feat: add support for multi-key channels in RelayInfo and access token caching

fix: correct Gemini channel model retrieval logic

controller/channel.go

@@ -36,11 +36,30 @@ type OpenAIModel struct {
 	Parent string `json:"parent"`
 }
 
+type GoogleOpenAICompatibleModels []struct {
+	Name                       string   `json:"name"`
+	Version                    string   `json:"version"`
+	DisplayName                string   `json:"displayName"`
+	Description                string   `json:"description,omitempty"`
+	InputTokenLimit            int      `json:"inputTokenLimit"`
+	OutputTokenLimit           int      `json:"outputTokenLimit"`
+	SupportedGenerationMethods []string `json:"supportedGenerationMethods"`
+	Temperature                float64  `json:"temperature,omitempty"`
+	TopP                       float64  `json:"topP,omitempty"`
+	TopK                       int      `json:"topK,omitempty"`
+	MaxTemperature             int      `json:"maxTemperature,omitempty"`
+}
+
 type OpenAIModelsResponse struct {
 	Data    []OpenAIModel `json:"data"`
 	Success bool          `json:"success"`
 }
 
+type GoogleOpenAICompatibleResponse struct {
+	Models        []GoogleOpenAICompatibleModels `json:"models"`
+	NextPageToken string                         `json:"nextPageToken"`
+}
+
 func parseStatusFilter(statusParam string) int {
 	switch strings.ToLower(statusParam) {
 	case "enabled", "1":
@@ -168,26 +187,59 @@ func FetchUpstreamModels(c *gin.Context) {
 	if channel.GetBaseURL() != "" {
 		baseURL = channel.GetBaseURL()
 	}
-	url := fmt.Sprintf("%s/v1/models", baseURL)
+
+	var url string
 	switch channel.Type {
 	case constant.ChannelTypeGemini:
-		url = fmt.Sprintf("%s/v1beta/openai/models", baseURL)
+		// curl https://example.com/v1beta/models?key=$GEMINI_API_KEY
+		url = fmt.Sprintf("%s/v1beta/openai/models?key=%s", baseURL, channel.Key)
 	case constant.ChannelTypeAli:
 		url = fmt.Sprintf("%s/compatible-mode/v1/models", baseURL)
+	default:
+		url = fmt.Sprintf("%s/v1/models", baseURL)
+	}
+
+	// 获取响应体 - 根据渠道类型决定是否添加 AuthHeader
+	var body []byte
+	if channel.Type == constant.ChannelTypeGemini {
+		body, err = GetResponseBody("GET", url, channel, nil) // I don't know why, but Gemini requires no AuthHeader
+	} else {
+		body, err = GetResponseBody("GET", url, channel, GetAuthHeader(channel.Key))
 	}
-	body, err := GetResponseBody("GET", url, channel, GetAuthHeader(channel.Key))
 	if err != nil {
 		common.ApiError(c, err)
 		return
 	}
 
 	var result OpenAIModelsResponse
-	if err = json.Unmarshal(body, &result); err != nil {
-		c.JSON(http.StatusOK, gin.H{
-			"success": false,
-			"message": fmt.Sprintf("解析响应失败: %s", err.Error()),
-		})
-		return
+	var parseSuccess bool
+
+	// 适配特殊格式
+	switch channel.Type {
+	case constant.ChannelTypeGemini:
+		var googleResult GoogleOpenAICompatibleResponse
+		if err = json.Unmarshal(body, &googleResult); err == nil {
+			// 转换Google格式到OpenAI格式
+			for _, model := range googleResult.Models {
+				for _, gModel := range model {
+					result.Data = append(result.Data, OpenAIModel{
+						ID: gModel.Name,
+					})
+				}
+			}
+			parseSuccess = true
+		}
+	}
+
+	// 如果解析失败，尝试OpenAI格式
+	if !parseSuccess {
+		if err = json.Unmarshal(body, &result); err != nil {
+			c.JSON(http.StatusOK, gin.H{
+				"success": false,
+				"message": fmt.Sprintf("解析响应失败: %s", err.Error()),
+			})
+			return
+		}
 	}
 
 	var ids []string
@@ -669,6 +721,7 @@ func DeleteChannelBatch(c *gin.Context) {
 type PatchChannel struct {
 	model.Channel
 	MultiKeyMode *string `json:"multi_key_mode"`
+	KeyMode      *string `json:"key_mode"` // 多key模式下密钥覆盖或者追加
 }
 
 func UpdateChannel(c *gin.Context) {
@@ -688,7 +741,7 @@ func UpdateChannel(c *gin.Context) {
 		return
 	}
 	// Preserve existing ChannelInfo to ensure multi-key channels keep correct state even if the client does not send ChannelInfo in the request.
-	originChannel, err := model.GetChannelById(channel.Id, false)
+	originChannel, err := model.GetChannelById(channel.Id, true)
 	if err != nil {
 		c.JSON(http.StatusOK, gin.H{
 			"success": false,
@@ -704,6 +757,69 @@ func UpdateChannel(c *gin.Context) {
 	if channel.MultiKeyMode != nil && *channel.MultiKeyMode != "" {
 		channel.ChannelInfo.MultiKeyMode = constant.MultiKeyMode(*channel.MultiKeyMode)
 	}
+
+	// 处理多key模式下的密钥追加/覆盖逻辑
+	if channel.KeyMode != nil && channel.ChannelInfo.IsMultiKey {
+		switch *channel.KeyMode {
+		case "append":
+			// 追加模式：将新密钥添加到现有密钥列表
+			if originChannel.Key != "" {
+				var newKeys []string
+				var existingKeys []string
+
+				// 解析现有密钥
+				if strings.HasPrefix(strings.TrimSpace(originChannel.Key), "[") {
+					// JSON数组格式
+					var arr []json.RawMessage
+					if err := json.Unmarshal([]byte(strings.TrimSpace(originChannel.Key)), &arr); err == nil {
+						existingKeys = make([]string, len(arr))
+						for i, v := range arr {
+							existingKeys[i] = string(v)
+						}
+					}
+				} else {
+					// 换行分隔格式
+					existingKeys = strings.Split(strings.Trim(originChannel.Key, "\n"), "\n")
+				}
+
+				// 处理 Vertex AI 的特殊情况
+				if channel.Type == constant.ChannelTypeVertexAi {
+					// 尝试解析新密钥为JSON数组
+					if strings.HasPrefix(strings.TrimSpace(channel.Key), "[") {
+						array, err := getVertexArrayKeys(channel.Key)
+						if err != nil {
+							c.JSON(http.StatusOK, gin.H{
+								"success": false,
+								"message": "追加密钥解析失败: " + err.Error(),
+							})
+							return
+						}
+						newKeys = array
+					} else {
+						// 单个JSON密钥
+						newKeys = []string{channel.Key}
+					}
+					// 合并密钥
+					allKeys := append(existingKeys, newKeys...)
+					channel.Key = strings.Join(allKeys, "\n")
+				} else {
+					// 普通渠道的处理
+					inputKeys := strings.Split(channel.Key, "\n")
+					for _, key := range inputKeys {
+						key = strings.TrimSpace(key)
+						if key != "" {
+							newKeys = append(newKeys, key)
+						}
+					}
+					// 合并密钥
+					allKeys := append(existingKeys, newKeys...)
+					channel.Key = strings.Join(allKeys, "\n")
+				}
+			}
+		case "replace":
+			// 覆盖模式：直接使用新密钥（默认行为，不需要特殊处理）
+		}
+	}
 	err = channel.Update()
 	if err != nil {
 		common.ApiError(c, err)

controller/relay.go

@@ -47,7 +47,7 @@ func relayHandler(c *gin.Context, relayMode int) *types.NewAPIError {
 		err = relay.TextHelper(c)
 	}
 
-	if constant2.ErrorLogEnabled && err != nil {
+	if constant2.ErrorLogEnabled && err != nil && types.IsRecordErrorLog(err) {
 		// 保存错误日志到mysql中
 		userId := c.GetInt("id")
 		tokenName := c.GetString("token_name")
@@ -62,6 +62,14 @@ func relayHandler(c *gin.Context, relayMode int) *types.NewAPIError {
 		other["channel_id"] = channelId
 		other["channel_name"] = c.GetString("channel_name")
 		other["channel_type"] = c.GetInt("channel_type")
+		adminInfo := make(map[string]interface{})
+		adminInfo["use_channel"] = c.GetStringSlice("use_channel")
+		isMultiKey := common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey)
+		if isMultiKey {
+			adminInfo["is_multi_key"] = true
+			adminInfo["multi_key_index"] = common.GetContextKeyInt(c, constant.ContextKeyChannelMultiKeyIndex)
+		}
+		other["admin_info"] = adminInfo
 		model.RecordErrorLog(c, userId, channelId, modelName, tokenName, err.MaskSensitiveError(), tokenId, 0, false, userGroup, other)
 	}
 

dto/gemini.go

@@ -1,4 +1,4 @@
-package gemini
+package dto
 
 import (
 	"encoding/json"
@@ -56,7 +56,7 @@ type FunctionCall struct {
 	Arguments    any    `json:"args"`
 }
 
