AI网关新增xAI格式TTS端点与过滤弃用模型开关

- POST /ai/v1/tts:xAI 官方 TTS 格式转换层(text/voice_id→input/voice),复用 Speech 编排,model 缺省 xai.grok-tts;实测 output_format/speed 透传生效
- 「过滤弃用模型」开关(GET/PUT /api/v1/ai-settings,settings 表持久化):开启后已宣布弃用(未退役)模型从列表与路由排除;同步入库与退役提醒不受影响
- AI 网关文档更名 docs/AI网关.md,补 tts 矩阵段与开关说明;README 引用同步
- CHANGELOG 0.5.0(0.4.0 已发版冻结);DASH_VERSION v0.5.0

internal/aiwire/moderations.go

@@ -43,3 +43,21 @@ type SpeechRequest struct {
 	ResponseFormat string `json:"response_format,omitempty"`
 	Language       string `json:"language,omitempty"`
 }
+
+// TtsRequest 描述 /ai/v1/tts 请求体的已知字段(xAI 官方 TTS 格式,文档用途)。
+// model 为网关扩展字段(缺省 xai.grok-tts);未列字段原样透传上游。
+type TtsRequest struct {
+	Model        string           `json:"model,omitempty"`
+	Text         string           `json:"text"`
+	Language     string           `json:"language"`
+	VoiceID      string           `json:"voice_id,omitempty"`
+	OutputFormat *TtsOutputFormat `json:"output_format,omitempty"`
+	Speed        float64          `json:"speed,omitempty"`
+}
+
+// TtsOutputFormat 是 xAI TTS 输出格式配置(缺省 MP3 24kHz/128kbps)。
+type TtsOutputFormat struct {
+	Codec      string `json:"codec,omitempty"`
+	SampleRate int    `json:"sample_rate,omitempty"`
+	BitRate    int    `json:"bit_rate,omitempty"`
+}

internal/api/aiadmin.go

@@ -372,3 +372,40 @@ func aiPathID(c *gin.Context) (uint, bool) {
 	}
 	return uint(id), true
 }
+
+// aiSettingsResponse 是 AI 网关全局设置(文档与响应共用)。
+type aiSettingsResponse struct {
+	FilterDeprecated bool `json:"filterDeprecated"`
+}
+
+// aiSettings 返回 AI 网关全局设置。
+//
+// @Summary  AI 网关全局设置
+// @Tags     AI 管理
+// @Success  200 {object} aiSettingsResponse
+// @Security BearerAuth
+// @Router   /api/v1/ai-settings [get]
+func (h *aiAdminHandler) aiSettings(c *gin.Context) {
+	c.JSON(http.StatusOK, aiSettingsResponse{FilterDeprecated: h.gw.FilterDeprecated()})
+}
+
+// updateAiSettings 更新 AI 网关全局设置(当前仅「过滤弃用模型」开关)。
+//
+// @Summary  更新 AI 网关全局设置
+// @Tags     AI 管理
+// @Param    body body aiSettingsResponse true "开启后已宣布弃用(即使未退役)的模型从列表与路由中排除"
+// @Success  200 {object} aiSettingsResponse
+// @Security BearerAuth
+// @Router   /api/v1/ai-settings [put]
+func (h *aiAdminHandler) updateAiSettings(c *gin.Context) {
+	var req aiSettingsResponse
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+		return
+	}
+	if err := h.gw.SetFilterDeprecated(c.Request.Context(), req.FilterDeprecated); err != nil {
+		respondError(c, err)
+		return
+	}
+	c.JSON(http.StatusOK, aiSettingsResponse{FilterDeprecated: h.gw.FilterDeprecated()})
+}

