Enable Anthropic's prompt caching (cache_control) for Claude models accessed (#850)

Author: reckless-huang

providers/claude/chat.go

@@ -141,17 +141,27 @@ func ConvertFromChatOpenai(request *types.ChatCompletionRequest) (*ClaudeRequest
 		}
 	}
 
+	// 处理 system 字段（支持 cache_control）
 	systemMessage := ""
 	mgsLen := len(request.Messages) - 1
 	isThink := (request.OneOtherArg == "thinking" || request.Reasoning != nil)
 
+	// 如果请求中已经有 system 字段（如数组格式带 cache_control），直接使用
+	if request.System != nil {
+		claudeRequest.System = request.System
+	}
+
+	// 处理 messages
 	for index, msg := range request.Messages {
 		if isThink && index == mgsLen && (msg.Role == types.ChatMessageRoleAssistant || msg.Role == types.ChatMessageRoleSystem) {
 			msg.Role = types.ChatMessageRoleUser
 		}
 
 		if msg.Role == types.ChatMessageRoleSystem {
-			systemMessage += msg.StringContent()
+			// 如果没有预设的 system 字段，从 messages 中提取
+			if request.System == nil {
+				systemMessage += msg.StringContent()
+			}
 			continue
 		}
 		messageContent, err := convertMessageContent(&msg)
@@ -163,7 +173,8 @@ func ConvertFromChatOpenai(request *types.ChatCompletionRequest) (*ClaudeRequest
 		}
 	}
 
-	if systemMessage != "" {
+	// 如果没有预设的 system 字段，且从 messages 中提取到了 system message
+	if request.System == nil && systemMessage != "" {
 		claudeRequest.System = systemMessage
 	}
 
@@ -295,10 +306,15 @@ func convertMessageContent(msg *types.ChatCompletionMessage) (*Message, error) {
 	openaiContent := msg.ParseContent()
 	for _, part := range openaiContent {
 		if part.Type == types.ContentTypeText {
-			content = append(content, MessageContent{
+			msgContent := MessageContent{
 				Type: "text",
 				Text: part.Text,
-			})
+			}
+			// 传递 cache_control 字段
+			if msg.CacheControl != nil {
+				msgContent.CacheControl = msg.CacheControl
+			}
+			content = append(content, msgContent)
 			continue
 		}
 		if part.Type == types.ContentTypeImageURL {

types/chat.go

@@ -58,6 +58,7 @@ type ChatCompletionMessage struct {
 	Annotations      any                              `json:"annotations,omitempty"`
 	Image            []MultimediaData                 `json:"image,omitempty"`
 	Images           []ChatMessagePart                `json:"images,omitempty"`
+	CacheControl     any                              `json:"cache_control,omitempty"`
 }
 
 func (m ChatCompletionMessage) StringContent() string {
@@ -179,6 +180,7 @@ type FormatJsonSchema struct {
 type ChatCompletionRequest struct {
 	Model               string                        `json:"model" binding:"required"`
 	Messages            []ChatCompletionMessage       `json:"messages" binding:"required"`
+	System              any                           `json:"system,omitempty"`
 	MaxTokens           int                           `json:"max_tokens,omitempty"`
 	MaxCompletionTokens int                           `json:"max_completion_tokens,omitempty"`
 	Temperature         *float64                      `json:"temperature,omitempty"`