-type FunctionResponse struct {
+type GeminiFunctionResponse struct {
 	Name     string                 `json:"name"`
 	Response map[string]interface{} `json:"response"`
 }
@@ -81,7 +81,7 @@ type GeminiPart struct {
 	Thought             bool                           `json:"thought,omitempty"`
 	InlineData          *GeminiInlineData              `json:"inlineData,omitempty"`
 	FunctionCall        *FunctionCall                  `json:"functionCall,omitempty"`
-	FunctionResponse    *FunctionResponse              `json:"functionResponse,omitempty"`
+	FunctionResponse    *GeminiFunctionResponse        `json:"functionResponse,omitempty"`
 	FileData            *GeminiFileData                `json:"fileData,omitempty"`
 	ExecutableCode      *GeminiPartExecutableCode      `json:"executableCode,omitempty"`
 	CodeExecutionResult *GeminiPartCodeExecutionResult `json:"codeExecutionResult,omitempty"`

middleware/distributor.go

@@ -269,6 +269,9 @@ func SetupContextForSelectedChannel(c *gin.Context, channel *model.Channel, mode
 	if channel.ChannelInfo.IsMultiKey {
 		common.SetContextKey(c, constant.ContextKeyChannelIsMultiKey, true)
 		common.SetContextKey(c, constant.ContextKeyChannelMultiKeyIndex, index)
+	} else {
+		// 必须设置为 false，否则在重试到单个 key 的时候会导致日志显示错误
+		common.SetContextKey(c, constant.ContextKeyChannelIsMultiKey, false)
 	}
 	// c.Request.Header.Set("Authorization", fmt.Sprintf("Bearer %s", key))
 	common.SetContextKey(c, constant.ContextKeyChannelKey, key)

model/ability.go

@@ -284,6 +284,21 @@ func FixAbility() (int, int, error) {
 		return 0, 0, errors.New("已经有一个修复任务在运行中，请稍后再试")
 	}
 	defer fixLock.Unlock()
+
+	// truncate abilities table
+	if common.UsingSQLite {
+		err := DB.Exec("DELETE FROM abilities").Error
+		if err != nil {
+			common.SysError(fmt.Sprintf("Delete abilities failed: %s", err.Error()))
+			return 0, 0, err
+		}
+	} else {
+		err := DB.Exec("TRUNCATE TABLE abilities").Error
+		if err != nil {
+			common.SysError(fmt.Sprintf("Truncate abilities failed: %s", err.Error()))
+			return 0, 0, err
+		}
+	}
 	var channels []*Channel
 	// Find all channels
 	err := DB.Model(&Channel{}).Find(&channels).Error

model/channel.go

@@ -46,6 +46,9 @@ type Channel struct {
 	ParamOverride     *string `json:"param_override" gorm:"type:text"`
 	// add after v0.8.5
 	ChannelInfo ChannelInfo `json:"channel_info" gorm:"type:json"`
+
+	// cache info
+	Keys []string `json:"-" gorm:"-"`
 }
 
 type ChannelInfo struct {
@@ -71,6 +74,9 @@ func (channel *Channel) getKeys() []string {
 	if channel.Key == "" {
 		return []string{}
 	}
+	if len(channel.Keys) > 0 {
+		return channel.Keys
+	}
 	trimmed := strings.TrimSpace(channel.Key)
 	// If the key starts with '[', try to parse it as a JSON array (e.g., for Vertex AI scenarios)
 	if strings.HasPrefix(trimmed, "[") {

model/channel_cache.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"math/rand"
 	"one-api/common"
+	"one-api/constant"
 	"one-api/setting"
 	"sort"
 	"strings"
@@ -66,6 +67,20 @@ func InitChannelCache() {
 
 	channelSyncLock.Lock()
 	group2model2channels = newGroup2model2channels
+	//channelsIDM = newChannelId2channel
+	for i, channel := range newChannelId2channel {
+		if channel.ChannelInfo.IsMultiKey {
+			channel.Keys = channel.getKeys()
+			if channel.ChannelInfo.MultiKeyMode == constant.MultiKeyModePolling {
+				if oldChannel, ok := channelsIDM[i]; ok {
+					// 存在旧的渠道，如果是多key且轮询，保留轮询索引信息
+					if oldChannel.ChannelInfo.IsMultiKey && oldChannel.ChannelInfo.MultiKeyMode == constant.MultiKeyModePolling {
+						channel.ChannelInfo.MultiKeyPollingIndex = oldChannel.ChannelInfo.MultiKeyPollingIndex
+					}
+				}
+			}
+		}
+	}
 	channelsIDM = newChannelId2channel
 	channelSyncLock.Unlock()
 	common.SysLog("channels synced from database")
@@ -203,9 +218,6 @@ func CacheGetChannel(id int) (*Channel, error) {
 	if !ok {
 		return nil, fmt.Errorf("渠道# %d，已不存在", id)
 	}
-	if c.Status != common.ChannelStatusEnabled {
-		return nil, fmt.Errorf("渠道# %d，已被禁用", id)
-	}
 	return c, nil
 }
 
@@ -224,9 +236,6 @@ func CacheGetChannelInfo(id int) (*ChannelInfo, error) {
 	if !ok {
 		return nil, fmt.Errorf("渠道# %d，已不存在", id)
 	}
-	if c.Status != common.ChannelStatusEnabled {
-		return nil, fmt.Errorf("渠道# %d，已被禁用", id)
-	}
 	return &c.ChannelInfo, nil
 }
 

relay/channel/adapter.go

@@ -26,6 +26,7 @@ type Adaptor interface {
 	GetModelList() []string
 	GetChannelName() string
 	ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error)
+	ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error)
 }
 
 type TaskAdaptor interface {

relay/channel/ali/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/aws/adaptor.go

@@ -22,6 +22,11 @@ type Adaptor struct {
 	RequestMode int
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
 	c.Set("request_model", request.Model)
 	c.Set("converted_request", request)

relay/channel/baidu/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/baidu_v2/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")
@@ -43,15 +48,15 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) error {
 	channel.SetupApiRequestHeader(info, c, req)
-        keyParts := strings.Split(info.ApiKey, "|")
+	keyParts := strings.Split(info.ApiKey, "|")
 	if len(keyParts) == 0 || keyParts[0] == "" {
-        	return errors.New("invalid API key: authorization token is required")
-        }
-	 if len(keyParts) > 1 {
-               if keyParts[1] != "" {
-                       req.Set("appid", keyParts[1])
-               }
-        }
+		return errors.New("invalid API key: authorization token is required")
+	}
+	if len(keyParts) > 1 {
+		if keyParts[1] != "" {
+			req.Set("appid", keyParts[1])
+		}
+	}
 	req.Set("Authorization", "Bearer "+keyParts[0])
 	return nil
 }

relay/channel/claude/adaptor.go

@@ -24,6 +24,11 @@ type Adaptor struct {
 	RequestMode int
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
 	return request, nil
 }

relay/channel/cloudflare/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/cohere/adaptor.go

@@ -17,6 +17,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/coze/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *common.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 // ConvertAudioRequest implements channel.Adaptor.
 func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *common.RelayInfo, request dto.AudioRequest) (io.Reader, error) {
 	return nil, errors.New("not implemented")

relay/channel/deepseek/adaptor.go

@@ -19,6 +19,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/dify/adaptor.go

@@ -24,6 +24,11 @@ type Adaptor struct {
 	BotType int
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/gemini/adaptor.go

@@ -20,6 +20,26 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
+	if len(request.Contents) > 0 {
+		for i, content := range request.Contents {
+			if i == 0 {
+				if request.Contents[0].Role == "" {
+					request.Contents[0].Role = "user"
+				}
+			}
+			for _, part := range content.Parts {
+				if part.FileData != nil {
+					if part.FileData.MimeType == "" && strings.Contains(part.FileData.FileUri, "www.youtube.com") {
+						part.FileData.MimeType = "video/webm"
+					}
+				}
+			}
+		}
+	}
+	return request, nil
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, req *dto.ClaudeRequest) (any, error) {
 	adaptor := openai.Adaptor{}
 	oaiReq, err := adaptor.ConvertClaudeRequest(c, info, req)
@@ -51,13 +71,13 @@ func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInf
 	}
 
 	// build gemini imagen request
-	geminiRequest := GeminiImageRequest{
-		Instances: []GeminiImageInstance{
+	geminiRequest := dto.GeminiImageRequest{
+		Instances: []dto.GeminiImageInstance{
 			{
 				Prompt: request.Prompt,
 			},
 		},
-		Parameters: GeminiImageParameters{
+		Parameters: dto.GeminiImageParameters{
 			SampleCount:      request.N,
 			AspectRatio:      aspectRatio,
 			PersonGeneration: "allow_adult", // default allow adult
@@ -138,9 +158,9 @@ func (a *Adaptor) ConvertEmbeddingRequest(c *gin.Context, info *relaycommon.Rela
 	}
 
