🎨 1.优化前端渲染功能（html和对话消息格式）

2.优化流式传输，新增流式渲染功能
3.优化正则处理逻辑
4.新增context budget管理系统
5.优化对话消息失败处理逻辑
6.新增前端卡功能（待完整测试）

server/service/app/conversation.go

@@ -3,6 +3,7 @@ package app
 import (
 	"bufio"
 	"bytes"
+	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
@@ -371,21 +372,15 @@ func (s *ConversationService) SendMessage(userID, conversationID uint, req *requ
 		return nil, err
 	}
 
-	// 获取对话历史（最近10条）
+	// 获取完整对话历史（context 管理由 callAIService 内部处理）
 	var messages []app.Message
 	err = global.GVA_DB.Where("conversation_id = ?", conversationID).
-		Order("created_at DESC").
-		Limit(10).
+		Order("created_at ASC").
 		Find(&messages).Error
 	if err != nil {
 		return nil, err
 	}
 
-	// 反转消息顺序（从旧到新）
-	for i, j := 0, len(messages)-1; i < j; i, j = i+1, j-1 {
-		messages[i], messages[j] = messages[j], messages[i]
-	}
-
 	// 调用 AI 服务获取回复
 	aiResponse, err := s.callAIService(conversation, character, messages)
 	if err != nil {
@@ -527,47 +522,26 @@ func (s *ConversationService) callAIService(conversation app.Conversation, chara
 		}
 	}
 
-	// 构建系统提示词（如果预设有系统提示词，则追加到角色卡提示词后）
-	systemPrompt := s.buildSystemPrompt(character)
-	if preset != nil && preset.SystemPrompt != "" {
-		systemPrompt = systemPrompt + "\n\n" + preset.SystemPrompt
-		global.GVA_LOG.Info("已追加预设的系统提示词")
+	// 构建消息列表（含 context 预算管理）
+	var presetSysPrompt string
+	if preset != nil {
+		presetSysPrompt = preset.SystemPrompt
 	}
+	wbEngine := &WorldbookEngine{}
+	apiMessages := s.buildAPIMessagesWithContextManagement(
+		messages, character, presetSysPrompt, wbEngine, conversation, &aiConfig, preset,
+	)
 
-	// 集成世界书触发引擎
-	if conversation.WorldbookEnabled && conversation.WorldbookID != nil {
-		global.GVA_LOG.Info(fmt.Sprintf("世界书已启用，ID: %d", *conversation.WorldbookID))
-
-		// 提取消息内容用于扫描
-		var messageContents []string
-		for _, msg := range messages {
-			messageContents = append(messageContents, msg.Content)
-		}
-
-		// 使用世界书引擎扫描并触发条目
-		engine := &WorldbookEngine{}
-		triggered, err := engine.ScanAndTrigger(*conversation.WorldbookID, messageContents)
-		if err != nil {
-			global.GVA_LOG.Warn(fmt.Sprintf("世界书触发失败: %v", err))
-		} else if len(triggered) > 0 {
-			global.GVA_LOG.Info(fmt.Sprintf("触发了 %d 个世界书条目", len(triggered)))
-			// 将触发的世界书内容注入到系统提示词
-			systemPrompt = engine.BuildPromptWithWorldbook(systemPrompt, triggered)
-		} else {
-			global.GVA_LOG.Info("没有触发任何世界书条目")
-		}
+	// 从 apiMessages 中提取 systemPrompt，供 Anthropic 独立参数使用
+	systemPrompt := ""
+	if len(apiMessages) > 0 && apiMessages[0]["role"] == "system" {
+		systemPrompt = apiMessages[0]["content"]
 	}
 
-	// 构建消息列表
-	apiMessages := s.buildAPIMessages(messages, systemPrompt)
-
 	// 打印发送给AI的完整内容
 	global.GVA_LOG.Info("========== 发送给AI的完整内容 ==========")
-	global.GVA_LOG.Info(fmt.Sprintf("系统提示词: %s", systemPrompt))
-	global.GVA_LOG.Info("消息列表:")
-	for i, msg := range apiMessages {
-		global.GVA_LOG.Info(fmt.Sprintf("  [%d] Role: %s, Content: %s", i, msg["role"], msg["content"]))
-	}
+	global.GVA_LOG.Info(fmt.Sprintf("系统提示词长度: %d 字符", len(systemPrompt)))
+	global.GVA_LOG.Info(fmt.Sprintf("历史消息条数: %d", len(apiMessages)-1))
 	global.GVA_LOG.Info("==========================================")
 
 	// 确定使用的模型：如果用户在设置中指定了AI配置，则使用该配置的默认模型
@@ -735,7 +709,7 @@ func (s *ConversationService) getWeekdayInChinese(weekday time.Weekday) string {
 }
 
