.env.example

@@ -0,0 +1,21 @@
+# CamTalk 环境变量模板
+# 复制为 .env 并填入实际值：cp .env.example .env
+# .env 已在 .gitignore 中，不会提交到版本控制
+
+# ---- AI 服务 API Key ----
+CAMTALK_AI_LLM_API_KEY=sk-xxx
+CAMTALK_AI_STT_API_KEY=
+CAMTALK_AI_TTS_API_KEY=
+
+# ---- 可选覆盖（默认值见 config.yaml）----
+# CAMTALK_AI_LLM_MODEL=gpt-4o
+# CAMTALK_AI_LLM_ENDPOINT=https://api.openai.com/v1
+# CAMTALK_AI_LLM_TIMEOUT=10
+# CAMTALK_AI_STT_ENDPOINT=wss://api.deepgram.com/v1/listen
+# CAMTALK_AI_TTS_ENDPOINT=https://api.openai.com/v1
+# CAMTALK_AI_TTS_VOICE=alloy
+# CAMTALK_AI_TTS_SPEED=1.0
+# CAMTALK_AI_TTS_TIMEOUT=5
+
+# ---- 应用 ----
+# APP_ENV=dev

backend/cmd/server/main.go

@@ -46,12 +46,12 @@ func main() {
 	defer sessionMgr.(*session.MemoryManager).Stop()
 
 	// 初始化 AI 服务
-	sttService := stt.NewDeepgramService(cfg.AI.STT.APIKey, cfg.AI.STT.Endpoint, logger.Log)
+	sttService := stt.NewDeepgramService(cfg.AI.STT.APIKey, cfg.AI.STT.Model, cfg.AI.STT.Endpoint, logger.Log)
 	llmService := llm.NewOpenAIService(cfg.AI.LLM.APIKey, cfg.AI.LLM.Model, cfg.AI.LLM.Endpoint, cfg.AI.LLM.Timeout, logger.Log)
-	ttsService := tts.NewOpenAIService(cfg.AI.TTS.APIKey, cfg.AI.TTS.Voice, cfg.AI.TTS.Endpoint, cfg.AI.TTS.Speed, cfg.AI.TTS.Timeout, logger.Log)
+	ttsService := tts.NewOpenAIService(cfg.AI.TTS.APIKey, cfg.AI.TTS.Model, cfg.AI.TTS.Voice, cfg.AI.TTS.Endpoint, cfg.AI.TTS.Speed, cfg.AI.TTS.Timeout, logger.Log)
 
 	// 初始化 Orchestrator
-	orch := orchestrator.New(sttService, llmService, ttsService, sessionMgr)
+	orch := orchestrator.New(sttService, llmService, ttsService, sessionMgr, cfg.AI.LLM.Model)
 
 	// Gin 模式
 	if cfg.App.Env == "prod" {

backend/config.yaml

@@ -15,18 +15,20 @@ redis:
 
 ai:
   stt:
-    provider: deepgram
+    provider: Xiaomi MiMo
-    endpoint: "wss://api.deepgram.com/v1/listen"
+    model: mimo-v2.5
+    endpoint: "https://token-plan-cn.xiaomimimo.com/v1"
   llm:
-    provider: openai
+    provider: dashscope
-    model: gpt-4o
+    model: qwen3-vl-plus
-    endpoint: "https://api.openai.com/v1"
+    endpoint: "https://dashscope.aliyuncs.com/compatible-mode/v1"
-    timeout: 10
+    timeout: 30
   tts:
-    provider: openai
+    provider: Xiaomi MiMo
+    model: mimo-v2.5
     voice: alloy
     speed: 1.0
-    endpoint: "https://api.openai.com/v1"
+    endpoint: "https://token-plan-cn.xiaomimimo.com/v1"
     timeout: 5
 
 storage:

backend/go.mod

@@ -27,6 +27,7 @@ require (
 	github.com/go-playground/validator/v10 v10.20.0 // indirect
 	github.com/go-viper/mapstructure/v2 v2.4.0 // indirect
 	github.com/goccy/go-json v0.10.2 // indirect
+	github.com/joho/godotenv v1.5.1 // indirect
 	github.com/json-iterator/go v1.1.12 // indirect
 	github.com/klauspost/cpuid/v2 v2.2.10 // indirect
 	github.com/leodido/go-urn v1.4.0 // indirect