 	// only process the first input
-	geminiRequest := GeminiEmbeddingRequest{
-		Content: GeminiChatContent{
-			Parts: []GeminiPart{
+	geminiRequest := dto.GeminiEmbeddingRequest{
+		Content: dto.GeminiChatContent{
+			Parts: []dto.GeminiPart{
 				{
 					Text: inputs[0],
 				},

relay/channel/gemini/relay-gemini-native.go

@@ -1,6 +1,7 @@
 package gemini
 
 import (
+	"github.com/pkg/errors"
 	"io"
 	"net/http"
 	"one-api/common"
@@ -28,7 +29,7 @@ func GeminiTextGenerationHandler(c *gin.Context, info *relaycommon.RelayInfo, re
 	}
 
 	// 解析为 Gemini 原生响应格式
-	var geminiResponse GeminiChatResponse
+	var geminiResponse dto.GeminiChatResponse
 	err = common.Unmarshal(responseBody, &geminiResponse)
 	if err != nil {
 		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
@@ -71,7 +72,7 @@ func GeminiTextGenerationStreamHandler(c *gin.Context, info *relaycommon.RelayIn
 	responseText := strings.Builder{}
 
 	helper.StreamScannerHandler(c, resp, info, func(data string) bool {
-		var geminiResponse GeminiChatResponse
+		var geminiResponse dto.GeminiChatResponse
 		err := common.UnmarshalJsonStr(data, &geminiResponse)
 		if err != nil {
 			common.LogError(c, "error unmarshalling stream response: "+err.Error())
@@ -110,10 +111,14 @@ func GeminiTextGenerationStreamHandler(c *gin.Context, info *relaycommon.RelayIn
 		if err != nil {
 			common.LogError(c, err.Error())
 		}
-
+		info.SendResponseCount++
 		return true
 	})
 
+	if info.SendResponseCount == 0 {
+		return nil, types.NewOpenAIError(errors.New("no response received from Gemini API"), types.ErrorCodeEmptyResponse, http.StatusInternalServerError)
+	}
+
 	if imageCount != 0 {
 		if usage.CompletionTokens == 0 {
 			usage.CompletionTokens = imageCount * 258

relay/channel/gemini/relay-gemini.go

@@ -81,7 +81,7 @@ func clampThinkingBudget(modelName string, budget int) int {
 	return budget
 }
 
-func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayInfo) {
+func ThinkingAdaptor(geminiRequest *dto.GeminiChatRequest, info *relaycommon.RelayInfo) {
 	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
 		modelName := info.UpstreamModelName
 		isNew25Pro := strings.HasPrefix(modelName, "gemini-2.5-pro") &&
@@ -93,7 +93,7 @@ func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayIn
 			if len(parts) == 2 && parts[1] != "" {
 				if budgetTokens, err := strconv.Atoi(parts[1]); err == nil {
 					clampedBudget := clampThinkingBudget(modelName, budgetTokens)
-					geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+					geminiRequest.GenerationConfig.ThinkingConfig = &dto.GeminiThinkingConfig{
 						ThinkingBudget:  common.GetPointer(clampedBudget),
 						IncludeThoughts: true,
 					}
@@ -113,11 +113,11 @@ func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayIn
 			}
 
 			if isUnsupported {
-				geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+				geminiRequest.GenerationConfig.ThinkingConfig = &dto.GeminiThinkingConfig{
 					IncludeThoughts: true,
 				}
 			} else {
-				geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+				geminiRequest.GenerationConfig.ThinkingConfig = &dto.GeminiThinkingConfig{
 					IncludeThoughts: true,
 				}
 				if geminiRequest.GenerationConfig.MaxOutputTokens > 0 {
@@ -128,7 +128,7 @@ func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayIn
 			}
 		} else if strings.HasSuffix(modelName, "-nothinking") {
 			if !isNew25Pro {
-				geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+				geminiRequest.GenerationConfig.ThinkingConfig = &dto.GeminiThinkingConfig{
 					ThinkingBudget: common.GetPointer(0),
 				}
 			}
@@ -137,11 +137,11 @@ func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayIn
 }
 
 // Setting safety to the lowest possible values since Gemini is already powerless enough
-func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*GeminiChatRequest, error) {
+func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*dto.GeminiChatRequest, error) {
 
-	geminiRequest := GeminiChatRequest{
-		Contents: make([]GeminiChatContent, 0, len(textRequest.Messages)),
-		GenerationConfig: GeminiChatGenerationConfig{
+	geminiRequest := dto.GeminiChatRequest{
+		Contents: make([]dto.GeminiChatContent, 0, len(textRequest.Messages)),
+		GenerationConfig: dto.GeminiChatGenerationConfig{
 			Temperature:     textRequest.Temperature,
 			TopP:            textRequest.TopP,
 			MaxOutputTokens: textRequest.MaxTokens,
@@ -158,9 +158,9 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 
 	ThinkingAdaptor(&geminiRequest, info)
 
-	safetySettings := make([]GeminiChatSafetySettings, 0, len(SafetySettingList))
+	safetySettings := make([]dto.GeminiChatSafetySettings, 0, len(SafetySettingList))
 	for _, category := range SafetySettingList {
-		safetySettings = append(safetySettings, GeminiChatSafetySettings{
+		safetySettings = append(safetySettings, dto.GeminiChatSafetySettings{
 			Category:  category,
 			Threshold: model_setting.GetGeminiSafetySetting(category),
 		})
@@ -198,17 +198,17 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 			functions = append(functions, tool.Function)
 		}
 		if codeExecution {
-			geminiRequest.Tools = append(geminiRequest.Tools, GeminiChatTool{
+			geminiRequest.Tools = append(geminiRequest.Tools, dto.GeminiChatTool{
 				CodeExecution: make(map[string]string),
 			})
 		}
 		if googleSearch {
-			geminiRequest.Tools = append(geminiRequest.Tools, GeminiChatTool{
+			geminiRequest.Tools = append(geminiRequest.Tools, dto.GeminiChatTool{
 				GoogleSearch: make(map[string]string),
 			})
 		}
 		if len(functions) > 0 {
-			geminiRequest.Tools = append(geminiRequest.Tools, GeminiChatTool{
+			geminiRequest.Tools = append(geminiRequest.Tools, dto.GeminiChatTool{
 				FunctionDeclarations: functions,
 			})
 		}
@@ -238,7 +238,7 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 			continue
 		} else if message.Role == "tool" || message.Role == "function" {
 			if len(geminiRequest.Contents) == 0 || geminiRequest.Contents[len(geminiRequest.Contents)-1].Role == "model" {
-				geminiRequest.Contents = append(geminiRequest.Contents, GeminiChatContent{
+				geminiRequest.Contents = append(geminiRequest.Contents, dto.GeminiChatContent{
 					Role: "user",
 				})
 			}
@@ -265,18 +265,18 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				}
 			}
 
-			functionResp := &FunctionResponse{
+			functionResp := &dto.GeminiFunctionResponse{
 				Name:     name,
 				Response: contentMap,
 			}
 
-			*parts = append(*parts, GeminiPart{
+			*parts = append(*parts, dto.GeminiPart{
 				FunctionResponse: functionResp,
 			})
 			continue
 		}
-		var parts []GeminiPart
-		content := GeminiChatContent{
+		var parts []dto.GeminiPart
+		content := dto.GeminiChatContent{
 			Role: message.Role,
 		}
 		// isToolCall := false
@@ -290,8 +290,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 						return nil, fmt.Errorf("invalid arguments for function %s, args: %s", call.Function.Name, call.Function.Arguments)
 					}
 				}
-				toolCall := GeminiPart{
-					FunctionCall: &FunctionCall{
+				toolCall := dto.GeminiPart{
+					FunctionCall: &dto.FunctionCall{
 						FunctionName: call.Function.Name,
 						Arguments:    args,
 					},
@@ -308,7 +308,7 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				if part.Text == "" {
 					continue
 				}
-				parts = append(parts, GeminiPart{
+				parts = append(parts, dto.GeminiPart{
 					Text: part.Text,
 				})
 			} else if part.Type == dto.ContentTypeImageURL {
@@ -331,8 +331,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 						return nil, fmt.Errorf("mime type is not supported by Gemini: '%s', url: '%s', supported types are: %v", fileData.MimeType, url, getSupportedMimeTypesList())
 					}
 