 // SendMessageStream 流式发送消息并获取 AI 回复
-func (s *ConversationService) SendMessageStream(userID, conversationID uint, req *request.SendMessageRequest, streamChan chan string, doneChan chan bool) error {
+func (s *ConversationService) SendMessageStream(ctx context.Context, userID, conversationID uint, req *request.SendMessageRequest, streamChan chan string, doneChan chan bool) error {
 	defer close(streamChan)
 	defer close(doneChan)
 
@@ -796,21 +770,15 @@ func (s *ConversationService) SendMessageStream(userID, conversationID uint, req
 		return err
 	}
 
-	// 获取对话历史（最近10条）
+	// 获取完整对话历史（context 管理由 buildAPIMessagesWithContextManagement 处理）
 	var messages []app.Message
 	err = global.GVA_DB.Where("conversation_id = ?", conversationID).
-		Order("created_at DESC").
-		Limit(10).
+		Order("created_at ASC").
 		Find(&messages).Error
 	if err != nil {
 		return err
 	}
 
-	// 反转消息顺序（从旧到新）
-	for i, j := 0, len(messages)-1; i < j; i, j = i+1, j-1 {
-		messages[i], messages[j] = messages[j], messages[i]
-	}
-
 	// 获取 AI 配置
 	var aiConfig app.AIConfig
 	var configID uint
@@ -857,42 +825,26 @@ func (s *ConversationService) SendMessageStream(userID, conversationID uint, req
 		}
 	}
 
-	// 构建系统提示词（应用预设）
-	systemPrompt := s.buildSystemPrompt(character)
-	if streamPreset != nil && streamPreset.SystemPrompt != "" {
-		systemPrompt = systemPrompt + "\n\n" + streamPreset.SystemPrompt
+	// 构建消息列表（含 context 预算管理）
+	var streamPresetSysPrompt string
+	if streamPreset != nil {
+		streamPresetSysPrompt = streamPreset.SystemPrompt
 	}
+	streamWbEngine := &WorldbookEngine{}
+	apiMessages := s.buildAPIMessagesWithContextManagement(
+		messages, character, streamPresetSysPrompt, streamWbEngine, conversation, &aiConfig, streamPreset,
+	)
 
-	// 集成世界书触发引擎（流式传输）
-	if conversation.WorldbookEnabled && conversation.WorldbookID != nil {
-		global.GVA_LOG.Info(fmt.Sprintf("[流式传输] 世界书已启用，ID: %d", *conversation.WorldbookID))
-
-		var messageContents []string
-		for _, msg := range messages {
-			messageContents = append(messageContents, msg.Content)
-		}
-
-		engine := &WorldbookEngine{}
-		triggeredEntries, wbErr := engine.ScanAndTrigger(*conversation.WorldbookID, messageContents)
-		if wbErr != nil {
-			global.GVA_LOG.Warn(fmt.Sprintf("[流式传输] 世界书触发失败: %v", wbErr))
-		} else if len(triggeredEntries) > 0 {
-			global.GVA_LOG.Info(fmt.Sprintf("[流式传输] 触发了 %d 个世界书条目", len(triggeredEntries)))
-			systemPrompt = engine.BuildPromptWithWorldbook(systemPrompt, triggeredEntries)
-		} else {
-			global.GVA_LOG.Info("[流式传输] 没有触发任何世界书条目")
-		}
+	// 从 apiMessages 中提取 systemPrompt，供 Anthropic 独立参数使用
+	systemPrompt := ""
+	if len(apiMessages) > 0 && apiMessages[0]["role"] == "system" {
+		systemPrompt = apiMessages[0]["content"]
 	}
 
-	apiMessages := s.buildAPIMessages(messages, systemPrompt)
-
 	// 打印发送给AI的完整内容（流式传输）
 	global.GVA_LOG.Info("========== [流式传输] 发送给AI的完整内容 ==========")
-	global.GVA_LOG.Info(fmt.Sprintf("系统提示词: %s", systemPrompt))
-	global.GVA_LOG.Info("消息列表:")
-	for i, msg := range apiMessages {
-		global.GVA_LOG.Info(fmt.Sprintf("  [%d] Role: %s, Content: %s", i, msg["role"], msg["content"]))
-	}
+	global.GVA_LOG.Info(fmt.Sprintf("[流式传输] 系统提示词长度: %d 字符", len(systemPrompt)))
+	global.GVA_LOG.Info(fmt.Sprintf("[流式传输] 历史消息条数: %d", len(apiMessages)-1))
 	global.GVA_LOG.Info("==========================================")
 