backend/go.sum

@@ -44,6 +44,8 @@ github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
 github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
+github.com/joho/godotenv v1.5.1 h1:7eLL/+HRGLY0ldzfGMeQkb7vMd0as4CfYvUVzLqw0N0=
+github.com/joho/godotenv v1.5.1/go.mod h1:f4LDr5Voq0i2e/R5DDNOoa2zzDfwtkZa6DnEwAbqwq4=
 github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
 github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
 github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=

backend/internal/ai/stt/deepgram.go

@@ -16,17 +16,22 @@ import (
 // DeepgramService 基于 Deepgram WebSocket API 的语音识别实现。
 type DeepgramService struct {
 	apiKey   string
+	model    string
 	endpoint string
 	logger   *zap.SugaredLogger
 }
 
 // NewDeepgramService 创建 Deepgram STT 服务。
-func NewDeepgramService(apiKey, endpoint string, logger *zap.SugaredLogger) *DeepgramService {
+func NewDeepgramService(apiKey, model, endpoint string, logger *zap.SugaredLogger) *DeepgramService {
+	if model == "" {
+		model = "nova-2"
+	}
 	if endpoint == "" {
 		endpoint = "wss://api.deepgram.com/v1/listen"
 	}
 	return &DeepgramService{
 		apiKey:   apiKey,
+		model:    model,
 		endpoint: endpoint,
 		logger:   logger,
 	}
@@ -128,7 +133,7 @@ func (d *DeepgramService) buildURL(opts Options) string {
 	q.Set("encoding", encoding)
 	q.Set("sample_rate", fmt.Sprintf("%d", sampleRate))
 	q.Set("language", language)
-	q.Set("model", "nova-2")
+	q.Set("model", d.model)
 	q.Set("punctuate", "true")
 	u.RawQuery = q.Encode()
 

backend/internal/ai/stt/deepgram_test.go

@@ -74,7 +74,7 @@ func TestDeepgramService_Recognize_Success(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewDeepgramService("test-key", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
+	svc := NewDeepgramService("test-key", "", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
 
 	text, err := svc.Recognize(context.Background(), []byte("fake-pcm-audio"), Options{
 		Encoding:   "pcm_s16le",
@@ -90,7 +90,7 @@ func TestDeepgramService_Recognize_Success(t *testing.T) {
 }
 
 func TestDeepgramService_Recognize_EmptyAudio(t *testing.T) {
-	svc := NewDeepgramService("test-key", "ws://localhost", zap.NewNop().Sugar())
+	svc := NewDeepgramService("test-key", "", "ws://localhost", zap.NewNop().Sugar())
 	_, err := svc.Recognize(context.Background(), nil, Options{})
 	if err == nil {
 		t.Fatal("Recognize() with empty audio should return error")
@@ -98,7 +98,7 @@ func TestDeepgramService_Recognize_EmptyAudio(t *testing.T) {
 }
 
 func TestDeepgramService_Recognize_ConnectError(t *testing.T) {
-	svc := NewDeepgramService("test-key", "ws://localhost:1", zap.NewNop().Sugar())
+	svc := NewDeepgramService("test-key", "", "ws://localhost:1", zap.NewNop().Sugar())
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
 	defer cancel()
 
@@ -117,7 +117,7 @@ func TestDeepgramService_Recognize_Timeout(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewDeepgramService("test-key", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
+	svc := NewDeepgramService("test-key", "", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
 
 	ctx, cancel := context.WithTimeout(context.Background(), 1*time.Second)
 	defer cancel()
@@ -147,7 +147,7 @@ func TestDeepgramService_Recognize_MultipleFinals(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewDeepgramService("test-key", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
+	svc := NewDeepgramService("test-key", "", wsToWss(srv.URL)+"/v1/listen", zap.NewNop().Sugar())
 
 	text, err := svc.Recognize(context.Background(), []byte("audio"), Options{})
 	if err != nil {
@@ -159,7 +159,7 @@ func TestDeepgramService_Recognize_MultipleFinals(t *testing.T) {
 }
 
 func TestDeepgramService_buildURL(t *testing.T) {
-	svc := NewDeepgramService("key", "wss://api.deepgram.com/v1/listen", zap.NewNop().Sugar())
+	svc := NewDeepgramService("key", "", "wss://api.deepgram.com/v1/listen", zap.NewNop().Sugar())
 
 	tests := []struct {
 		name string

backend/internal/ai/tts/openai.go

@@ -15,6 +15,7 @@ import (
 // OpenAIService 基于 OpenAI TTS API 的语音合成实现。
 type OpenAIService struct {
 	apiKey   string
+	model    string
 	voice    string
 	speed    float64
 	endpoint string
@@ -24,7 +25,10 @@ type OpenAIService struct {
 }
 
