SpringAI2.0 对话记忆管理ChatMemory、Advisor 链与长期记忆架构前言多轮对话的核心挑战在构建 AI 应用时实现自然的对话体验至关重要。用户期望 AI 能够记住之前的对话上下文理解上下文而不是每次对话都从零开始。Spring AI 2.0 提供了一套完整的对话记忆管理机制让开发者能够轻松实现多轮对话和长期记忆。作为架构师我见过很多 AI 应用因为缺乏有效的记忆管理而导致用户体验差。Spring AI 2.0 通过 ChatMemory、Advisor 链等机制让记忆管理变得简单而强大。一、MessageChatMemoryAdvisor 与 VectorStoreChatMemoryAdvisor 差异对比1.1 两种记忆类型Spring AI 2.0 提供两种主要的 ChatMemory 实现MessageChatMemoryAdvisor基于内存或 Redis 的简单记忆存储VectorStoreChatMemoryAdvisor基于向量存储的语义记忆检索1.2 MessageChatMemoryAdvisorServicepublicclassSimpleMemoryService{privatefinalChatClientchatClient;privatefinalChatMemorychatMemory;// 使用 MessageChatMemoryAdvisorpublicStringchat(StringconversationId,Stringmessage){returnchatClient.prompt().user(message).advisors(newMessageChatMemoryAdvisor(chatMemory),newPromptChatMemoryAdvisor(chatMemory)).advisors(a-a.param(CHAT_MEMORY_CONVERSATION_ID_KEY,conversationId).param(CHAT_MEMORY_RETRIEVE_SIZE_KEY,100)).call().content();}}1.3 VectorStoreChatMemoryAdvisorServicepublicclassSemanticMemoryService{privatefinalChatClientchatClient;privatefinalVectorStorememoryStore;// 使用 VectorStoreChatMemoryAdvisorpublicStringchatWithSemanticMemory(StringconversationId,Stringmessage){returnchatClient.prompt().user(message).advisors(newVectorStoreChatMemoryAdvisor(memoryStore,SearchRequest.builder().withTopK(10).withSimilarityThreshold(0.7).build())).advisors(a-a.param(CHAT_MEMORY_CONVERSATION_ID_KEY,conversationId)).call().content();}}1.4 对比表格特性MessageChatMemoryAdvisorVectorStoreChatMemoryAdvisor存储方式线性存储向量检索检索策略最近 N 条语义相似度内存占用较大较小适用场景短期对话长期记忆性能高中等二、记忆存储后端选型2.1 RedisChatMemoryConfigurationpublicclassRedisChatMemoryConfig{BeanpublicChatMemoryredisChatMemory(RedisConnectionFactoryconnectionFactory){RedisChatMemoryConfigconfigRedisChatMemoryConfig.builder().withConversationKeyPrefix(chat:).build();returnnewRedisChatMemory(connectionFactory,config);}}2.2 Neo4jChatMemoryConfigurationpublicclassNeo4jChatMemoryConfig{BeanpublicChatMemoryneo4jChatMemory(Driverneo4jDriver){Neo4jChatMemoryConfigconfigNeo4jChatMemoryConfig.builder().withMemoryNodeLabel(Memory).build();returnnewNeo4jChatMemory(neo4jDriver,config);}}2.3 JDBC 持久化ConfigurationpublicclassJdbcChatMemoryConfig{BeanpublicChatMemoryjdbcChatMemory(JdbcTemplatejdbcTemplate){returnnewJdbcChatMemory(jdbcTemplate);}}三、上下文修剪与记忆压缩3.1 上下文修剪ServicepublicclassPruningMemoryService{privatefinalChatMemorychatMemory;privatefinalintmaxMessages50;publicvoidaddMessage(StringconversationId,Messagemessage){ListMessagemessageschatMemory.get(conversationId,maxMessages10);if(messages.size()maxMessages){// 保留最近的消息ListMessagerecentmessages.subList(messages.size()-maxMessages,messages.size());// 更新记忆chatMemory.clear(conversationId);chatMemory.add(conversationId,recent);}chatMemory.add(conversationId,List.of(message));}}3.2 记忆压缩ServicepublicclassCompressedMemoryService{privatefinalChatClientchatClient;privatefinalChatMemorychatMemory;// 定期压缩记忆Scheduled(cron0 0 */4 * * ?)// 每 4 小时publicvoidcompressMemories(){ListStringconversationIdsgetAllConversationIds();for(StringconversationId:conversationIds){try{compressConversation(conversationId);}catch(Exceptione){log.error(压缩记忆失败{},conversationId,e);}}}privatevoidcompressConversation(StringconversationId){ListMessagemessageschatMemory.get(conversationId,null);if(messagesnull||messages.size()20){return;// 消息太少不需要压缩}// 提取关键信息StringsummarychatClient.prompt().user( 请总结以下对话的要点保持简洁 %s .formatted(formatMessages(messages))).call().content();// 创建压缩消息MessagesummaryMessagenewSystemMessage(之前的对话摘要summary);// 清空旧记忆只保留摘要和最近的消息chatMemory.clear(conversationId);chatMemory.add(conversationId,List.of(summaryMessage));ListMessagerecentmessages.subList(messages.size()-10,messages.size());chatMemory.add(conversationId,recent);}privateStringformatMessages(ListMessagemessages){returnmessages.stream().map(msg-[%s] %s.formatted(msg.getMessageType(),msg.getContent())).collect(Collectors.joining(\n));}}四、Advisor 执行顺序控制4.1 执行顺序用户请求 ↓ [Advisor 1] 提示词增强 ↓ [Advisor 2] 记忆检索历史 ↓ [Advisor 