 	// 确定使用的模型
@@ -910,15 +862,21 @@ func (s *ConversationService) SendMessageStream(userID, conversationID uint, req
 	var fullContent string
 	switch aiConfig.Provider {
 	case "openai", "custom":
-		fullContent, err = s.callOpenAIAPIStream(&aiConfig, model, apiMessages, streamPreset, streamChan)
+		fullContent, err = s.callOpenAIAPIStream(ctx, &aiConfig, model, apiMessages, streamPreset, streamChan)
 	case "anthropic":
-		fullContent, err = s.callAnthropicAPIStream(&aiConfig, model, apiMessages, systemPrompt, streamPreset, streamChan)
+		fullContent, err = s.callAnthropicAPIStream(ctx, &aiConfig, model, apiMessages, systemPrompt, streamPreset, streamChan)
 	default:
 		return fmt.Errorf("不支持的 AI 提供商: %s", aiConfig.Provider)
 	}
 
 	if err != nil {
 		global.GVA_LOG.Error(fmt.Sprintf("========== [流式传输] AI返回错误 ==========\n%v\n==========================================", err))
+		// AI 调用失败，回滚已写入的用户消息，避免孤立记录残留在数据库
+		if delErr := global.GVA_DB.Delete(&userMessage).Error; delErr != nil {
+			global.GVA_LOG.Error(fmt.Sprintf("[流式传输] 回滚用户消息失败: %v", delErr))
+		} else {
+			global.GVA_LOG.Info("[流式传输] 已回滚用户消息")
+		}
 		return err
 	}
 
@@ -962,8 +920,9 @@ func (s *ConversationService) SendMessageStream(userID, conversationID uint, req
 }
 
 // callOpenAIAPIStream 调用 OpenAI API 流式传输
-func (s *ConversationService) callOpenAIAPIStream(config *app.AIConfig, model string, messages []map[string]string, preset *app.AIPreset, streamChan chan string) (string, error) {
-	client := &http.Client{Timeout: 120 * time.Second}
+func (s *ConversationService) callOpenAIAPIStream(ctx context.Context, config *app.AIConfig, model string, messages []map[string]string, preset *app.AIPreset, streamChan chan string) (string, error) {
+	// 不设 Timeout：生命周期由调用方传入的 ctx 控制（客户端断连时自动取消）
+	client := &http.Client{}
 
 	if model == "" {
 		model = config.DefaultModel
@@ -1025,7 +984,7 @@ func (s *ConversationService) callOpenAIAPIStream(config *app.AIConfig, model st
 	}
 
 	endpoint := config.BaseURL + "/chat/completions"
-	req, err := http.NewRequest("POST", endpoint, bytes.NewBuffer(bodyBytes))
+	req, err := http.NewRequestWithContext(ctx, "POST", endpoint, bytes.NewBuffer(bodyBytes))
 	if err != nil {
 		return "", fmt.Errorf("创建请求失败: %v", err)
 	}
@@ -1035,6 +994,10 @@ func (s *ConversationService) callOpenAIAPIStream(config *app.AIConfig, model st
 
 	resp, err := client.Do(req)
 	if err != nil {
+		// 客户端主动断开时 ctx 被取消，不算真正的错误
+		if ctx.Err() != nil {
+			return "", nil
+		}
 		return "", fmt.Errorf("请求失败: %v", err)
 	}
 	defer resp.Body.Close()
@@ -1050,49 +1013,51 @@ func (s *ConversationService) callOpenAIAPIStream(config *app.AIConfig, model st
 
 	for {
 		line, err := reader.ReadString('\n')
+		// 先处理本次读到的数据（EOF 时可能仍携带最后一行内容）
+		if line != "" {
+			trimmed := strings.TrimSpace(line)
+			if trimmed != "" && trimmed != "data: [DONE]" && strings.HasPrefix(trimmed, "data: ") {
+				data := strings.TrimPrefix(trimmed, "data: ")
+
+				var streamResp struct {
+					Choices []struct {
+						Delta struct {
+							Content string `json:"content"`
+						} `json:"delta"`
+					} `json:"choices"`
+				}
+
+				if jsonErr := json.Unmarshal([]byte(data), &streamResp); jsonErr == nil {
+					if len(streamResp.Choices) > 0 {
+						content := streamResp.Choices[0].Delta.Content
+						if content != "" {
+							fullContent.WriteString(content)
+							streamChan <- content
+						}
+					}
+				}
+			}
+		}
+		// 再检查读取错误
 		if err != nil {
 			if err == io.EOF {
 				break
 			}
+			// ctx 被取消（客户端断开）时不算真正的流读取错误
+			if ctx.Err() != nil {
+				return fullContent.String(), nil
+			}
 			return "", fmt.Errorf("读取流失败: %v", err)
 		}
-
-		line = strings.TrimSpace(line)
-		if line == "" || line == "data: [DONE]" {
-			continue
-		}
-
-		if strings.HasPrefix(line, "data: ") {
-			data := strings.TrimPrefix(line, "data: ")
-
-			var streamResp struct {
-				Choices []struct {
-					Delta struct {
-						Content string `json:"content"`
-					} `json:"delta"`
-				} `json:"choices"`
-			}
-
-			if err := json.Unmarshal([]byte(data), &streamResp); err != nil {
-				continue
-			}
-
-			if len(streamResp.Choices) > 0 {
-				content := streamResp.Choices[0].Delta.Content
-				if content != "" {
-					fullContent.WriteString(content)
-					streamChan <- content
-				}
-			}
-		}
 	}
 