internal/api/aigateway_extras.go

@@ -29,12 +29,38 @@ func (h *aiGatewayHandler) audioSpeech(c *gin.Context) {
 		aiError(c, http.StatusBadRequest, "invalid_request_error", err.Error())
 		return
 	}
+	h.speechRespond(c, "speech", modelName, raw, body)
+}
+
+// tts 是 xAI 官方格式 TTS 端点(/ai/v1/tts),转换为上游 OpenAI 兼容形态后复用 Speech 编排。
+//
+// @Summary  xAI 官方格式文本转语音端点
+// @Tags     AI 网关
+// @Param    body body aiwire.TtsRequest true "xAI TTS 请求体(text/language 必填,voice_id 缺省 eve;model 为网关扩展,缺省 xai.grok-tts;未列字段原样透传)"
+// @Success  200 {file} binary "音频字节(Content-Type 透传上游,默认 audio/mpeg)"
+// @Router   /ai/v1/tts [post]
+func (h *aiGatewayHandler) tts(c *gin.Context) {
+	raw, err := c.GetRawData()
+	if err != nil {
+		aiError(c, http.StatusBadRequest, "invalid_request_error", err.Error())
+		return
+	}
+	modelName, body, err := service.TtsBodyConvert(raw)
+	if err != nil {
+		aiError(c, http.StatusBadRequest, "invalid_request_error", err.Error())
+		return
+	}
+	h.speechRespond(c, "tts", modelName, raw, body)
+}
+
+// speechRespond 是 audioSpeech / tts 的公共主体:白名单校验、上游调用、日志与音频响应。
+func (h *aiGatewayHandler) speechRespond(c *gin.Context, endpoint, modelName string, raw, body []byte) {
 	if !checkKeyModel(c, modelName) {
 		return
 	}
 	start := time.Now()
 	audio, contentType, meta, err := h.gw.Speech(c.Request.Context(), modelName, body, keyGroup(c))
-	entry := h.logEntry(c, "speech", modelName, false, meta, start)
+	entry := h.logEntry(c, endpoint, modelName, false, meta, start)
 	if err != nil {
 		upstreamError(c, err)
 		entry.ErrMsg = err.Error()
@@ -43,7 +69,7 @@ func (h *aiGatewayHandler) audioSpeech(c *gin.Context) {
 	}
 	entry.Status = http.StatusOK
 	callID := h.gw.LogCall(entry)
-	h.maybeLogContent(c, callID, "speech", modelName, false, string(raw), nil)
+	h.maybeLogContent(c, callID, endpoint, modelName, false, string(raw), nil)
 	if contentType == "" {
 		contentType = "audio/mpeg"
 	}

internal/api/aigateway_test.go

@@ -103,6 +103,10 @@ func TestAiGatewayKeyModelRestrict(t *testing.T) {
 			`{"model":"ghost-model","query":"q"}`, 400, []string{"invalid_request_error"}},
 		{"moderations 空 input 拒绝", "open-key-12345", "/ai/v1/moderations",
 			`{"input":[]}`, 400, []string{"invalid_request_error"}},
+		{"tts 缺 language 拒绝", "open-key-12345", "/ai/v1/tts",
+			`{"text":"你好"}`, 400, []string{"invalid_request_error"}},
+		{"tts 白名单外拦截", "limited-key-1234", "/ai/v1/tts",
+			`{"model":"meta.llama-3.3-70b-instruct","text":"你好","language":"zh"}`, 404, deny},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {

internal/api/routes_ai.go

@@ -16,6 +16,7 @@ func registerAiGateway(r *gin.Engine, aiGateway *service.AiGatewayService) {
 	ai.POST("/messages", aih.messages)
 	ai.POST("/embeddings", aih.embeddings)
 	ai.POST("/audio/speech", aih.audioSpeech)
+	ai.POST("/tts", aih.tts)
 	ai.POST("/rerank", aih.rerank)
 	ai.POST("/moderations", aih.moderations)
 	ai.GET("/models", aih.listModels)
@@ -36,6 +37,8 @@ func registerAiAdmin(secured *gin.RouterGroup, aiGateway *service.AiGatewayServi
 	secured.POST("/ai-channels/:id/probe", aiadmin.probeChannel)
 	secured.POST("/ai-channels/:id/sync-models", aiadmin.syncChannelModels)
 	secured.GET("/ai-models", aiadmin.gatewayModels)
+	secured.GET("/ai-settings", aiadmin.aiSettings)
+	secured.PUT("/ai-settings", aiadmin.updateAiSettings)
 	secured.GET("/ai-blacklist", aiadmin.listBlacklist)
 	secured.POST("/ai-blacklist", aiadmin.addBlacklist)
 	secured.DELETE("/ai-blacklist/:id", aiadmin.removeBlacklist)

