis-cli", "ping" ]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 5s
-    ports:
-      - 6379:6379
+    # 默认不对外暴露，如需本地调试取消注释
+    # ports:
+    #   - "127.0.0.1:6380:6379"
+
 # =============================================================================
 # Volumes
 # =============================================================================
--- a/deploy/nginx/nginx.conf
+++ b/deploy/nginx/nginx.conf
@ -0,0 +1,155 @@
+# =============================================================================
+# Nginx 负载均衡配置 for Sub2API（Cloudflare 前置 + Origin Certificate）
+# =============================================================================
+# 架构：Cloudflare (HTTPS) → VPS:443 (Docker nginx + CF Origin Cert) → sub2api
+# - SSL 由 Cloudflare Origin Certificate 端到端加密
+# - 真实客户端 IP 从 CF-Connecting-IP 头还原
+# =============================================================================
+
+worker_processes  auto;
+error_log  /var/log/nginx/error.log warn;
+pid        /var/run/nginx.pid;
+
+events {
+    worker_connections  4096;
+    use epoll;
+    multi_accept on;
+}
+
+http {
+    include       /etc/nginx/mime.types;
+    default_type  application/octet-stream;
+
+    log_format  main  '$realip_remote_addr - $remote_user [$time_local] "$request" '
+                      '$status $body_bytes_sent '
+                      'rt=$request_time urt=$upstream_response_time cf_ray=$http_cf_ray';
+    access_log  /var/log/nginx/access.log  main;
+
+    sendfile        on;
+    tcp_nopush      on;
+    tcp_nodelay     on;
+    keepalive_timeout  75s;
+    keepalive_requests 1000;
+    server_tokens  off;
+
+    client_max_body_size       64m;
+    client_body_buffer_size    2m;   # 避免请求体落盘（默认 8k/16k 对 AI 请求太小）
+
+    gzip on;
+    gzip_vary on;
+    gzip_proxied any;
+    gzip_comp_level 4;
+    gzip_min_length 1024;
+    gzip_types text/plain text/css application/json application/javascript
+               text/xml application/xml text/javascript;
+
+    # -------------------------------------------------------------------------
+    # 还原真实客户端 IP（Cloudflare IP 段）
+    # -------------------------------------------------------------------------
+    set_real_ip_from 103.21.244.0/22;
+    set_real_ip_from 103.22.200.0/22;
+    set_real_ip_from 103.31.4.0/22;
+    set_real_ip_from 104.16.0.0/13;
+    set_real_ip_from 104.24.0.0/14;
+    set_real_ip_from 108.162.192.0/18;
+    set_real_ip_from 131.0.72.0/22;
+    set_real_ip_from 141.101.64.0/18;
+    set_real_ip_from 162.158.0.0/15;
+    set_real_ip_from 172.64.0.0/13;
+    set_real_ip_from 173.245.48.0/20;
+    set_real_ip_from 188.114.96.0/20;
+    set_real_ip_from 190.93.240.0/20;
+    set_real_ip_from 197.234.240.0/22;
+    set_real_ip_from 198.41.128.0/17;
+    set_real_ip_from 2400:cb00::/32;
+    set_real_ip_from 2606:4700::/32;
+    set_real_ip_from 2803:f800::/32;
+    set_real_ip_from 2405:b500::/32;
+    set_real_ip_from 2405:8100::/32;
+    set_real_ip_from 2a06:98c0::/29;
+    set_real_ip_from 2c0f:f248::/32;
+    real_ip_header    CF-Connecting-IP;
+    real_ip_recursive on;
+
+    # WebSocket upgrade 映射
+    map $http_upgrade $connection_upgrade {
+        default upgrade;
+        ''      close;
+    }
+
+    # -------------------------------------------------------------------------
+    # 上游实例池（least_conn）
+    # -------------------------------------------------------------------------
+    upstream sub2api_backend {
+        # ip_hash 保证同一客户端 IP 始终路由到同一实例
+        # 解决 OAuth session 内存不共享问题（session 在实例内存中）
+        ip_hash;
+        server sub2api:8080;
+    }
+
+    # =========================================================================
+    # HTTP → HTTPS 跳转
+    # =========================================================================
+    server {
+        listen 80;
+        server_name _;
+
+        location = /health {
+            proxy_pass            http://sub2api_backend;
+            proxy_set_header      Host $host;
+            proxy_connect_timeout 5s;
+            proxy_read_timeout    10s;
+            access_log            off;
+        }
+
+        location / {
+            return 301 https://$host$request_uri;
+        }
+    }
+
+    # =========================================================================