 	return fullContent.String(), nil
 }
 
 // callAnthropicAPIStream 调用 Anthropic API 流式传输
-func (s *ConversationService) callAnthropicAPIStream(config *app.AIConfig, model string, messages []map[string]string, systemPrompt string, preset *app.AIPreset, streamChan chan string) (string, error) {
-	client := &http.Client{Timeout: 120 * time.Second}
+func (s *ConversationService) callAnthropicAPIStream(ctx context.Context, config *app.AIConfig, model string, messages []map[string]string, systemPrompt string, preset *app.AIPreset, streamChan chan string) (string, error) {
+	// 不设 Timeout：生命周期由调用方传入的 ctx 控制（客户端断连时自动取消）
+	client := &http.Client{}
 
 	if model == "" {
 		model = config.DefaultModel
@@ -1152,7 +1117,7 @@ func (s *ConversationService) callAnthropicAPIStream(config *app.AIConfig, model
 	}
 
 	endpoint := config.BaseURL + "/messages"
-	req, err := http.NewRequest("POST", endpoint, bytes.NewBuffer(bodyBytes))
+	req, err := http.NewRequestWithContext(ctx, "POST", endpoint, bytes.NewBuffer(bodyBytes))
 	if err != nil {
 		return "", fmt.Errorf("创建请求失败: %v", err)
 	}
@@ -1163,6 +1128,10 @@ func (s *ConversationService) callAnthropicAPIStream(config *app.AIConfig, model
 
 	resp, err := client.Do(req)
 	if err != nil {
+		// 客户端主动断开时 ctx 被取消，不算真正的错误
+		if ctx.Err() != nil {
+			return "", nil
+		}
 		return "", fmt.Errorf("请求失败: %v", err)
 	}
 	defer resp.Body.Close()
@@ -1178,38 +1147,39 @@ func (s *ConversationService) callAnthropicAPIStream(config *app.AIConfig, model
 
 	for {
 		line, err := reader.ReadString('\n')
+		// 先处理本次读到的数据（EOF 时可能仍携带最后一行内容）
+		if line != "" {
+			trimmed := strings.TrimSpace(line)
+			if trimmed != "" && strings.HasPrefix(trimmed, "data: ") {
+				data := strings.TrimPrefix(trimmed, "data: ")
+
+				var streamResp struct {
+					Type  string `json:"type"`
+					Delta struct {
+						Type string `json:"type"`
+						Text string `json:"text"`
+					} `json:"delta"`
+				}
+
+				if jsonErr := json.Unmarshal([]byte(data), &streamResp); jsonErr == nil {
+					if streamResp.Type == "content_block_delta" && streamResp.Delta.Text != "" {
+						fullContent.WriteString(streamResp.Delta.Text)
+						streamChan <- streamResp.Delta.Text
+					}
+				}
+			}
+		}
+		// 再检查读取错误
 		if err != nil {
 			if err == io.EOF {
 				break
 			}
+			// ctx 被取消（客户端断开）时不算真正的流读取错误
+			if ctx.Err() != nil {
+				return fullContent.String(), nil
+			}
 			return "", fmt.Errorf("读取流失败: %v", err)
 		}
-
-		line = strings.TrimSpace(line)
-		if line == "" {
-			continue
-		}
-
-		if strings.HasPrefix(line, "data: ") {
-			data := strings.TrimPrefix(line, "data: ")
-
-			var streamResp struct {
-				Type  string `json:"type"`
-				Delta struct {
-					Type string `json:"type"`
-					Text string `json:"text"`
-				} `json:"delta"`
-			}
-
-			if err := json.Unmarshal([]byte(data), &streamResp); err != nil {
-				continue
-			}
-
-			if streamResp.Type == "content_block_delta" && streamResp.Delta.Text != "" {
-				fullContent.WriteString(streamResp.Delta.Text)
-				streamChan <- streamResp.Delta.Text
-			}
-		}
 	}
 
 	return fullContent.String(), nil
@@ -1243,19 +1213,16 @@ func (s *ConversationService) RegenerateMessage(userID, conversationID uint) (*r
 		})
 	}
 