internal/service/aigateway.go

@@ -12,6 +12,7 @@ import (
 	"sort"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"gorm.io/gorm"
@@ -58,11 +59,39 @@ type AiGatewayService struct {
 	lastTouch map[uint]time.Time
 	// onChannelsChanged 在渠道增删后触发,由 main 装配为探测任务同步钩子
 	onChannelsChanged func(context.Context)
+	// filterDeprecated 是「过滤弃用模型」开关(内存镜像,持久化在 settings 表)
+	filterDeprecated atomic.Bool
 }
 
 // NewAiGatewayService 组装依赖;调用 StartCleanup 后开始调用日志周期清理。
 func NewAiGatewayService(db *gorm.DB, configs *OciConfigService, client oci.Client) *AiGatewayService {
-	return &AiGatewayService{db: db, configs: configs, client: client, lastTouch: map[uint]time.Time{}}
+	s := &AiGatewayService{db: db, configs: configs, client: client, lastTouch: map[uint]time.Time{}}
+	var row model.Setting
+	if err := db.Where("key = ?", settingAiFilterDeprecated).First(&row).Error; err == nil {
+		s.filterDeprecated.Store(row.Value == "1")
+	}
+	return s
+}
+
+// settingAiFilterDeprecated 是「过滤弃用模型」开关的配置键,值 "1"/"0",缺省关闭。
+const settingAiFilterDeprecated = "ai_filter_deprecated"
+
+// FilterDeprecated 返回「过滤弃用模型」开关状态。
+func (s *AiGatewayService) FilterDeprecated() bool { return s.filterDeprecated.Load() }
+
+// SetFilterDeprecated 持久化并即时生效开关:开启后已宣布弃用
+// (deprecated_at 非空,即使未退役)的模型从列表与路由中排除。
+func (s *AiGatewayService) SetFilterDeprecated(ctx context.Context, on bool) error {
+	value := "0"
+	if on {
+		value = "1"
+	}
+	err := s.db.WithContext(ctx).Save(&model.Setting{Key: settingAiFilterDeprecated, Value: value}).Error
+	if err != nil {
+		return fmt.Errorf("保存过滤弃用模型开关: %w", err)
+	}
+	s.filterDeprecated.Store(on)
+	return nil
 }
 
 // SetOnChannelsChanged 注册渠道数量变化钩子(渠道创建/删除成功后调用)。
@@ -511,6 +540,9 @@ func (s *AiGatewayService) GatewayModels(ctx context.Context, group string) (aiw
 	if group != "" {
 		q = q.Where("ai_channels.channel_group = ?", group)
 	}
+	if s.FilterDeprecated() {
+		q = q.Where("ai_model_caches.deprecated_at IS NULL")
+	}
 	var rows []model.AiModelCache
 	err := q.Order("ai_model_caches.name ASC").Find(&rows).Error
 	list := aiwire.ModelList{Object: "list", Data: []aiwire.Model{}}

internal/service/aigateway_chat.go

@@ -164,6 +164,9 @@ func (s *AiGatewayService) modelChannels(ctx context.Context, modelName, capabil
 	} else {
 		q = q.Where("capability = ?", capability)
 	}
+	if s.FilterDeprecated() {
+		q = q.Where("deprecated_at IS NULL")
+	}
 	var rows []model.AiModelCache
 	if err := q.Find(&rows).Error; err != nil {
 		return nil, nil, err

internal/service/aigateway_extras.go

@@ -40,6 +40,42 @@ func SpeechBodyNormalize(raw []byte) (string, []byte, error) {
 	return modelName, out, nil
 }
 