-					parts = append(parts, GeminiPart{
-						InlineData: &GeminiInlineData{
+					parts = append(parts, dto.GeminiPart{
+						InlineData: &dto.GeminiInlineData{
 							MimeType: fileData.MimeType, // 使用原始的 MimeType，因为大小写可能对API有意义
 							Data:     fileData.Base64Data,
 						},
@@ -342,8 +342,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 					if err != nil {
 						return nil, fmt.Errorf("decode base64 image data failed: %s", err.Error())
 					}
-					parts = append(parts, GeminiPart{
-						InlineData: &GeminiInlineData{
+					parts = append(parts, dto.GeminiPart{
+						InlineData: &dto.GeminiInlineData{
 							MimeType: format,
 							Data:     base64String,
 						},
@@ -357,8 +357,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				if err != nil {
 					return nil, fmt.Errorf("decode base64 file data failed: %s", err.Error())
 				}
-				parts = append(parts, GeminiPart{
-					InlineData: &GeminiInlineData{
+				parts = append(parts, dto.GeminiPart{
+					InlineData: &dto.GeminiInlineData{
 						MimeType: format,
 						Data:     base64String,
 					},
@@ -371,8 +371,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				if err != nil {
 					return nil, fmt.Errorf("decode base64 audio data failed: %s", err.Error())
 				}
-				parts = append(parts, GeminiPart{
-					InlineData: &GeminiInlineData{
+				parts = append(parts, dto.GeminiPart{
+					InlineData: &dto.GeminiInlineData{
 						MimeType: "audio/" + part.GetInputAudio().Format,
 						Data:     base64String,
 					},
@@ -392,8 +392,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 	}
 
 	if len(system_content) > 0 {
-		geminiRequest.SystemInstructions = &GeminiChatContent{
-			Parts: []GeminiPart{
+		geminiRequest.SystemInstructions = &dto.GeminiChatContent{
+			Parts: []dto.GeminiPart{
 				{
 					Text: strings.Join(system_content, "\n"),
 				},
@@ -636,7 +636,7 @@ func unescapeMapOrSlice(data interface{}) interface{} {
 	return data
 }
 
-func getResponseToolCall(item *GeminiPart) *dto.ToolCallResponse {
+func getResponseToolCall(item *dto.GeminiPart) *dto.ToolCallResponse {
 	var argsBytes []byte
 	var err error
 	if result, ok := item.FunctionCall.Arguments.(map[string]interface{}); ok {
@@ -658,7 +658,7 @@ func getResponseToolCall(item *GeminiPart) *dto.ToolCallResponse {
 	}
 }
 
-func responseGeminiChat2OpenAI(c *gin.Context, response *GeminiChatResponse) *dto.OpenAITextResponse {
+func responseGeminiChat2OpenAI(c *gin.Context, response *dto.GeminiChatResponse) *dto.OpenAITextResponse {
 	fullTextResponse := dto.OpenAITextResponse{
 		Id:      helper.GetResponseID(c),
 		Object:  "chat.completion",
@@ -725,10 +725,9 @@ func responseGeminiChat2OpenAI(c *gin.Context, response *GeminiChatResponse) *dt
 	return &fullTextResponse
 }
 
-func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.ChatCompletionsStreamResponse, bool, bool) {
+func streamResponseGeminiChat2OpenAI(geminiResponse *dto.GeminiChatResponse) (*dto.ChatCompletionsStreamResponse, bool) {
 	choices := make([]dto.ChatCompletionsStreamResponseChoice, 0, len(geminiResponse.Candidates))
 	isStop := false
-	hasImage := false
 	for _, candidate := range geminiResponse.Candidates {
 		if candidate.FinishReason != nil && *candidate.FinishReason == "STOP" {
 			isStop = true
@@ -759,7 +758,6 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
 				if strings.HasPrefix(part.InlineData.MimeType, "image") {
 					imgText := "![image](data:" + part.InlineData.MimeType + ";base64," + part.InlineData.Data + ")"
 					texts = append(texts, imgText)
-					hasImage = true
 				}
 			} else if part.FunctionCall != nil {
 				isTools = true
@@ -796,7 +794,7 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
 	var response dto.ChatCompletionsStreamResponse
 	response.Object = "chat.completion.chunk"
 	response.Choices = choices
-	return &response, isStop, hasImage
+	return &response, isStop
 }
 
 func handleStream(c *gin.Context, info *relaycommon.RelayInfo, resp *dto.ChatCompletionsStreamResponse) error {
@@ -824,23 +822,31 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 	// responseText := ""
 	id := helper.GetResponseID(c)
 	createAt := common.GetTimestamp()
+	responseText := strings.Builder{}
 	var usage = &dto.Usage{}
 	var imageCount int
 
-	respCount := 0
-
 	helper.StreamScannerHandler(c, resp, info, func(data string) bool {
-		var geminiResponse GeminiChatResponse
+		var geminiResponse dto.GeminiChatResponse
 		err := common.UnmarshalJsonStr(data, &geminiResponse)
 		if err != nil {
 			common.LogError(c, "error unmarshalling stream response: "+err.Error())
 			return false
 		}
 
-		response, isStop, hasImage := streamResponseGeminiChat2OpenAI(&geminiResponse)
-		if hasImage {
-			imageCount++
+		for _, candidate := range geminiResponse.Candidates {
+			for _, part := range candidate.Content.Parts {
+				if part.InlineData != nil && part.InlineData.MimeType != "" {
+					imageCount++
+				}
+				if part.Text != "" {
+					responseText.WriteString(part.Text)
+				}
+			}
 		}
+
+		response, isStop := streamResponseGeminiChat2OpenAI(&geminiResponse)
+
 		response.Id = id
 		response.Created = createAt
 		response.Model = info.UpstreamModelName
@@ -858,7 +864,7 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 			}
 		}
 
-		if respCount == 0 {
+		if info.SendResponseCount == 0 {
 			// send first response
 			err = handleStream(c, info, helper.GenerateStartEmptyResponse(id, createAt, info.UpstreamModelName, nil))
 			if err != nil {
@@ -873,11 +879,10 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 		if isStop {
 			_ = handleStream(c, info, helper.GenerateStopResponse(id, createAt, info.UpstreamModelName, constant.FinishReasonStop))
 		}
-		respCount++
 		return true
 	})
 
-	if respCount == 0 {
+	if info.SendResponseCount == 0 {
 		// 空补全，报错不计费
 		// empty response, throw an error
 		return nil, types.NewOpenAIError(errors.New("no response received from Gemini API"), types.ErrorCodeEmptyResponse, http.StatusInternalServerError)
@@ -892,6 +897,16 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 	usage.PromptTokensDetails.TextTokens = usage.PromptTokens
 	usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
 
+	if usage.CompletionTokens == 0 {
+		str := responseText.String()
+		if len(str) > 0 {
+			usage = service.ResponseText2Usage(responseText.String(), info.UpstreamModelName, info.PromptTokens)
+		} else {
+			// 空补全，不需要使用量
+			usage = &dto.Usage{}
+		}
+	}
+
 	response := helper.GenerateFinalUsageResponse(id, createAt, info.UpstreamModelName, *usage)
 	err := handleFinalStream(c, info, response)
 	if err != nil {
@@ -913,7 +928,7 @@ func GeminiChatHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.R
 	if common.DebugEnabled {
 		println(string(responseBody))
 	}
-	var geminiResponse GeminiChatResponse
+	var geminiResponse dto.GeminiChatResponse
 	err = common.Unmarshal(responseBody, &geminiResponse)
 	if err != nil {
 		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
@@ -959,7 +974,7 @@ func GeminiEmbeddingHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *h
 		return nil, types.NewOpenAIError(readErr, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
 	}
 
-	var geminiResponse GeminiEmbeddingResponse
+	var geminiResponse dto.GeminiEmbeddingResponse
 	if jsonErr := common.Unmarshal(responseBody, &geminiResponse); jsonErr != nil {
 		return nil, types.NewOpenAIError(jsonErr, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
 	}
@@ -1005,7 +1020,7 @@ func GeminiImageHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.
 	}
 	_ = resp.Body.Close()
 
-	var geminiResponse GeminiImageResponse
+	var geminiResponse dto.GeminiImageResponse
 	if jsonErr := common.Unmarshal(responseBody, &geminiResponse); jsonErr != nil {
 		return nil, types.NewOpenAIError(jsonErr, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
 	}

relay/channel/jimeng/adaptor.go

@@ -4,7 +4,6 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
-	"github.com/gin-gonic/gin"
 	"io"
 	"net/http"
 	"one-api/dto"
@@ -13,11 +12,18 @@ import (
 	relaycommon "one-api/relay/common"
 	relayconstant "one-api/relay/constant"
 	"one-api/types"
+
+	"github.com/gin-gonic/gin"
 )
 
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	return nil, errors.New("not implemented")
 }

relay/channel/jina/adaptor.go

@@ -19,6 +19,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/mistral/adaptor.go

@@ -16,6 +16,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/mokaai/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/ollama/adaptor.go