-	// 获取删除后的消息历史
+	// 获取删除后的完整消息历史（context 管理由 callAIService 内部处理）
 	var messages []app.Message
 	err = global.GVA_DB.Where("conversation_id = ?", conversationID).
-		Order("created_at DESC").Limit(10).Find(&messages).Error
+		Order("created_at ASC").Find(&messages).Error
 	if err != nil {
 		return nil, err
 	}
 	if len(messages) == 0 {
 		return nil, errors.New("没有可用的消息历史")
 	}
-	for i, j := 0, len(messages)-1; i < j; i, j = i+1, j-1 {
-		messages[i], messages[j] = messages[j], messages[i]
-	}
 
 	aiResponse, err := s.callAIService(conversation, character, messages)
 	if err != nil {
@@ -1282,7 +1249,7 @@ func (s *ConversationService) RegenerateMessage(userID, conversationID uint) (*r
 }
 
 // RegenerateMessageStream 流式重新生成最后一条 AI 回复
-func (s *ConversationService) RegenerateMessageStream(userID, conversationID uint, streamChan chan string, doneChan chan bool) error {
+func (s *ConversationService) RegenerateMessageStream(ctx context.Context, userID, conversationID uint, streamChan chan string, doneChan chan bool) error {
 	defer close(streamChan)
 	defer close(doneChan)
 
@@ -1312,19 +1279,16 @@ func (s *ConversationService) RegenerateMessageStream(userID, conversationID uin
 		})
 	}
 
-	// 获取删除后的消息历史
+	// 获取删除后的完整消息历史（context 管理由 buildAPIMessagesWithContextManagement 处理）
 	var messages []app.Message
 	err = global.GVA_DB.Where("conversation_id = ?", conversationID).
-		Order("created_at DESC").Limit(10).Find(&messages).Error
+		Order("created_at ASC").Find(&messages).Error
 	if err != nil {
 		return err
 	}
 	if len(messages) == 0 {
 		return errors.New("没有可用的消息历史")
 	}
-	for i, j := 0, len(messages)-1; i < j; i, j = i+1, j-1 {
-		messages[i], messages[j] = messages[j], messages[i]
-	}
 
 	// 获取 AI 配置
 	var aiConfig app.AIConfig
@@ -1367,11 +1331,21 @@ func (s *ConversationService) RegenerateMessageStream(userID, conversationID uin
 		}
 	}
 
-	systemPrompt := s.buildSystemPrompt(character)
-	if preset != nil && preset.SystemPrompt != "" {
-		systemPrompt = systemPrompt + "\n\n" + preset.SystemPrompt
+	// 构建消息列表（含 context 预算管理）
+	var regenPresetSysPrompt string
+	if preset != nil {
+		regenPresetSysPrompt = preset.SystemPrompt
+	}
+	regenWbEngine := &WorldbookEngine{}
+	apiMessages := s.buildAPIMessagesWithContextManagement(
+		messages, character, regenPresetSysPrompt, regenWbEngine, conversation, &aiConfig, preset,
+	)
+
+	// 从 apiMessages 中提取 systemPrompt，供 Anthropic 独立参数使用
+	systemPrompt := ""
+	if len(apiMessages) > 0 && apiMessages[0]["role"] == "system" {
+		systemPrompt = apiMessages[0]["content"]
 	}
-	apiMessages := s.buildAPIMessages(messages, systemPrompt)
 
 	model := aiConfig.DefaultModel
 	if model == "" {
@@ -1384,14 +1358,27 @@ func (s *ConversationService) RegenerateMessageStream(userID, conversationID uin
 	var fullContent string
 	switch aiConfig.Provider {
 	case "openai", "custom":
-		fullContent, err = s.callOpenAIAPIStream(&aiConfig, model, apiMessages, preset, streamChan)
+		fullContent, err = s.callOpenAIAPIStream(ctx, &aiConfig, model, apiMessages, preset, streamChan)
 	case "anthropic":
-		fullContent, err = s.callAnthropicAPIStream(&aiConfig, model, apiMessages, systemPrompt, preset, streamChan)
+		fullContent, err = s.callAnthropicAPIStream(ctx, &aiConfig, model, apiMessages, systemPrompt, preset, streamChan)
 	default:
 		return fmt.Errorf("不支持的 AI 提供商: %s", aiConfig.Provider)
 	}
 
 	if err != nil {
+		// AI 调用失败，恢复刚才删除的 assistant 消息，避免数据永久丢失
+		if lastAssistantMsg.ID > 0 {
+			if restoreErr := global.GVA_DB.Unscoped().Model(&lastAssistantMsg).Update("deleted_at", nil).Error; restoreErr != nil {
+				global.GVA_LOG.Error(fmt.Sprintf("[重新生成] 恢复 assistant 消息失败: %v", restoreErr))
+			} else {
+				// 回滚 conversation 统计
+				global.GVA_DB.Model(&conversation).Updates(map[string]interface{}{
+					"message_count": gorm.Expr("message_count + 1"),
+					"token_count":   gorm.Expr("token_count + ?", lastAssistantMsg.TokenCount),
+				})
+				global.GVA_LOG.Info("[重新生成] 已恢复 assistant 消息")
+			}
+		}
 		return err
 	}
 