+// ttsDefaultModel 是 /ai/v1/tts 缺省注入的模型(xAI 官方格式无 model 字段)。
+const ttsDefaultModel = "xai.grok-tts"
+
+// TtsBodyConvert 把 xAI 官方 TTS 格式转换为上游 OpenAI 兼容 audio/speech 形态:
+// text→input、voice_id→voice,text 与 language 必填(对齐 xAI 官方);
+// model 为网关扩展字段,缺省注入 ttsDefaultModel;其余字段原样保留。
+func TtsBodyConvert(raw []byte) (string, []byte, error) {
+	dec := json.NewDecoder(strings.NewReader(string(raw)))
+	dec.UseNumber()
+	var body map[string]any
+	if err := dec.Decode(&body); err != nil {
+		return "", nil, fmt.Errorf("解析请求体: %w", err)
+	}
+	text, _ := body["text"].(string)
+	language, _ := body["language"].(string)
+	if strings.TrimSpace(text) == "" || strings.TrimSpace(language) == "" {
+		return "", nil, fmt.Errorf("text 与 language 不能为空")
+	}
+	modelName, _ := body["model"].(string)
+	if strings.TrimSpace(modelName) == "" {
+		modelName = ttsDefaultModel
+		body["model"] = modelName
+	}
+	delete(body, "text")
+	body["input"] = text
+	if voice, ok := body["voice_id"].(string); ok && strings.TrimSpace(voice) != "" {
+		body["voice"] = voice
+	}
+	delete(body, "voice_id")
+	out, err := json.Marshal(body)
+	if err != nil {
+		return "", nil, fmt.Errorf("重组请求体: %w", err)
+	}
+	return modelName, out, nil
+}
+
 // Speech 编排 TTS 调用:按 TTS 能力选渠道,可重试错误换渠道(上限 3 次)。
 func (s *AiGatewayService) Speech(ctx context.Context, modelName string, body []byte, group string) ([]byte, string, ChatMeta, error) {
 	meta := ChatMeta{}

internal/service/aigateway_extras_test.go

@@ -45,6 +45,64 @@ func TestSpeechBodyNormalize(t *testing.T) {
 	}
 }
 
+// TestTtsBodyConvert 断言 xAI 官方 TTS 格式到 OpenAI 兼容形态的转换。
+func TestTtsBodyConvert(t *testing.T) {
+	tests := []struct {
+		name      string
+		raw       string
+		wantErr   bool
+		wantModel string
+	}{
+		{"缺 text 拒绝", `{"language":"zh"}`, true, ""},
+		{"缺 language 拒绝", `{"text":"你好"}`, true, ""},
+		{"缺省注入默认模型", `{"text":"你好","language":"zh"}`, false, "xai.grok-tts"},
+		{"model 扩展字段可覆盖", `{"model":"xai.other-tts","text":"你好","language":"auto"}`, false, "xai.other-tts"},
+		{"非 JSON 拒绝", `<html>`, true, ""},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			modelName, _, err := TtsBodyConvert([]byte(tt.raw))
+			if (err != nil) != tt.wantErr {
+				t.Fatalf("err = %v, wantErr %v", err, tt.wantErr)
+			}
+			if err != nil {
+				return
+			}
+			if modelName != tt.wantModel {
+				t.Fatalf("model = %s, want %s", modelName, tt.wantModel)
+			}
+		})
+	}
+}
+
+// TestTtsBodyConvertMapping 断言字段映射与未知字段保留。
+func TestTtsBodyConvertMapping(t *testing.T) {
+	raw := `{"text":"你好","language":"zh","voice_id":"ara","speed":1.2,` +
+		`"output_format":{"codec":"mp3","sample_rate":44100}}`
+	_, body, err := TtsBodyConvert([]byte(raw))
+	if err != nil {
+		t.Fatalf("err = %v", err)
+	}
+	var out map[string]any
+	_ = json.Unmarshal(body, &out)
+	if out["input"] != "你好" || out["voice"] != "ara" {
+		t.Fatalf("input/voice 映射错误: %v", out)
+	}
+	if _, ok := out["text"]; ok {
+		t.Fatal("text 字段应被移除")
+	}
+	if _, ok := out["voice_id"]; ok {
+		t.Fatal("voice_id 字段应被移除")
+	}
+	of, _ := out["output_format"].(map[string]any)
+	if of == nil || of["codec"] != "mp3" {
+		t.Fatalf("output_format 应原样保留: %v", out["output_format"])
+	}
+	if out["language"] != "zh" || out["speed"] == nil {
+		t.Fatalf("language/speed 应保留: %v", out)
+	}
+}
+
 // TestModerationInputs 断言 input 的 string / []string 解析与边界校验。
 func TestModerationInputs(t *testing.T) {
 	tests := []struct {