@@ -17,6 +17,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
 	openaiAdaptor := openai.Adaptor{}
 	openaiRequest, err := openaiAdaptor.ConvertClaudeRequest(c, info, request)

relay/channel/openai/adaptor.go

@@ -34,6 +34,15 @@ type Adaptor struct {
 	ResponseFormat string
 }
 
+func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
+	// 使用 service.GeminiToOpenAIRequest 转换请求格式
+	openaiRequest, err := service.GeminiToOpenAIRequest(request, info)
+	if err != nil {
+		return nil, err
+	}
+	return a.ConvertOpenAIRequest(c, info, openaiRequest)
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
 	//if !strings.Contains(request.Model, "claude") {
 	//	return nil, fmt.Errorf("you are using openai channel type with path /v1/messages, only claude model supported convert, but got %s", request.Model)
@@ -64,7 +73,7 @@ func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
 }
 
 func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
-	if info.RelayFormat == relaycommon.RelayFormatClaude {
+	if info.RelayFormat == relaycommon.RelayFormatClaude || info.RelayFormat == relaycommon.RelayFormatGemini {
 		return fmt.Sprintf("%s/v1/chat/completions", info.BaseUrl), nil
 	}
 	if info.RelayMode == relayconstant.RelayModeRealtime {

relay/channel/openai/helper.go

@@ -2,6 +2,8 @@ package openai
 
 import (
 	"encoding/json"
+	"errors"
+	"net/http"
 	"one-api/common"
 	"one-api/dto"
 	relaycommon "one-api/relay/common"
@@ -16,11 +18,14 @@ import (
 // 辅助函数
 func HandleStreamFormat(c *gin.Context, info *relaycommon.RelayInfo, data string, forceFormat bool, thinkToContent bool) error {
 	info.SendResponseCount++
+
 	switch info.RelayFormat {
 	case relaycommon.RelayFormatOpenAI:
 		return sendStreamData(c, info, data, forceFormat, thinkToContent)
 	case relaycommon.RelayFormatClaude:
 		return handleClaudeFormat(c, data, info)
+	case relaycommon.RelayFormatGemini:
+		return handleGeminiFormat(c, data, info)
 	}
 	return nil
 }
@@ -41,6 +46,36 @@ func handleClaudeFormat(c *gin.Context, data string, info *relaycommon.RelayInfo
 	return nil
 }
 
+func handleGeminiFormat(c *gin.Context, data string, info *relaycommon.RelayInfo) error {
+	var streamResponse dto.ChatCompletionsStreamResponse
+	if err := common.Unmarshal(common.StringToByteSlice(data), &streamResponse); err != nil {
+		common.LogError(c, "failed to unmarshal stream response: "+err.Error())
+		return err
+	}
+
+	geminiResponse := service.StreamResponseOpenAI2Gemini(&streamResponse, info)
+
+	// 如果返回 nil，表示没有实际内容，跳过发送
+	if geminiResponse == nil {
+		return nil
+	}
+
+	geminiResponseStr, err := common.Marshal(geminiResponse)
+	if err != nil {
+		common.LogError(c, "failed to marshal gemini response: "+err.Error())
+		return err
+	}
+
+	// send gemini format response
+	c.Render(-1, common.CustomEvent{Data: "data: " + string(geminiResponseStr)})
+	if flusher, ok := c.Writer.(http.Flusher); ok {
+		flusher.Flush()
+	} else {
+		return errors.New("streaming error: flusher not found")
+	}
+	return nil
+}
+
 func ProcessStreamResponse(streamResponse dto.ChatCompletionsStreamResponse, responseTextBuilder *strings.Builder, toolCount *int) error {
 	for _, choice := range streamResponse.Choices {
 		responseTextBuilder.WriteString(choice.Delta.GetContentString())
@@ -185,6 +220,37 @@ func HandleFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, lastStream
 		for _, resp := range claudeResponses {
 			_ = helper.ClaudeData(c, *resp)
 		}
+
+	case relaycommon.RelayFormatGemini:
+		var streamResponse dto.ChatCompletionsStreamResponse
+		if err := common.Unmarshal(common.StringToByteSlice(lastStreamData), &streamResponse); err != nil {
+			common.SysError("error unmarshalling stream response: " + err.Error())
+			return
+		}
+
+		// 这里处理的是 openai 最后一个流响应，其 delta 为空，有 finish_reason 字段
+		// 因此相比较于 google 官方的流响应，由 openai 转换而来会多一个 parts 为空，finishReason 为 STOP 的响应
+		// 而包含最后一段文本输出的响应（倒数第二个）的 finishReason 为 null
+		// 暂不知是否有程序会不兼容。
+
+		geminiResponse := service.StreamResponseOpenAI2Gemini(&streamResponse, info)
+
+		// openai 流响应开头的空数据
+		if geminiResponse == nil {
+			return
+		}
+
+		geminiResponseStr, err := common.Marshal(geminiResponse)
+		if err != nil {
+			common.SysError("error marshalling gemini response: " + err.Error())
+			return
+		}
+
+		// 发送最终的 Gemini 响应
+		c.Render(-1, common.CustomEvent{Data: "data: " + string(geminiResponseStr)})
+		if flusher, ok := c.Writer.(http.Flusher); ok {
+			flusher.Flush()
+		}
 	}
 }
 

relay/channel/openai/relay-openai.go

@@ -223,6 +223,13 @@ func OpenaiHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Respo
 			return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
 		}
 		responseBody = claudeRespStr
+	case relaycommon.RelayFormatGemini:
+		geminiResp := service.ResponseOpenAI2Gemini(&simpleResponse, info)
+		geminiRespStr, err := common.Marshal(geminiResp)
+		if err != nil {
+			return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		responseBody = geminiRespStr
 	}
 
 	common.IOCopyBytesGracefully(c, resp, responseBody)

relay/channel/palm/adaptor.go

@@ -17,6 +17,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/perplexity/adaptor.go

@@ -17,6 +17,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/siliconflow/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, req *dto.ClaudeRequest) (any, error) {
 	adaptor := openai.Adaptor{}
 	return adaptor.ConvertClaudeRequest(c, info, req)

relay/channel/tencent/adaptor.go

@@ -25,6 +25,11 @@ type Adaptor struct {
 	Timestamp int64
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/vertex/adaptor.go

@@ -44,6 +44,11 @@ type Adaptor struct {
 	AccountCredentials Credentials
 }
 
+func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
+	geminiAdaptor := gemini.Adaptor{}
+	return geminiAdaptor.ConvertGeminiRequest(c, info, request)
+}
+
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
 	if v, ok := claudeModelMap[info.UpstreamModelName]; ok {
 		c.Set("request_model", v)

relay/channel/vertex/service_account.go

@@ -36,7 +36,12 @@ var Cache = asynccache.NewAsyncCache(asynccache.Options{
 })
 
 func getAccessToken(a *Adaptor, info *relaycommon.RelayInfo) (string, error) {
-	cacheKey := fmt.Sprintf("access-token-%d", info.ChannelId)
+	var cacheKey string
+	if info.ChannelIsMultiKey {
+		cacheKey = fmt.Sprintf("access-token-%d-%d", info.ChannelId, info.ChannelMultiKeyIndex)
+	} else {
+		cacheKey = fmt.Sprintf("access-token-%d", info.ChannelId)
+	}
 	val, err := Cache.Get(cacheKey)
 	if err == nil {
 		return val.(string), nil

relay/channel/volcengine/adaptor.go

@@ -23,6 +23,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/xai/adaptor.go

@@ -19,6 +19,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	//panic("implement me")

relay/channel/xunfei/adaptor.go

@@ -17,6 +17,11 @@ type Adaptor struct {
 	request *dto.GeneralOpenAIRequest
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/zhipu/adaptor.go

@@ -16,6 +16,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/channel/zhipu_4v/adaptor.go

@@ -18,6 +18,11 @@ import (
 type Adaptor struct {
 }
 
+func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
 func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
 	//TODO implement me
 	panic("implement me")

relay/common/relay_info.go

@@ -60,17 +60,19 @@ type ResponsesUsageInfo struct {
 }
 
 type RelayInfo struct {
-	ChannelType       int
-	ChannelId         int
-	TokenId           int
-	TokenKey          string
-	UserId            int
-	UsingGroup        string // 使用的分组
-	UserGroup         string // 用户所在分组
-	TokenUnlimited    bool
-	StartTime         time.Time
-	FirstResponseTime time.Time
-	isFirstResponse   bool
+	ChannelType          int
+	ChannelId            int
+	ChannelIsMultiKey    bool // 是否多密钥
+	ChannelMultiKeyIndex int  // 多密钥索引
+	TokenId              int
+	TokenKey             string
+	UserId               int
+	UsingGroup           string // 使用的分组
+	UserGroup            string // 用户所在分组
+	TokenUnlimited       bool
+	StartTime            time.Time
+	FirstResponseTime    time.Time
+	isFirstResponse      bool
 	//SendLastReasoningResponse bool
 	ApiType           int
 	IsStream          bool
@@ -260,6 +262,9 @@ func GenRelayInfo(c *gin.Context) *RelayInfo {
 			IsFirstThinkingContent:  true,
 			SendLastThinkingContent: false,
 		},
+
+		ChannelIsMultiKey:    common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey),
+		ChannelMultiKeyIndex: common.GetContextKeyInt(c, constant.ContextKeyChannelMultiKeyIndex),
 	}
 	if strings.HasPrefix(c.Request.URL.Path, "/pg") {
 		info.IsPlayground = true

relay/gemini_handler.go

@@ -20,8 +20,8 @@ import (
 	"github.com/gin-gonic/gin"
 )
 
-func getAndValidateGeminiRequest(c *gin.Context) (*gemini.GeminiChatRequest, error) {
-	request := &gemini.GeminiChatRequest{}
+func getAndValidateGeminiRequest(c *gin.Context) (*dto.GeminiChatRequest, error) {
+	request := &dto.GeminiChatRequest{}
 	err := common.UnmarshalBodyReusable(c, request)
 	if err != nil {
 		return nil, err
@@ -44,7 +44,7 @@ func checkGeminiStreamMode(c *gin.Context, relayInfo *relaycommon.RelayInfo) {
 	// }
 }
 