 // NewOpenAIService 创建 OpenAI TTS 服务。
-func NewOpenAIService(apiKey, voice, endpoint string, speed float64, timeoutSec int, logger *zap.SugaredLogger) *OpenAIService {
+func NewOpenAIService(apiKey, model, voice, endpoint string, speed float64, timeoutSec int, logger *zap.SugaredLogger) *OpenAIService {
+	if model == "" {
+		model = "tts-1"
+	}
 	if voice == "" {
 		voice = "alloy"
 	}
@@ -40,6 +44,7 @@ func NewOpenAIService(apiKey, voice, endpoint string, speed float64, timeoutSec
 	}
 	return &OpenAIService{
 		apiKey:   apiKey,
+		model:    model,
 		voice:    voice,
 		speed:    speed,
 		endpoint: endpoint,
@@ -109,7 +114,7 @@ func (o *OpenAIService) synthesize(ctx context.Context, text, voice string, spee
 	defer cancel()
 
 	body := ttsRequest{
-		Model:          "tts-1",
+		Model:          o.model,
 		Input:          text,
 		Voice:          voice,
 		ResponseFormat: "mp3",

backend/internal/ai/tts/openai_test.go

@@ -58,7 +58,7 @@ func TestOpenAIService_SynthesizeStream_Success(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	textStream := sendSentences("你好", "世界", "！")
 
@@ -110,7 +110,7 @@ func TestOpenAIService_SynthesizeStream_APIError(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	textStream := sendSentences("你好")
 
@@ -142,7 +142,7 @@ func TestOpenAIService_SynthesizeStream_Timeout(t *testing.T) {
 	defer srv.Close()
 
 	// 1 秒超时
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 1, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 1, zap.NewNop().Sugar())
 
 	textStream := sendSentences("很长的句子")
 
@@ -177,7 +177,7 @@ func TestOpenAIService_SynthesizeStream_EmptyText(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	// 空句子应该被跳过
 	textStream := sendSentences("", "你好", "")
@@ -210,7 +210,7 @@ func TestOpenAIService_SynthesizeStream_ContextCancelled(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	// 发送多个句子，但在第一个后取消
 	textStream := make(chan string, 3)
@@ -252,7 +252,7 @@ func TestOpenAIService_SynthesizeStream_PartialFailure(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	textStream := sendSentences("第一句", "第二句", "第三句")
 
@@ -286,7 +286,7 @@ func TestOpenAIService_SynthesizeStream_CustomVoice(t *testing.T) {
 	})
 	defer srv.Close()
 
-	svc := NewOpenAIService("test-key", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
+	svc := NewOpenAIService("test-key", "", "alloy", srv.URL, 1.0, 5, zap.NewNop().Sugar())
 
 	textStream := sendSentences("你好")
 

backend/internal/config/config.go

@@ -5,6 +5,7 @@ import (
 	"os"
 	"strings"
 
+	"github.com/joho/godotenv"
 	"github.com/spf13/viper"
 )
 
@@ -50,6 +51,7 @@ type AIConfig struct {
 type STTConfig struct {
 	Provider string `mapstructure:"provider"`
 	APIKey   string `mapstructure:"api_key"`
+	Model    string `mapstructure:"model"`
 	Endpoint string `mapstructure:"endpoint"`
 }
 
@@ -64,6 +66,7 @@ type LLMConfig struct {
 type TTSConfig struct {
 	Provider string  `mapstructure:"provider"`
 	APIKey   string  `mapstructure:"api_key"`
+	Model    string  `mapstructure:"model"`
 	Voice    string  `mapstructure:"voice"`
 	Speed    float64 `mapstructure:"speed"`
 	Endpoint string  `mapstructure:"endpoint"`
@@ -98,12 +101,14 @@ func Load() (*Config, error) {
 	v.SetDefault("redis.addr", "localhost:6379")
 	v.SetDefault("redis.db", 0)
 	v.SetDefault("ai.stt.provider", "deepgram")
+	v.SetDefault("ai.stt.model", "nova-2")
 	v.SetDefault("ai.stt.endpoint", "wss://api.deepgram.com/v1/listen")
 	v.SetDefault("ai.llm.provider", "openai")
 	v.SetDefault("ai.llm.model", "gpt-4o")
 	v.SetDefault("ai.llm.endpoint", "https://api.openai.com/v1")
 	v.SetDefault("ai.llm.timeout", 10)
 	v.SetDefault("ai.tts.provider", "openai")
+	v.SetDefault("ai.tts.model", "tts-1")
 	v.SetDefault("ai.tts.voice", "alloy")
 	v.SetDefault("ai.tts.speed", 1.0)
 	v.SetDefault("ai.tts.endpoint", "https://api.openai.com/v1")
@@ -125,6 +130,11 @@ func Load() (*Config, error) {
 		_ = v.MergeInConfig()
 	}
 