internal/service/aigateway_test.go

@@ -136,7 +136,7 @@ func (f *gatewayStubClient) GenAiProbeChat(ctx context.Context, cred oci.Credent
 func newTestGateway(t *testing.T, client oci.Client) (*AiGatewayService, *OciConfigService) {
 	t.Helper()
 	svc := newTestService(t, client)
-	if err := svc.db.AutoMigrate(&model.AiKey{}, &model.AiChannel{}, &model.AiModelCache{}, &model.AiModelBlacklist{}, &model.AiCallLog{}); err != nil {
+	if err := svc.db.AutoMigrate(&model.Setting{}, &model.AiKey{}, &model.AiChannel{}, &model.AiModelCache{}, &model.AiModelBlacklist{}, &model.AiCallLog{}); err != nil {
 		t.Fatalf("auto migrate ai tables: %v", err)
 	}
 	return NewAiGatewayService(svc.db, svc, client), svc
@@ -228,6 +228,46 @@ func seedChannel(t *testing.T, gw *AiGatewayService, cfgID uint, region string,
 	return ch
 }
 
+// TestFilterDeprecatedModels 断言「过滤弃用模型」开关对列表 / 路由的过滤与持久化。
+func TestFilterDeprecatedModels(t *testing.T) {
+	gw, svc := newTestGateway(t, &gatewayStubClient{fakeClient: &fakeClient{tenancy: oci.TenancyInfo{Name: "t"}}})
+	cfg := importAliveConfig(t, svc)
+	ch := seedChannel(t, gw, cfg.ID, "eu-frankfurt-1", 1, 1)
+	dep := time.Now().Add(-24 * time.Hour)
+	old := &model.AiModelCache{ChannelID: ch.ID, ModelOcid: "ocid1..dep", Name: "meta.llama-old",
+		Vendor: "meta", SyncedAt: time.Now(), DeprecatedAt: &dep}
+	if err := gw.db.Create(old).Error; err != nil {
+		t.Fatalf("seed deprecated cache: %v", err)
+	}
+	ctx := context.Background()
+
+	list, _ := gw.GatewayModels(ctx, "")
+	if len(list.Data) != 2 {
+		t.Fatalf("开关关:应含弃用模型,got %d", len(list.Data))
+	}
+	if err := gw.SetFilterDeprecated(ctx, true); err != nil {
+		t.Fatalf("SetFilterDeprecated: %v", err)
+	}
+	list, _ = gw.GatewayModels(ctx, "")
+	if len(list.Data) != 1 || list.Data[0].ID != "meta.llama-3.3-70b-instruct" {
+		t.Fatalf("开关开:弃用模型应被过滤,got %+v", list.Data)
+	}
+	if _, err := gw.pick(ctx, "meta.llama-old", "", "CHAT", map[uint]bool{}); !errors.Is(err, ErrAiUnknownModel) {
+		t.Errorf("开关开:弃用模型路由应不可达,err = %v", err)
+	}
+	// 持久化:重建 service 后开关仍生效
+	gw2 := NewAiGatewayService(gw.db, svc, &gatewayStubClient{})
+	if !gw2.FilterDeprecated() {
+		t.Error("重建后开关状态应保持开启")
+	}
+	if err := gw.SetFilterDeprecated(ctx, false); err != nil {
+		t.Fatalf("关闭开关: %v", err)
+	}
+	if list, _ = gw.GatewayModels(ctx, ""); len(list.Data) != 2 {
+		t.Errorf("开关关:弃用模型应恢复,got %d", len(list.Data))
+	}
+}
+
 func TestPickPriorityAndBreaker(t *testing.T) {
 	gw, svc := newTestGateway(t, &gatewayStubClient{fakeClient: &fakeClient{tenancy: oci.TenancyInfo{Name: "t"}}})
 	cfg := importAliveConfig(t, svc)