-func checkGeminiInputSensitive(textRequest *gemini.GeminiChatRequest) ([]string, error) {
+func checkGeminiInputSensitive(textRequest *dto.GeminiChatRequest) ([]string, error) {
 	var inputTexts []string
 	for _, content := range textRequest.Contents {
 		for _, part := range content.Parts {
@@ -61,7 +61,7 @@ func checkGeminiInputSensitive(textRequest *gemini.GeminiChatRequest) ([]string,
 	return sensitiveWords, err
 }
 
-func getGeminiInputTokens(req *gemini.GeminiChatRequest, info *relaycommon.RelayInfo) int {
+func getGeminiInputTokens(req *dto.GeminiChatRequest, info *relaycommon.RelayInfo) int {
 	// 计算输入 token 数量
 	var inputTexts []string
 	for _, content := range req.Contents {
@@ -78,9 +78,13 @@ func getGeminiInputTokens(req *gemini.GeminiChatRequest, info *relaycommon.Relay
 	return inputTokens
 }
 
-func isNoThinkingRequest(req *gemini.GeminiChatRequest) bool {
+func isNoThinkingRequest(req *dto.GeminiChatRequest) bool {
 	if req.GenerationConfig.ThinkingConfig != nil && req.GenerationConfig.ThinkingConfig.ThinkingBudget != nil {
-		return *req.GenerationConfig.ThinkingConfig.ThinkingBudget == 0
+		configBudget := req.GenerationConfig.ThinkingConfig.ThinkingBudget
+		if configBudget != nil && *configBudget == 0 {
+			// 如果思考预算为 0，则认为是非思考请求
+			return true
+		}
 	}
 	return false
 }
@@ -202,7 +206,12 @@ func GeminiHelper(c *gin.Context) (newAPIError *types.NewAPIError) {
 		}
 		requestBody = bytes.NewReader(body)
 	} else {
-		jsonData, err := common.Marshal(req)
+		// 使用 ConvertGeminiRequest 转换请求格式
+		convertedRequest, err := adaptor.ConvertGeminiRequest(c, relayInfo, req)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+		}
+		jsonData, err := common.Marshal(convertedRequest)
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}

relay/relay-text.go

@@ -305,10 +305,10 @@ func preConsumeQuota(c *gin.Context, preConsumedQuota int, relayInfo *relaycommo
 		return 0, 0, types.NewError(err, types.ErrorCodeQueryDataError, types.ErrOptionWithSkipRetry())
 	}
 	if userQuota <= 0 {
-		return 0, 0, types.NewErrorWithStatusCode(errors.New("user quota is not enough"), types.ErrorCodeInsufficientUserQuota, http.StatusForbidden, types.ErrOptionWithSkipRetry())
+		return 0, 0, types.NewErrorWithStatusCode(errors.New("user quota is not enough"), types.ErrorCodeInsufficientUserQuota, http.StatusForbidden, types.ErrOptionWithSkipRetry(), types.ErrOptionWithNoRecordErrorLog())
 	}
 	if userQuota-preConsumedQuota < 0 {
-		return 0, 0, types.NewErrorWithStatusCode(fmt.Errorf("pre-consume quota failed, user quota: %s, need quota: %s", common.FormatQuota(userQuota), common.FormatQuota(preConsumedQuota)), types.ErrorCodeInsufficientUserQuota, http.StatusForbidden, types.ErrOptionWithSkipRetry())
+		return 0, 0, types.NewErrorWithStatusCode(fmt.Errorf("pre-consume quota failed, user quota: %s, need quota: %s", common.FormatQuota(userQuota), common.FormatQuota(preConsumedQuota)), types.ErrorCodeInsufficientUserQuota, http.StatusForbidden, types.ErrOptionWithSkipRetry(), types.ErrOptionWithNoRecordErrorLog())
 	}
 	relayInfo.UserQuota = userQuota
 	if userQuota > 100*preConsumedQuota {
@@ -332,7 +332,7 @@ func preConsumeQuota(c *gin.Context, preConsumedQuota int, relayInfo *relaycommo
 	if preConsumedQuota > 0 {
 		err := service.PreConsumeTokenQuota(relayInfo, preConsumedQuota)
 		if err != nil {
-			return 0, 0, types.NewErrorWithStatusCode(err, types.ErrorCodePreConsumeTokenQuotaFailed, http.StatusForbidden, types.ErrOptionWithSkipRetry())
+			return 0, 0, types.NewErrorWithStatusCode(err, types.ErrorCodePreConsumeTokenQuotaFailed, http.StatusForbidden, types.ErrOptionWithSkipRetry(), types.ErrOptionWithNoRecordErrorLog())
 		}
 		err = model.DecreaseUserQuota(relayInfo.UserId, preConsumedQuota)
 		if err != nil {