@@ -1437,9 +1424,333 @@ func (s *ConversationService) buildAPIMessages(messages []app.Message, systemPro
 	return apiMessages
 }
 
+// estimateTokens 粗略估算文本的 token 数（字符数 / 3，适用于中英混合文本）
+func estimateTokens(text string) int {
+	if text == "" {
+		return 0
+	}
+	// 中文字符约 1 char = 1 token，英文约 4 chars = 1 token
+	// 取中间值 1 char ≈ 0.75 token，即 chars * 4 / 3 的倒数 ≈ chars / 1.5
+	// 保守估算用 chars / 2 防止超出
+	n := len([]rune(text))
+	return (n + 1) / 2
+}
+
+// contextConfig 保存从 AIConfig.Settings 中解析出的上下文配置
+type contextConfig struct {
+	contextLength int // 模型上下文窗口大小（token 数）
+	maxTokens     int // 最大输出 token 数
+}
+
+// getContextConfig 从 AIConfig 中读取上下文配置，如果没有配置则使用默认值
+func getContextConfig(aiConfig *app.AIConfig, preset *app.AIPreset) contextConfig {
+	cfg := contextConfig{
+		contextLength: 200000, // 保守默认值
+		maxTokens:     2000,
+	}
+
+	// 从 preset 读取 max_tokens
+	if preset != nil && preset.MaxTokens > 0 {
+		cfg.maxTokens = preset.MaxTokens
+	}
+
+	// 从 AIConfig.Settings 读取 context_length
+	if len(aiConfig.Settings) > 0 {
+		var settings map[string]interface{}
+		if err := json.Unmarshal(aiConfig.Settings, &settings); err == nil {
+			if cl, ok := settings["context_length"].(float64); ok && cl > 0 {
+				cfg.contextLength = int(cl)
+			}
+		}
+	}
+
+	return cfg
+}
+
+// buildContextManagedSystemPrompt 按优先级构建 system prompt，超出 budget 时截断低优先级内容
+// 优先级（从高到低）：
+//  1. 核心人设（Name/Description/Personality/Scenario/SystemPrompt）
+//  2. Preset.SystemPrompt
+//  3. Worldbook 触发条目
+//  4. CharacterBook 内嵌条目
+//  5. MesExample（对话示例，最容易被截断）
+//
+// 返回构建好的 systemPrompt 以及消耗的 token 数
+func (s *ConversationService) buildContextManagedSystemPrompt(
+	character app.AICharacter,
+	presetSystemPrompt string,
+	worldbookEngine *WorldbookEngine,
+	conversation app.Conversation,
+	messageContents []string,
+	budget int,
+) (string, int) {
+	used := 0
+
+	// ── 优先级1：核心人设 ─────────────────────────────────────────────
+	core := fmt.Sprintf("你是 %s。", character.Name)
+	if character.Description != "" {
+		core += fmt.Sprintf("\n\n描述：%s", character.Description)
+	}
+	if character.Personality != "" {
+		core += fmt.Sprintf("\n\n性格：%s", character.Personality)
+	}
+	if character.Scenario != "" {
+		core += fmt.Sprintf("\n\n场景：%s", character.Scenario)
+	}
+	if character.SystemPrompt != "" {
+		core += fmt.Sprintf("\n\n系统提示：%s", character.SystemPrompt)
+	}
+	core += "\n\n请根据以上设定进行角色扮演，保持角色的性格和说话方式。"
+	core = s.applyMacroVariables(core, character)
+
+	coreTokens := estimateTokens(core)
+	if coreTokens >= budget {
+		// 极端情况：核心人设本身就超出 budget，截断到 budget
+		runes := []rune(core)
+		limit := budget * 2
+		if limit > len(runes) {
+			limit = len(runes)
+		}
+		core = string(runes[:limit])
+		global.GVA_LOG.Warn(fmt.Sprintf("[context] 核心人设超出 budget，已截断至 %d chars", limit))
+		return core, budget
+	}
+	used += coreTokens
+	prompt := core
+
+	// ── 优先级2：Preset.SystemPrompt ────────────────────────────────
+	if presetSystemPrompt != "" {
+		tokens := estimateTokens(presetSystemPrompt)
+		if used+tokens <= budget {
+			prompt += "\n\n" + presetSystemPrompt
+			used += tokens
+		} else {
+			// 尝试部分插入
+			remaining := budget - used
+			if remaining > 50 {
+				runes := []rune(presetSystemPrompt)
+				limit := remaining * 2
+				if limit > len(runes) {
+					limit = len(runes)
+				}
+				prompt += "\n\n" + string(runes[:limit])
+				used = budget
+			}
+			global.GVA_LOG.Warn(fmt.Sprintf("[context] Preset.SystemPrompt 因 budget 不足被截断（需要 %d tokens，剩余 %d）", tokens, budget-used))
+		}
+	}
+
+	if used >= budget {
+		return prompt, used