3] RAG 检索 ↓ [Advisor 4] 工具调用 ↓ ChatModel 处理 ↓ [Advisor 5] 响应后处理 ↓ 用户响应4.2 先历史后检索ServicepublicclassHistoryFirstService{publicStringchatHistoryFirst(StringconversationId,Stringmessage){returnchatClient.prompt().user(message).advisors(// 先检索历史对话newPromptChatMemoryAdvisor(chatMemory),// 再检索知识库newQuestionAnswerAdvisor(vectorStore)).advisors(a-a.param(CHAT_MEMORY_CONVERSATION_ID_KEY,conversationId).param(CHAT_MEMORY_RETRIEVE_SIZE_KEY,50)).call().content();}}4.3 先检索后历史ServicepublicclassKnowledgeFirstService{publicStringchatKnowledgeFirst(StringconversationId,Stringmessage){returnchatClient.prompt().user(message).advisors(// 先检索知识库newQuestionAnswerAdvisor(vectorStore),// 再检索历史对话newPromptChatMemoryAdvisor(chatMemory)).advisors(a-a.param(CHAT_MEMORY_CONVERSATION_ID_KEY,conversationId).param(CHAT_MEMORY_RETRIEVE_SIZE_KEY,20)).call().content();}}五、长期记忆架构5.1 三层记忆架构┌─────────────────────────────────────┐ │ 工作记忆10 条最近消息 │ │ - Redis ChatMemory │ │ - 快速访问 │ └─────────────────────────────────────┘ ↓ ┌─────────────────────────────────────┐ │ 短期记忆最近 24 小时 │ │ - VectorStoreChatMemory │ │ - 语义检索 │ └─────────────────────────────────────┘ ↓ ┌─────────────────────────────────────┐ │ 长期记忆用户偏好、重要事件 │ │ - 专用数据库 │ │ - 定期更新 │ └─────────────────────────────────────┘5.2 长期记忆服务ServicepublicclassLongTermMemoryService{privatefinalVectorStorememoryStore;privatefinalJdbcTemplatejdbcTemplate;privatefinalChatClientchatClient;// 保存重要信息到长期记忆publicvoidsaveImportantInfo(StringuserId,Stringinfo,doubleimportance){// 存储到向量数据库DocumentmemorynewDocument(info,Map.of(user_id,userId,importance,importance,created_at,Instant.now(),type,long_term));memoryStore.add(List.of(memory));// 更新用户画像updateUserProfile(userId,info);}// 从长期记忆检索publicListDocumentretrieveLongTermMemory(StringuserId,Stringquery){returnmemoryStore.similaritySearch(SearchRequest.query(query).withTopK(5).withFilterExpression(user_id %s.formatted(userId)));}// 更新用户画像privatevoidupdateUserProfile(StringuserId,Stringinfo){Stringsql INSERT INTO user_profiles (user_id, preferences, updated_at) VALUES (?, ?, ?) ON CONFLICT (user_id) DO UPDATE SET preferences user_profiles.preferences || ?, updated_at ? ;jdbcTemplate.update(sql,userId,info,Instant.now(),info,Instant.now());}}六、实战案例智能客服记忆系统ServicepublicclassCustomerServiceMemorySystem{privatefinalChatClientchatClient;privatefinalChatMemorychatMemory;privatefinalLongTermMemoryServicelongTermMemory;publicStringhandleCustomerInquiry(StringcustomerId,Stringinquiry){// 1. 检索长期记忆ListDocumentuserProfilelongTermMemory.retrieveLongTermMemory(customerId,用户偏好);// 2. 构建系统提示词StringsystemPromptbuildSystemPrompt(userProfile);// 3. 调用 ChatClientStringresponsechatClient.prompt().system(systemPrompt).user(inquiry).advisors(newPromptChatMemoryAdvisor(chatMemory),newQuestionAnswerAdvisor(knowledgeBase)).advisors(a-a.param(CHAT_MEMORY_CONVERSATION_ID_KEY,customerId)).call().content();// 4. 提取并保存重要信息extractAndSaveImportantInfo(customerId,inquiry,response);returnresponse;}privateStringbuildSystemPrompt(ListDocumentuserProfile){if(userProfile.isEmpty()){return你是一个专业的客服代表;}StringpreferencesuserProfile.stream().map(Document::getContent).collect(Collectors.joining(\n));return 你是一个专业的客服代表。 用户偏好 %s 请根据用户偏好提供个性化的服务。 .formatted(preferences);}privatevoidextractAndSaveImportantInfo(StringcustomerId,Stringinquiry,Stringresponse){Stringprompt 从以下对话中提取重要的用户信息 用户提问%s 客服回答%s 重要信息包括姓名、联系方式、偏好、投诉等 .formatted(inquiry,response);StringextractedInfochatClient.prompt().user(prompt).call().content();if(!extractedInfo.isBlank()){longTermMemory.saveImportantInfo(customerId,extractedInfo,0.8// 重要性评分);}}}总结Spring AI 2.0 的对话记忆管理机制让构建自然的 AI 对话应用变得简单。通过合理选择 ChatMemory 实现、优化 Advisor 执行顺序、实现记忆压缩我们可以构建出高效、智能的对话系统。作为架构师建议混合使用结合短期和长期记忆定期压缩避免记忆无限增长语义检索使用向量存储提高检索精度个性化基于用户偏好定制对话性能优化合理设置记忆大小和检索数量参考资料Spring AI ChatMemory 文档https://docs.spring.io/spring-ai/reference/api/chatclient-memory.htmlRedis Memory 文档https://docs.spring.io/spring-ai/reference/api/vectordbs/redis-vector-store.html