service/convert.go

@@ -448,3 +448,353 @@ func toJSONString(v interface{}) string {
 	}
 	return string(b)
 }
+
+func GeminiToOpenAIRequest(geminiRequest *dto.GeminiChatRequest, info *relaycommon.RelayInfo) (*dto.GeneralOpenAIRequest, error) {
+	openaiRequest := &dto.GeneralOpenAIRequest{
+		Model:  info.UpstreamModelName,
+		Stream: info.IsStream,
+	}
+
+	// 转换 messages
+	var messages []dto.Message
+	for _, content := range geminiRequest.Contents {
+		message := dto.Message{
+			Role: convertGeminiRoleToOpenAI(content.Role),
+		}
+
+		// 处理 parts
+		var mediaContents []dto.MediaContent
+		var toolCalls []dto.ToolCallRequest
+		for _, part := range content.Parts {
+			if part.Text != "" {
+				mediaContent := dto.MediaContent{
+					Type: "text",
+					Text: part.Text,
+				}
+				mediaContents = append(mediaContents, mediaContent)
+			} else if part.InlineData != nil {
+				mediaContent := dto.MediaContent{
+					Type: "image_url",
+					ImageUrl: &dto.MessageImageUrl{
+						Url:      fmt.Sprintf("data:%s;base64,%s", part.InlineData.MimeType, part.InlineData.Data),
+						Detail:   "auto",
+						MimeType: part.InlineData.MimeType,
+					},
+				}
+				mediaContents = append(mediaContents, mediaContent)
+			} else if part.FileData != nil {
+				mediaContent := dto.MediaContent{
+					Type: "image_url",
+					ImageUrl: &dto.MessageImageUrl{
+						Url:      part.FileData.FileUri,
+						Detail:   "auto",
+						MimeType: part.FileData.MimeType,
+					},
+				}
+				mediaContents = append(mediaContents, mediaContent)
+			} else if part.FunctionCall != nil {
+				// 处理 Gemini 的工具调用
+				toolCall := dto.ToolCallRequest{
+					ID:   fmt.Sprintf("call_%d", len(toolCalls)+1), // 生成唯一ID
+					Type: "function",
+					Function: dto.FunctionRequest{
+						Name:      part.FunctionCall.FunctionName,
+						Arguments: toJSONString(part.FunctionCall.Arguments),
+					},
+				}
+				toolCalls = append(toolCalls, toolCall)
+			} else if part.FunctionResponse != nil {
+				// 处理 Gemini 的工具响应，创建单独的 tool 消息
+				toolMessage := dto.Message{
+					Role:       "tool",
+					ToolCallId: fmt.Sprintf("call_%d", len(toolCalls)), // 使用对应的调用ID
+				}
+				toolMessage.SetStringContent(toJSONString(part.FunctionResponse.Response))
+				messages = append(messages, toolMessage)
+			}
+		}
+
+		// 设置消息内容
+		if len(toolCalls) > 0 {
+			// 如果有工具调用，设置工具调用
+			message.SetToolCalls(toolCalls)
+		} else if len(mediaContents) == 1 && mediaContents[0].Type == "text" {
+			// 如果只有一个文本内容，直接设置字符串
+			message.Content = mediaContents[0].Text
+		} else if len(mediaContents) > 0 {
+			// 如果有多个内容或包含媒体，设置为数组
+			message.SetMediaContent(mediaContents)
+		}
+
+		// 只有当消息有内容或工具调用时才添加
+		if len(message.ParseContent()) > 0 || len(message.ToolCalls) > 0 {
+			messages = append(messages, message)
+		}
+	}
+
+	openaiRequest.Messages = messages
+
+	if geminiRequest.GenerationConfig.Temperature != nil {
+		openaiRequest.Temperature = geminiRequest.GenerationConfig.Temperature
+	}
+	if geminiRequest.GenerationConfig.TopP > 0 {
+		openaiRequest.TopP = geminiRequest.GenerationConfig.TopP
+	}
+	if geminiRequest.GenerationConfig.TopK > 0 {
+		openaiRequest.TopK = int(geminiRequest.GenerationConfig.TopK)
+	}
+	if geminiRequest.GenerationConfig.MaxOutputTokens > 0 {
+		openaiRequest.MaxTokens = geminiRequest.GenerationConfig.MaxOutputTokens
+	}
+	// gemini stop sequences 最多 5 个，openai stop 最多 4 个
+	if len(geminiRequest.GenerationConfig.StopSequences) > 0 {
+		openaiRequest.Stop = geminiRequest.GenerationConfig.StopSequences[:4]
+	}
+	if geminiRequest.GenerationConfig.CandidateCount > 0 {
+		openaiRequest.N = geminiRequest.GenerationConfig.CandidateCount
+	}
+
+	// 转换工具调用
+	if len(geminiRequest.Tools) > 0 {
+		var tools []dto.ToolCallRequest
+		for _, tool := range geminiRequest.Tools {
+			if tool.FunctionDeclarations != nil {
+				// 将 Gemini 的 FunctionDeclarations 转换为 OpenAI 的 ToolCallRequest
+				functionDeclarations, ok := tool.FunctionDeclarations.([]dto.FunctionRequest)
+				if ok {
+					for _, function := range functionDeclarations {
+						openAITool := dto.ToolCallRequest{
+							Type: "function",
+							Function: dto.FunctionRequest{
+								Name:        function.Name,
+								Description: function.Description,
+								Parameters:  function.Parameters,
+							},
+						}
+						tools = append(tools, openAITool)
+					}
+				}
+			}
+		}
+		if len(tools) > 0 {
+			openaiRequest.Tools = tools
+		}
+	}
+
+	// gemini system instructions
+	if geminiRequest.SystemInstructions != nil {
+		// 将系统指令作为第一条消息插入
+		systemMessage := dto.Message{
+			Role:    "system",
+			Content: extractTextFromGeminiParts(geminiRequest.SystemInstructions.Parts),
+		}
+		openaiRequest.Messages = append([]dto.Message{systemMessage}, openaiRequest.Messages...)
+	}
+
+	return openaiRequest, nil
+}
+
+func convertGeminiRoleToOpenAI(geminiRole string) string {
+	switch geminiRole {
+	case "user":
+		return "user"
+	case "model":
+		return "assistant"
+	case "function":
+		return "function"
+	default:
+		return "user"
+	}
+}
+
+func extractTextFromGeminiParts(parts []dto.GeminiPart) string {
+	var texts []string
+	for _, part := range parts {
+		if part.Text != "" {
+			texts = append(texts, part.Text)
+		}
+	}
+	return strings.Join(texts, "\n")
+}
+
+// ResponseOpenAI2Gemini 将 OpenAI 响应转换为 Gemini 格式
+func ResponseOpenAI2Gemini(openAIResponse *dto.OpenAITextResponse, info *relaycommon.RelayInfo) *dto.GeminiChatResponse {
+	geminiResponse := &dto.GeminiChatResponse{
+		Candidates: make([]dto.GeminiChatCandidate, 0, len(openAIResponse.Choices)),
+		PromptFeedback: dto.GeminiChatPromptFeedback{
+			SafetyRatings: []dto.GeminiChatSafetyRating{},
+		},
+		UsageMetadata: dto.GeminiUsageMetadata{
+			PromptTokenCount:     openAIResponse.PromptTokens,
+			CandidatesTokenCount: openAIResponse.CompletionTokens,
+			TotalTokenCount:      openAIResponse.PromptTokens + openAIResponse.CompletionTokens,
+		},
+	}
+
+	for _, choice := range openAIResponse.Choices {
+		candidate := dto.GeminiChatCandidate{
+			Index:         int64(choice.Index),
+			SafetyRatings: []dto.GeminiChatSafetyRating{},
+		}
+
+		// 设置结束原因
+		var finishReason string
+		switch choice.FinishReason {
+		case "stop":
+			finishReason = "STOP"
+		case "length":
+			finishReason = "MAX_TOKENS"
+		case "content_filter":
+			finishReason = "SAFETY"
+		case "tool_calls":
+			finishReason = "STOP"
+		default:
+			finishReason = "STOP"
+		}
+		candidate.FinishReason = &finishReason
+
+		// 转换消息内容
+		content := dto.GeminiChatContent{
+			Role:  "model",
+			Parts: make([]dto.GeminiPart, 0),
+		}
+
+		// 处理工具调用
+		toolCalls := choice.Message.ParseToolCalls()
+		if len(toolCalls) > 0 {
+			for _, toolCall := range toolCalls {
+				// 解析参数
+				var args map[string]interface{}
+				if toolCall.Function.Arguments != "" {
+					if err := json.Unmarshal([]byte(toolCall.Function.Arguments), &args); err != nil {
+						args = map[string]interface{}{"arguments": toolCall.Function.Arguments}
+					}
+				} else {
+					args = make(map[string]interface{})
+				}
+
+				part := dto.GeminiPart{
+					FunctionCall: &dto.FunctionCall{
+						FunctionName: toolCall.Function.Name,
+						Arguments:    args,
+					},
+				}
+				content.Parts = append(content.Parts, part)
+			}
+		} else {
+			// 处理文本内容
+			textContent := choice.Message.StringContent()
+			if textContent != "" {
+				part := dto.GeminiPart{
+					Text: textContent,
+				}
+				content.Parts = append(content.Parts, part)
+			}
+		}
+
+		candidate.Content = content
+		geminiResponse.Candidates = append(geminiResponse.Candidates, candidate)
+	}
+
+	return geminiResponse
+}
+
+// StreamResponseOpenAI2Gemini 将 OpenAI 流式响应转换为 Gemini 格式
+func StreamResponseOpenAI2Gemini(openAIResponse *dto.ChatCompletionsStreamResponse, info *relaycommon.RelayInfo) *dto.GeminiChatResponse {
+	// 检查是否有实际内容或结束标志
+	hasContent := false
+	hasFinishReason := false
+	for _, choice := range openAIResponse.Choices {
+		if len(choice.Delta.GetContentString()) > 0 || (choice.Delta.ToolCalls != nil && len(choice.Delta.ToolCalls) > 0) {
+			hasContent = true
+		}
+		if choice.FinishReason != nil {
+			hasFinishReason = true
+		}
+	}
+
+	// 如果没有实际内容且没有结束标志，跳过。主要针对 openai 流响应开头的空数据
+	if !hasContent && !hasFinishReason {
+		return nil
+	}
+
+	geminiResponse := &dto.GeminiChatResponse{
+		Candidates: make([]dto.GeminiChatCandidate, 0, len(openAIResponse.Choices)),
+		PromptFeedback: dto.GeminiChatPromptFeedback{
+			SafetyRatings: []dto.GeminiChatSafetyRating{},
+		},
+		UsageMetadata: dto.GeminiUsageMetadata{
+			PromptTokenCount:     info.PromptTokens,
+			CandidatesTokenCount: 0, // 流式响应中可能没有完整的 usage 信息
+			TotalTokenCount:      info.PromptTokens,
+		},
+	}
+
+	for _, choice := range openAIResponse.Choices {
+		candidate := dto.GeminiChatCandidate{
+			Index:         int64(choice.Index),
+			SafetyRatings: []dto.GeminiChatSafetyRating{},
+		}
+
+		// 设置结束原因
+		if choice.FinishReason != nil {
+			var finishReason string
+			switch *choice.FinishReason {
+			case "stop":
+				finishReason = "STOP"
+			case "length":
+				finishReason = "MAX_TOKENS"
+			case "content_filter":
+				finishReason = "SAFETY"
+			case "tool_calls":
+				finishReason = "STOP"
+			default:
+				finishReason = "STOP"
+			}
+			candidate.FinishReason = &finishReason
+		}
+
+		// 转换消息内容
+		content := dto.GeminiChatContent{
+			Role:  "model",
+			Parts: make([]dto.GeminiPart, 0),
+		}
+
+		// 处理工具调用
+		if choice.Delta.ToolCalls != nil {
+			for _, toolCall := range choice.Delta.ToolCalls {
+				// 解析参数
+				var args map[string]interface{}
+				if toolCall.Function.Arguments != "" {
+					if err := json.Unmarshal([]byte(toolCall.Function.Arguments), &args); err != nil {
+						args = map[string]interface{}{"arguments": toolCall.Function.Arguments}
+					}
+				} else {
+					args = make(map[string]interface{})
+				}
+
+				part := dto.GeminiPart{
+					FunctionCall: &dto.FunctionCall{
+						FunctionName: toolCall.Function.Name,
+						Arguments:    args,
+					},
+				}
+				content.Parts = append(content.Parts, part)
+			}
+		} else {
+			// 处理文本内容
+			textContent := choice.Delta.GetContentString()
+			if textContent != "" {
+				part := dto.GeminiPart{
+					Text: textContent,
+				}
+				content.Parts = append(content.Parts, part)
+			}
+		}
+
+		candidate.Content = content
+		geminiResponse.Candidates = append(geminiResponse.Candidates, candidate)
+	}
+
+	return geminiResponse
+}

types/error.go

@@ -76,12 +76,13 @@ const (
 )
 
 type NewAPIError struct {
-	Err        error
-	RelayError any
-	skipRetry  bool
-	errorType  ErrorType
-	errorCode  ErrorCode
-	StatusCode int
+	Err            error
+	RelayError     any
+	skipRetry      bool
+	recordErrorLog *bool
+	errorType      ErrorType
+	errorCode      ErrorCode
+	StatusCode     int
 }
 
 func (e *NewAPIError) GetErrorCode() ErrorCode {
@@ -278,3 +279,20 @@ func ErrOptionWithSkipRetry() NewAPIErrorOptions {
 		e.skipRetry = true
 	}
 }
+
+func ErrOptionWithNoRecordErrorLog() NewAPIErrorOptions {
+	return func(e *NewAPIError) {
+		e.recordErrorLog = common.GetPointer(false)
+	}
+}
+
+func IsRecordErrorLog(e *NewAPIError) bool {
+	if e == nil {
+		return false
+	}
+	if e.recordErrorLog == nil {
+		// default to true if not set
+		return true
+	}
+	return *e.recordErrorLog
+}

web/src/components/common/JSONEditor.js

@@ -65,7 +65,8 @@ const JSONEditor = ({
         const keyCount = Object.keys(parsed).length;
         return keyCount > 10 ? 'manual' : 'visual';
       } catch (error) {
-        return 'visual';
+        // JSON无效时默认显示手动编辑模式
+        return 'manual';
       }
     }
     return 'visual';
@@ -201,6 +202,18 @@ const JSONEditor = ({
 