+	}
+
+	// ── 优先级3：世界书触发条目 ──────────────────────────────────────
+	if conversation.WorldbookEnabled && conversation.WorldbookID != nil && worldbookEngine != nil {
+		triggeredEntries, wbErr := worldbookEngine.ScanAndTrigger(*conversation.WorldbookID, messageContents)
+		if wbErr != nil {
+			global.GVA_LOG.Warn(fmt.Sprintf("[context] 世界书触发失败: %v", wbErr))
+		} else if len(triggeredEntries) > 0 {
+			wbHeader := "\n\n[World Information]"
+			wbSection := wbHeader
+			for _, te := range triggeredEntries {
+				if te.Entry == nil || te.Entry.Content == "" {
+					continue
+				}
+				line := fmt.Sprintf("\n- %s", te.Entry.Content)
+				lineTokens := estimateTokens(line)
+				if used+estimateTokens(wbSection)+lineTokens <= budget {
+					wbSection += line
+					used += lineTokens
+				} else {
+					global.GVA_LOG.Warn(fmt.Sprintf("[context] 世界书条目 (id=%d) 因 budget 不足被跳过", te.Entry.ID))
+					break
+				}
+			}
+			if wbSection != wbHeader {
+				prompt += wbSection
+			}
+		}
+	}
+
+	if used >= budget {
+		return prompt, used
+	}
+
+	// ── 优先级4：CharacterBook 内嵌条目 ──────────────────────────────
+	if len(character.CharacterBook) > 0 {
+		var characterBook map[string]interface{}
+		if err := json.Unmarshal(character.CharacterBook, &characterBook); err == nil {
+			if entries, ok := characterBook["entries"].([]interface{}); ok && len(entries) > 0 {
+				cbSection := "\n\n世界设定："
+				addedAny := false
+				for _, entry := range entries {
+					entryMap, ok := entry.(map[string]interface{})
+					if !ok {
+						continue
+					}
+					enabled := true
+					if enabledVal, ok := entryMap["enabled"].(bool); ok {
+						enabled = enabledVal
+					}
+					if !enabled {
+						continue
+					}
+					content, ok := entryMap["content"].(string)
+					if !ok || content == "" {
+						continue
+					}
+					line := fmt.Sprintf("\n- %s", content)
+					lineTokens := estimateTokens(line)
+					if used+estimateTokens(cbSection)+lineTokens <= budget {
+						cbSection += line
+						used += lineTokens
+						addedAny = true
+					} else {
+						global.GVA_LOG.Warn("[context] CharacterBook 条目因 budget 不足被跳过")
+						break
+					}
+				}
+				if addedAny {
+					prompt += cbSection
+				}
+			}
+		}
+	}
+
+	if used >= budget {
+		return prompt, used
+	}
+
+	// ── 优先级5：MesExample（对话示例，最低优先级）──────────────────
+	if character.MesExample != "" {
+		mesTokens := estimateTokens(character.MesExample)
+		prefix := "\n\n对话示例：\n"
+		prefixTokens := estimateTokens(prefix)
+		if used+prefixTokens+mesTokens <= budget {
+			prompt += prefix + character.MesExample
+			used += prefixTokens + mesTokens
+		} else {
+			// 尝试截断 MesExample
+			remaining := budget - used - prefixTokens
+			if remaining > 100 {
+				runes := []rune(character.MesExample)
+				limit := remaining * 2
+				if limit > len(runes) {
+					limit = len(runes)
+				}
+				prompt += prefix + string(runes[:limit])
+				used = budget
+				global.GVA_LOG.Warn(fmt.Sprintf("[context] MesExample 被截断（原始 %d tokens，保留约 %d tokens）", mesTokens, remaining))
+			} else {
+				global.GVA_LOG.Warn("[context] MesExample 因 budget 不足被完全跳过")
+			}
+		}
+	}
+
+	return prompt, used
+}
+
+// trimMessagesToBudget 从历史消息中按 token 预算选取最近的消息
+// 优先保留最新的消息，从后往前丢弃旧消息直到 token 数在 budget 内
+func trimMessagesToBudget(messages []app.Message, budget int) []app.Message {
+	if budget <= 0 {
+		return nil
+	}
+
+	// messages 已经是从旧到新的顺序
+	// 从最新消息开始往前累加，直到超出 budget
+	selected := make([]app.Message, 0, len(messages))
+	used := 0
+
+	for i := len(messages) - 1; i >= 0; i-- {