+	// 加载 .env 文件（不覆盖已有环境变量）
+	// 按优先级尝试：当前目录、上级目录（兼容从 backend/ 或项目根目录启动）
+	_ = godotenv.Load()
+	_ = godotenv.Load("../.env")
+
 	// 环境变量覆盖
 	v.SetEnvPrefix("CAMTALK")
 	v.SetEnvKeyReplacer(strings.NewReplacer(".", "_"))

backend/internal/orchestrator/pipeline.go

@@ -22,6 +22,7 @@ type Pipeline struct {
 	llmService llm.Service
 	ttsService tts.Service
 	sessionMgr session.Manager
+	model      string // LLM 模型名，用于 llm_done 上报
 }
 
 // New 创建 Pipeline 实例。
@@ -30,12 +31,14 @@ func New(
 	llmService llm.Service,
 	ttsService tts.Service,
 	sessionMgr session.Manager,
+	model string,
 ) *Pipeline {
 	return &Pipeline{
 		sttService: sttService,
 		llmService: llmService,
 		ttsService: ttsService,
 		sessionMgr: sessionMgr,
+		model:      model,
 	}
 }
 
@@ -206,7 +209,7 @@ func (p *Pipeline) ProcessQuery(
 		Type:      "llm_done",
 		RequestID: req.RequestID,
 		FullText:  fullText,
-		Model:     "gpt-4o",
+		Model:     p.model,
 		LatencyMs: latency,
 	}); err != nil {
 		log.Errorw("发送 llm_done 失败", "error", err)

backend/internal/orchestrator/pipeline_test.go

@@ -254,7 +254,7 @@ func TestProcessQuery_Success(t *testing.T) {
 	mockSender.On("SendTTSAudio", mock.Anything).Return(nil)
 
 	// 创建 Pipeline
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	// 执行
 	ctx := context.Background()
@@ -305,7 +305,7 @@ func TestProcessQuery_STTError(t *testing.T) {
 
 	mockSender.On("SendError", mock.Anything).Return(nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)
@@ -357,7 +357,7 @@ func TestProcessQuery_LLMError(t *testing.T) {
 
 	mockSender.On("SendError", mock.Anything).Return(nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)
@@ -415,7 +415,7 @@ func TestProcessQuery_TTSError(t *testing.T) {
 	mockTTS.On("SynthesizeStream", mock.Anything, mock.Anything, mock.Anything).
 		Return(nil, errors.New("TTS service unavailable"))
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)
@@ -485,7 +485,7 @@ func TestProcessQuery_ContextCancelled(t *testing.T) {
 	}()
 	mockTTS.On("SynthesizeStream", mock.Anything, mock.Anything, mock.Anything).Return((<-chan tts.Chunk)(ttsCh), nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	// 创建可取消的上下文
 	ctx, cancel := context.WithCancel(context.Background())
@@ -545,7 +545,7 @@ func TestProcessQuery_DisabledTTS(t *testing.T) {
 	mockSender.On("SendLLMChunk", mock.Anything).Return(nil)
 	mockSender.On("SendLLMDone", mock.Anything).Return(nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)
@@ -617,7 +617,7 @@ func TestProcessQuery_InvalidAudio(t *testing.T) {
 
 	mockSender.On("SendError", mock.Anything).Return(nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)
@@ -650,7 +650,7 @@ func TestProcessQuery_SessionNotFound(t *testing.T) {
 
 	mockSender.On("SendError", mock.Anything).Return(nil)
 
-	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession)
+	pipeline := New(mockSTT, mockLLM, mockTTS, mockSession, "gpt-4o")
 
 	ctx := context.Background()
 	err := pipeline.ProcessQuery(ctx, "session-123", req, nil, mockSender)