   // 渲染键值对编辑器
   const renderKeyValueEditor = () => {
+    if (typeof jsonData !== 'object' || jsonData === null) {
+      return (
+        <div className="text-center py-6 px-4">
+          <div className="text-gray-400 mb-2">
+            <IconCode size={32} />
+          </div>
+          <Text type="tertiary" className="text-gray-500 text-sm">
+            {t('无效的JSON数据，请检查格式')}
+          </Text>
+        </div>
+      );
+    }
     const entries = Object.entries(jsonData);
     
     return (

web/src/components/table/channels/modals/EditChannelModal.jsx

@@ -154,6 +154,7 @@ const EditChannelModal = (props) => {
   const [isMultiKeyChannel, setIsMultiKeyChannel] = useState(false);
   const [channelSearchValue, setChannelSearchValue] = useState('');
   const [useManualInput, setUseManualInput] = useState(false); // 是否使用手动输入模式
+  const [keyMode, setKeyMode] = useState('append'); // 密钥模式：replace（覆盖）或 append（追加）
   // 渠道额外设置状态
   const [channelSettings, setChannelSettings] = useState({
     force_format: false,
@@ -560,6 +561,12 @@ const EditChannelModal = (props) => {
         pass_through_body_enabled: false,
         system_prompt: '',
       });
+      // 重置密钥模式状态
+      setKeyMode('append');
+      // 清空表单中的key_mode字段
+      if (formApiRef.current) {
+        formApiRef.current.setValue('key_mode', undefined);
+      }
     }
   }, [props.visible, channelId]);
 
@@ -725,6 +732,7 @@ const EditChannelModal = (props) => {
       res = await API.put(`/api/channel/`, {
         ...localInputs,
         id: parseInt(channelId),
+        key_mode: isMultiKeyChannel ? keyMode : undefined, // 只在多key模式下传递
       });
     } else {
       res = await API.post(`/api/channel/`, {
@@ -787,55 +795,59 @@ const EditChannelModal = (props) => {
   const batchAllowed = !isEdit || isMultiKeyChannel;
   const batchExtra = batchAllowed ? (
     <Space>
-      <Checkbox
-        disabled={isEdit}
-        checked={batch}
-        onChange={(e) => {
-          const checked = e.target.checked;
+      {!isEdit && (
+        <Checkbox
+          disabled={isEdit}
+          checked={batch}
+          onChange={(e) => {
+            const checked = e.target.checked;
 
-          if (!checked && vertexFileList.length > 1) {
-            Modal.confirm({
-              title: t('切换为单密钥模式'),
-              content: t('将仅保留第一个密钥文件，其余文件将被移除，是否继续？'),
-              onOk: () => {
-                const firstFile = vertexFileList[0];
-                const firstKey = vertexKeys[0] ? [vertexKeys[0]] : [];
+            if (!checked && vertexFileList.length > 1) {
+              Modal.confirm({
+                title: t('切换为单密钥模式'),
+                content: t('将仅保留第一个密钥文件，其余文件将被移除，是否继续？'),
+                onOk: () => {
+                  const firstFile = vertexFileList[0];
+                  const firstKey = vertexKeys[0] ? [vertexKeys[0]] : [];
 
-                setVertexFileList([firstFile]);
-                setVertexKeys(firstKey);
+                  setVertexFileList([firstFile]);
+                  setVertexKeys(firstKey);
 
-                formApiRef.current?.setValue('vertex_files', [firstFile]);
-                setInputs((prev) => ({ ...prev, vertex_files: [firstFile] }));
+                  formApiRef.current?.setValue('vertex_files', [firstFile]);
+                  setInputs((prev) => ({ ...prev, vertex_files: [firstFile] }));
 
-                setBatch(false);
-                setMultiToSingle(false);
-                setMultiKeyMode('random');
-              },
-              onCancel: () => {
-                setBatch(true);
-              },
-              centered: true,
-            });
-            return;
-          }
-
-          setBatch(checked);
-          if (!checked) {
-            setMultiToSingle(false);
-            setMultiKeyMode('random');
-          } else {
-            // 批量模式下禁用手动输入，并清空手动输入的内容
-            setUseManualInput(false);
-            if (inputs.type === 41) {
-              // 清空手动输入的密钥内容
-              if (formApiRef.current) {
-                formApiRef.current.setValue('key', '');
-              }
-              handleInputChange('key', '');
+                  setBatch(false);
+                  setMultiToSingle(false);
+                  setMultiKeyMode('random');
+                },
+                onCancel: () => {
+                  setBatch(true);
+                },
+                centered: true,
+              });
+              return;
             }
-          }
-        }}
-      >{t('批量创建')}</Checkbox>
+
+            setBatch(checked);
+            if (!checked) {
+              setMultiToSingle(false);
+              setMultiKeyMode('random');
+            } else {
+              // 批量模式下禁用手动输入，并清空手动输入的内容
+              setUseManualInput(false);
+              if (inputs.type === 41) {
+                // 清空手动输入的密钥内容
+                if (formApiRef.current) {
+                  formApiRef.current.setValue('key', '');
+                }
+                handleInputChange('key', '');
+              }
+            }
+          }}
+        >
+          {t('批量创建')}
+        </Checkbox>
+      )}
       {batch && (
         <Checkbox disabled={isEdit} checked={multiToSingle} onChange={() => {
           setMultiToSingle(prev => !prev);
@@ -1032,7 +1044,16 @@ const EditChannelModal = (props) => {
                         autosize
                         autoComplete='new-password'
                         onChange={(value) => handleInputChange('key', value)}
-                        extraText={batchExtra}
+                        extraText={
+                          <div className="flex items-center gap-2">
+                            {isEdit && isMultiKeyChannel && keyMode === 'append' && (
+                              <Text type="warning" size="small">
+                                {t('追加模式：新密钥将添加到现有密钥列表的末尾')}
+                              </Text>
+                            )}
+                            {batchExtra}
+                          </div>
+                        }
                         showClear
                       />
                     )
@@ -1099,6 +1120,11 @@ const EditChannelModal = (props) => {
                                   <Text type="tertiary" size="small">
                                     {t('请输入完整的 JSON 格式密钥内容')}
                                   </Text>
+                                  {isEdit && isMultiKeyChannel && keyMode === 'append' && (
+                                    <Text type="warning" size="small">
+                                      {t('追加模式：新密钥将添加到现有密钥列表的末尾')}
+                                    </Text>
+                                  )}
                                   {batchExtra}
                                 </div>
                               }
@@ -1132,13 +1158,44 @@ const EditChannelModal = (props) => {
                           rules={isEdit ? [] : [{ required: true, message: t('请输入密钥') }]}
                           autoComplete='new-password'
                           onChange={(value) => handleInputChange('key', value)}
-                          extraText={batchExtra}
+                          extraText={
+                            <div className="flex items-center gap-2">
+                              {isEdit && isMultiKeyChannel && keyMode === 'append' && (
+                                <Text type="warning" size="small">
+                                  {t('追加模式：新密钥将添加到现有密钥列表的末尾')}
+                                </Text>
+                              )}
+                              {batchExtra}
+                            </div>
+                          }
                           showClear
                         />
                       )}
                     </>
                   )}
 
+                {isEdit && isMultiKeyChannel && (
+                        <Form.Select
+                          field='key_mode'
+                          label={t('密钥更新模式')}
+                          placeholder={t('请选择密钥更新模式')}
+                          optionList={[
+                            { label: t('追加到现有密钥'), value: 'append' },
+                            { label: t('覆盖现有密钥'), value: 'replace' },
+                          ]}
+                          style={{ width: '100%' }}
+                          value={keyMode}
+                          onChange={(value) => setKeyMode(value)}
+                          extraText={
+                            <Text type="tertiary" size="small">
+                              {keyMode === 'replace' 
+                                ? t('覆盖模式：将完全替换现有的所有密钥') 
+                                : t('追加模式：将新密钥添加到现有密钥列表末尾')
+                              }
+                            </Text>
+                          }
+                        />
+                  )}
                   {batch && multiToSingle && (
                     <>
                       <Form.Select