+		msg := messages[i]
+		if msg.Role == "system" {
+			continue
+		}
+		t := estimateTokens(msg.Content)
+		if used+t > budget {
+			global.GVA_LOG.Warn(fmt.Sprintf("[context] 历史消息已截断，保留最近 %d 条（共 %d 条），使用 %d tokens", len(selected), len(messages), used))
+			break
+		}
+		used += t
+		selected = append([]app.Message{msg}, selected...) // 保持时序
+	}
+
+	return selected
+}
+
+// buildAPIMessagesWithContextManagement 整合 context 管理，构建最终的 messages 列表
+// 返回 apiMessages 及各部分 token 统计日志
+func (s *ConversationService) buildAPIMessagesWithContextManagement(
+	allMessages []app.Message,
+	character app.AICharacter,
+	presetSystemPrompt string,
+	worldbookEngine *WorldbookEngine,
+	conversation app.Conversation,
+	aiConfig *app.AIConfig,
+	preset *app.AIPreset,
+) []map[string]string {
+	cfg := getContextConfig(aiConfig, preset)
+
+	// 安全边际：为输出保留 max_tokens，另加 200 token 缓冲
+	inputBudget := cfg.contextLength - cfg.maxTokens - 200
+	if inputBudget <= 0 {
+		inputBudget = cfg.contextLength / 2
+	}
+
+	// 为历史消息分配预算：system prompt 最多占用 60% 的 input budget
+	systemBudget := inputBudget * 60 / 100
+	historyBudget := inputBudget - systemBudget
+
+	// 提取消息内容用于世界书扫描
+	var messageContents []string
+	for _, msg := range allMessages {
+		messageContents = append(messageContents, msg.Content)
+	}
+
+	// 构建 system prompt（含 worldbook 注入，按优先级截断）
+	systemPrompt, systemTokens := s.buildContextManagedSystemPrompt(
+		character,
+		presetSystemPrompt,
+		worldbookEngine,
+		conversation,
+		messageContents,
+		systemBudget,
+	)
+
+	// 如果 system prompt 实际用量比预算少，把节省的预算让给历史消息
+	if systemTokens < systemBudget {
+		historyBudget += systemBudget - systemTokens
+	}
+
+	global.GVA_LOG.Info(fmt.Sprintf("[context] 配置：context_length=%d, max_tokens=%d, input_budget=%d, system=%d tokens, history_budget=%d",
+		cfg.contextLength, cfg.maxTokens, inputBudget, systemTokens, historyBudget))
+
+	// 按 token 预算裁剪历史消息
+	trimmedMessages := trimMessagesToBudget(allMessages, historyBudget)
+
+	// 构建最终 messages
+	apiMessages := make([]map[string]string, 0, len(trimmedMessages)+1)
+	apiMessages = append(apiMessages, map[string]string{
+		"role":    "system",
+		"content": systemPrompt,
+	})
+	for _, msg := range trimmedMessages {
+		if msg.Role == "system" {
+			continue
+		}
+		apiMessages = append(apiMessages, map[string]string{
+			"role":    msg.Role,
+			"content": msg.Content,
+		})
+	}
+
+	return apiMessages
+}
+
 // callOpenAIAPI 调用 OpenAI API
 func (s *ConversationService) callOpenAIAPI(config *app.AIConfig, model string, messages []map[string]string, preset *app.AIPreset) (string, error) {
-	client := &http.Client{Timeout: 120 * time.Second}
+	client := &http.Client{Timeout: 10 * time.Minute}
 
 	// 使用配置的模型或默认模型
 	if model == "" {
@@ -1559,7 +1870,7 @@ func (s *ConversationService) callOpenAIAPI(config *app.AIConfig, model string,
 
 // callAnthropicAPI 调用 Anthropic API
 func (s *ConversationService) callAnthropicAPI(config *app.AIConfig, model string, messages []map[string]string, systemPrompt string, preset *app.AIPreset) (string, error) {
-	client := &http.Client{Timeout: 120 * time.Second}
+	client := &http.Client{Timeout: 10 * time.Minute}
 
 	// 使用配置的模型或默认模型
 	if model == "" {

server/service/app/regex_script.go

@@ -69,6 +69,10 @@ func (s *RegexScriptService) GetRegexScriptList(userID uint, req *request.GetReg
 		db = db.Where("scope = ?", *req.Scope)
 	}
 
+	if req.OwnerCharID != nil {
+		db = db.Where("owner_char_id = ?", *req.OwnerCharID)
+	}
+
 	if err := db.Count(&total).Error; err != nil {
 		return nil, 0, err
 	}