+    # HTTPS（Cloudflare Origin Certificate）
+    # =========================================================================
+    server {
+        listen 443 ssl;
+        http2  on;
+        server_name _;
+
+        ssl_certificate     /etc/nginx/certs/fullchain.pem;
+        ssl_certificate_key /etc/nginx/certs/privkey.pem;
+
+        ssl_protocols       TLSv1.2 TLSv1.3;
+        ssl_ciphers         ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305;
+        ssl_prefer_server_ciphers off;
+        ssl_session_cache   shared:SSL:10m;
+        ssl_session_timeout 1d;
+
+        location = /health {
+            proxy_pass            http://sub2api_backend;
+            proxy_set_header      Host $host;
+            proxy_connect_timeout 5s;
+            proxy_read_timeout    10s;
+            access_log            off;
+        }
+
+        location / {
+            proxy_pass         http://sub2api_backend;
+
+            proxy_http_version 1.1;
+            proxy_set_header   Upgrade           $http_upgrade;
+            proxy_set_header   Connection        $connection_upgrade;
+
+            proxy_set_header   Host              $host;
+            proxy_set_header   X-Real-IP         $remote_addr;
+            proxy_set_header   X-Forwarded-For   $proxy_add_x_forwarded_for;
+            proxy_set_header   X-Forwarded-Proto $http_x_forwarded_proto;
+
+            proxy_buffering    off;
+            proxy_cache        off;
+
+            proxy_connect_timeout  10s;
+            proxy_send_timeout     600s;
+            proxy_read_timeout     600s;
+        }
+    }
+}
--- a/frontend/src/composables/useModelWhitelist.ts
+++ b/frontend/src/composables/useModelWhitelist.ts
@ -43,6 +43,7 @@ export const claudeModels = [
  'claude-sonnet-4-5-20250929', 'claude-haiku-4-5-20251001',
  'claude-opus-4-5-20251101',
  'claude-opus-4-6',
+  'claude-opus-4-7',
  'claude-sonnet-4-6',
  'claude-2.1', 'claude-2.0', 'claude-instant-1.2'
 ]
@ -66,6 +67,7 @@ const antigravityModels = [
  // Claude 4.5+ 系列
  'claude-opus-4-6',
  'claude-opus-4-6-thinking',
+  'claude-opus-4-7',
  'claude-opus-4-5-thinking',
  'claude-sonnet-4-6',
  'claude-sonnet-4-5',
@ -250,6 +252,7 @@ const anthropicPresetMappings = [
  { label: 'Sonnet 4.6', from: 'claude-sonnet-4-6', to: 'claude-sonnet-4-6', color: 'bg-indigo-100 text-indigo-700 hover:bg-indigo-200 dark:bg-indigo-900/30 dark:text-indigo-400' },
  { label: 'Opus 4.5', from: 'claude-opus-4-5-20251101', to: 'claude-opus-4-5-20251101', color: 'bg-purple-100 text-purple-700 hover:bg-purple-200 dark:bg-purple-900/30 dark:text-purple-400' },
  { label: 'Opus 4.6', from: 'claude-opus-4-6', to: 'claude-opus-4-6', color: 'bg-purple-100 text-purple-700 hover:bg-purple-200 dark:bg-purple-900/30 dark:text-purple-400' },
+  { label: 'Opus 4.7', from: 'claude-opus-4-7', to: 'claude-opus-4-7', color: 'bg-purple-100 text-purple-700 hover:bg-purple-200 dark:bg-purple-900/30 dark:text-purple-400' },
  { label: 'Haiku 3.5', from: 'claude-3-5-haiku-20241022', to: 'claude-3-5-haiku-20241022', color: 'bg-green-100 text-green-700 hover:bg-green-200 dark:bg-green-900/30 dark:text-green-400' },
  { label: 'Haiku 4.5', from: 'claude-haiku-4-5-20251001', to: 'claude-haiku-4-5-20251001', color: 'bg-emerald-100 text-emerald-700 hover:bg-emerald-200 dark:bg-emerald-900/30 dark:text-emerald-400' },
  { label: 'Opus->Sonnet', from: 'claude-opus-4-6', to: 'claude-sonnet-4-5-20250929', color: 'bg-amber-100 text-amber-700 hover:bg-amber-200 dark:bg-amber-900/30 dark:text-amber-400' }
@ -309,12 +312,14 @@ const antigravityPresetMappings = [
  { label: 'Sonnet 4.6', from: 'claude-sonnet-4-6', to: 'claude-sonnet-4-6', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' },
  { label: 'Sonnet 4.5', from: 'claude-sonnet-4-5', to: 'claude-sonnet-4-5', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' },
  { label: 'Opus 4.6', from: 'claude-opus-4-6', to: 'claude-opus-4-6-thinking', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' },
-  { label: 'Opus 4.6-thinking', from: 'claude-opus-4-6-thinking', to: 'claude-opus-4-6-thinking', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' }
+  { label: 'Opus 4.6-thinking', from: 'claude-opus-4-6-thinking', to: 'claude-opus-4-6-thinking', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' },
+  { label: 'Opus 4.7', from: 'claude-opus-4-7', to: 'claude-opus-4-7', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' }
 ]

 // Bedrock 预设映射（与后端 DefaultBedrockModelMapping 保持一致）
 const bedrockPresetMappings = [
  { label: 'Opus 4.6', from: 'claude-opus-4-6', to: 'us.anthropic.claude-opus-4-6-v1', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' },
+  { label: 'Opus 4.7', from: 'claude-opus-4-7', to: 'us.anthropic.claude-opus-4-7-v1', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' },
  { label: 'Sonnet 4.6', from: 'claude-sonnet-4-6', to: 'us.anthropic.claude-sonnet-4-6', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' },
  { label: 'Opus 4.5', from: 'claude-opus-4-5-thinking', to: 'us.anthropic.claude-opus-4-5-20251101-v1:0', color: 'bg-pink-100 text-pink-700 hover:bg-pink-200 dark:bg-pink-900/30 dark:text-pink-400' },
  { label: 'Sonnet 4.5', from: 'claude-sonnet-4-5', to: 'us.anthropic.claude-sonnet-4-5-20250929-v1:0', color: 'bg-cyan-100 text-cyan-700 hover:bg-cyan-200 dark:bg-cyan-900/30 dark:text-cyan-400' },
Author	SHA1	Message	Date
win	888b7eeb21	feat: add opus-4-7 support + nginx load balancer docker-compose Some checks failed CI / test (push) Failing after 3s Details CI / golangci-lint (push) Failing after 3s Details Security Scan / backend-security (push) Failing after 2s Details Security Scan / frontend-security (push) Failing after 2m0s Details	2026-04-17 11:43:38 +08:00
win	56a955e1d6	chore: merge upstream v0.1.114, keep Antigravity customizations	2026-04-17 11:42:27 +08:00
github-actions[bot]	6cfdf4ec05	chore: sync VERSION to 0.1.114 [skip ci]	2026-04-17 02:51:18 +00:00
Wesley Liddick	358ff6a608	Merge pull request #1683 from FjlI5/dev-main fix：修复上游账号为OpenAI API key时Claude Code调用缓存率低的问题	2026-04-17 10:28:12 +08:00
Wesley Liddick	41fbdba104	Merge pull request #1687 from touwaeriol/refactor/upstream-response-limit-dedup refactor: extract ReadUpstreamResponseBody to deduplicate response read + too-large handling	2026-04-17 10:19:14 +08:00
Wesley Liddick	c22d11cedd	Merge pull request #1702 from StarryKira/fix/outbox-watermark-context-dedup-1691 fix: fix outbox watermark context expiry and add in-batch group rebuild dedup	2026-04-17 10:18:56 +08:00
shaw	5d586a9f3a	fix: 上游返回 KYC 身份验证要求时停止账号调度	2026-04-17 10:17:50 +08:00
shaw	a789c8c4c7	feat: 支持opus-4.7	2026-04-17 09:37:25 +08:00
Elysia	697c41a3f6	fix: create fresh context per watermark write retry attempt Each retry in the SetOutboxWatermark loop now gets its own 5s context. Previously a shared context could already be expired when the second or third attempt ran, making the retries pointless. Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>	2026-04-16 20:41:40 +08:00
Elysia	e44baa1094	fix: fix outbox watermark context expiry and add in-batch group rebuild dedup Fixes #1691 - pollOutbox() reused a 10s context for SetOutboxWatermark after event processing could take much longer, causing "outbox watermark write failed: context deadline exceeded". The watermark never advanced so the same 200 events were reprocessed every poll cycle, spiking CPU. Now uses an independent 5s context with up to 3 retries (200ms apart). - When multiple Codex accounts sharing the same 21-22 groups are all rate-limited in quick succession, each account_changed event triggered redundant bucket rebuild attempts for the same groups. Introduce batchSeenKey{groupID, platform} and thread a seen map through the handler chain; rebuildBucketsForPlatform skips (group, platform) pairs already rebuilt within the same poll batch (~80% fewer rebuild calls in the 5-accounts-same-groups scenario). Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>	2026-04-16 19:09:40 +08:00
erio	10699eeb34	refactor: extract ReadUpstreamResponseBody to deduplicate upstream response read + too-large error handling Consolidates 9 call sites of resolveUpstreamResponseReadLimit + readUpstreamResponseBodyLimited + ErrUpstreamResponseBodyTooLarge error handling into a single ReadUpstreamResponseBody function with TooLargeWriter callback for API-format-specific error responses (Anthropic, OpenAI, countTokens).	2026-04-16 01:53:22 +08:00
fjl5	6c89d8d35c	add prompt_cache_key injection for messages→responses	2026-04-15 23:56:56 +08:00
 @ -1 +1 @@
 .1.113
 .1.114