news 2026/5/8 12:16:56

.Net——AI智能体开发基于 Microsoft Agent Framework 实现第三方聊天历史存储

作者头像

张小明

前端开发工程师

1.2k 24
文章封面图
.Net——AI智能体开发基于 Microsoft Agent Framework 实现第三方聊天历史存储

基于 Microsoft Agent Framework 实现第三方聊天历史存储

理解 Microsoft Agent Framework

Microsoft Agent Framework 是一个用于构建对话式 AI 应用的框架,支持自然语言处理和上下文管理。其核心组件包括 Bot Framework SDK 和 Azure Bot Service,但默认聊天历史通常存储在 Azure 服务中。

具体实现可参考NetCoreKevin的kevin.AI.AgentFramework和KevinAIChatMessageStore服务模块

基于NET构建的现代化AI智能体Saas企业级架构:

项目地址:github:https://github.com/junkai-li/NetCoreKevin
Gitee: https://gitee.com/netkevin-li/NetCoreKevin

配置自定义存储提供程序
  1. **实现ChatMessageStore**
    创建自定义存储类继承Microsoft.Bot.Builder.IStorage,需实现以下方法:

    publicsealedclassKevinChatMessageStore:ChatMessageStore{privateIKevinAIChatMessageStore_chatMessageStore;publicstringThreadDbKey{get;privateset;}publicKevinChatMessageStore(IKevinAIChatMessageStorevectorStore,JsonElementserializedStoreState,stringAIChatsId,JsonSerializerOptions?jsonSerializerOptions=null){this._chatMessageStore=vectorStore??thrownewArgumentNullException(nameof(vectorStore));this.ThreadDbKey=AIChatsId;}publicoverrideasyncTaskAddMessagesAsync(IEnumerable<ChatMessage>messages,CancellationTokencancellationToken){await_chatMessageStore.AddMessagesAsync(messages.Select(x=>newChatHistoryItemDto(){Key=this.ThreadDbKey+x.MessageId,Timestamp=DateTimeOffset.UtcNow,ThreadId=this.ThreadDbKey,MessageId=x.MessageId,Role=x.Role.Value,SerializedMessage=JsonSerializer.Serialize(x),MessageText=x.Text}).ToList(),cancellationToken);// 设置前景色为红色// 保存原始颜色,以便之后恢复ConsoleColororiginalColor=Console.ForegroundColor;Console.ForegroundColor=ConsoleColor.Red;Console.WriteLine("聊天消息记录:",Color.Red);messages.Select(x=>x.Text).ToList().ForEach(t=>Console.WriteLine(t));// 设置前景色为红色Console.ForegroundColor=ConsoleColor.Red;Console.WriteLine("聊天消息记录添加完成",Color.Red);// 恢复原始颜色Console.ForegroundColor=originalColor;}publicoverrideasyncTask<IEnumerable<ChatMessage>>GetMessagesAsync(CancellationTokencancellationToken){vardata=await_chatMessageStore.GetMessagesAsync(this.ThreadDbKey,cancellationToken);varmessages=data.ConvertAll(x=>JsonSerializer.Deserialize<ChatMessage>(x.SerializedMessage!)!);messages.Reverse();ConsoleColororiginalColor=Console.ForegroundColor;Console.ForegroundColor=ConsoleColor.Red;Console.WriteLine("所有聊天消息记录开始:",Color.Red);messages.Select(x=>x.Text).ToList().ForEach(t=>Console.WriteLine(t));Console.WriteLine("所有聊天消息记录结束:",Color.Red);// 恢复原始颜色Console.ForegroundColor=originalColor;returnmessages;}publicoverrideJsonElementSerialize(JsonSerializerOptions?jsonSerializerOptions=null)=>// We have to serialize the thread id, so that on deserialization you can retrieve the messages using the same thread id.JsonSerializer.SerializeToElement(this.ThreadDbKey);}
  2. 实现IKevinAIChatMessageStore

    TaskAddMessagesAsync(List<ChatHistoryItemDto>chatHistoryItems,CancellationTokencancellationToken);Task<List<ChatHistoryItemDto>>GetMessagesAsync(stringthreadId,CancellationTokencancellationToken);
  3. 实现注入到AI中间件中
    1.定义AI服务:

    /// <summary>/// AI服务/// </summary>publicclassAIAgentService:IAIAgentService{publicAIAgentService(){}publicasyncTask<AIAgent>CreateOpenAIAgent(stringname,stringprompt,stringdescription,stringurl,stringmodel,stringkeySecret,IList<AITool>?tools=null,ChatResponseFormat?chatResponseFormat=null,Func<IChatClient,IChatClient>?clientFactory=null,ILoggerFactory?loggerFactory=null,IServiceProvider?services=null){OpenAIClientOptionsopenAIClientOptions=newOpenAIClientOptions();openAIClientOptions.Endpoint=newUri(url);varai=newOpenAIClient(newApiKeyCredential(keySecret),openAIClientOptions);if(chatResponseFormat!=default){ChatOptionschatOptions=new(){ResponseFormat=chatResponseFormat};returnai.GetChatClient(model).CreateAIAgent(newChatClientAgentOptions(){Name=name,Instructions=prompt,ChatOptions=chatOptions,Description=description});}returnai.GetChatClient(model).CreateAIAgent(instructions:prompt,name:name,prompt,tools,clientFactory,loggerFactory,services);}publicasyncTask<AIAgent>CreateOpenAIAgent(stringmsg,stringurl,stringmodel,stringkeySecret,ChatClientAgentOptionschatClientAgentOptions){OpenAIClientOptionsopenAIClientOptions=newOpenAIClientOptions();openAIClientOptions.Endpoint=newUri(url);varai=newOpenAIClient(newApiKeyCredential(keySecret),openAIClientOptions);returnai.GetChatClient(model).CreateAIAgent(chatClientAgentOptions);}/// <summary>/// 智能体转换为McpServerTool/// </summary>/// <param name="aIAgent">智能体</param>/// <returns></returns>/// <exception cref="NotImplementedException"></exception>publicMcpServerToolAIAgentAsMcpServerTool(AIAgentaIAgent){returnMcpServerTool.Create(aIAgent.AsAIFunction());}/// <summary>/// 获取代理/// </summary>/// <returns></returns>publicIChatClientGetChatClient(stringurl,stringmodel,stringkeySecret){OpenAIClientOptionsopenAIClientOptions=newOpenAIClientOptions();openAIClientOptions.Endpoint=newUri(url);varai=newOpenAIClient(newApiKeyCredential(model),openAIClientOptions);returnai.GetChatClient(keySecret).AsIChatClient();}publicasyncTask<(AIAgent,AgentRunResponse)>CreateOpenAIAgentAndSendMSG(stringmsg,stringname,stringprompt,stringdescription,stringurl,stringmodel,stringkeySecret,IList<AITool>?tools=null,ChatResponseFormat?chatResponseFormat=null,Func<IChatClient,IChatClient>?clientFactory=null,ILoggerFactory?loggerFactory=null,IServiceProvider?services=null){OpenAIClientOptionsopenAIClientOptions=newOpenAIClientOptions();openAIClientOptions.Endpoint=newUri(url);varai=newOpenAIClient(newApiKeyCredential(keySecret),openAIClientOptions);varaiAgent=ai.GetChatClient(model).CreateAIAgent(instructions:prompt,name:name,prompt,tools,clientFactory,loggerFactory,services);if(chatResponseFormat!=default){ChatOptionschatOptions=new(){ResponseFormat=chatResponseFormat};aiAgent=ai.GetChatClient(model).CreateAIAgent(newChatClientAgentOptions(){Name=name,Instructions=prompt,ChatOptions=chatOptions,Description=description});}varreslut=awaitaiAgent.RunAsync(msg);return(aiAgent,reslut);}publicasyncTask<(AIAgent,AgentRunResponse)>CreateOpenAIAgentAndSendMSG(stringmsg,stringurl,stringmodel,stringkeySecret,ChatClientAgentOptionschatClientAgentOptions){OpenAIClientOptionsopenAIClientOptions=newOpenAIClientOptions();openAIClientOptions.Endpoint=newUri(url);varai=newOpenAIClient(newApiKeyCredential(keySecret),openAIClientOptions);varaiAgent=ai.GetChatClient(model).CreateAIAgent(chatClientAgentOptions);varreslut=awaitaiAgent.RunAsync(msg);return(aiAgent,reslut);}}

2.使用AI服务

addAi.Content=(awaitaIAgentService.CreateOpenAIAgentAndSendMSG(add.Content,aIModels.EndPoint,aIModels.ModelName,aIModels.ModelKey,newChatClientAgentOptions{Name=aiapp.Name,Instructions=aIPrompts.Prompt,Description=aIPrompts.Description??"你是一个智能体,请根据你的提示词进行相关回答",ChatMessageStoreFactory=ctx=>{// Create a new chat message store for this agent that stores the messages in a vector store.returnnewKevinChatMessageStore(kevinAIChatMessageStore,ctx.SerializedState,par.AIChatsId.ToString(),ctx.JsonSerializerOptions);}})).Item2.Text;
数据库设计建议

对于关系型数据库(如 SQL Server),建议的表结构:

/// <summary>/// 专门用于存储AI聊天记录的表/// </summary>[Table("TAIChatMessageStore")][Description("专门用于存储AI聊天记录的表")][Index(nameof(Key))][Index(nameof(ThreadId))][Index(nameof(Role))][Index(nameof(MessageId))]publicclass TAIChatMessageStore : CUD_User {[MaxLength(200)]publicstringKey{ get;set;}[MaxLength(100)]publicstring ThreadId { get;set;}[Description("消息时间stamp")]publicDateTimeOffset?Timestamp{ get;set;}/// <summary>/// 角色标识/// </summary>[MaxLength(50)]publicstring Role { get;set;}publicstring SerializedMessage { get;set;}/// <summary>/// 消息内容/// </summary>publicstring? MessageText { get;set;}/// <summary>/// 消息id/// </summary>[Description("消息id")][MaxLength(100)]publicstring? MessageId { get;set;} }
实现数据持久化
  1. 写入示例
    使用 Entity Framework Core 保存数据:

    publicasyncTaskAddMessagesAsync(List<ChatHistoryItemDto>chatHistoryItems,CancellationTokencancellationToken){varadddata=chatHistoryItems.Select(t=>newTAIChatMessageStore{Id=SnowflakeIdService.GetNextId(),CreateTime=DateTime.Now,CreateUserId=CurrentUser.UserId,IsDelete=false,TenantId=CurrentUser.TenantId,ThreadId=t.ThreadId,Timestamp=t.Timestamp,Role=t.Role,Key=t.Key,SerializedMessage=t.SerializedMessage,MessageText=t.MessageText,MessageId=t.MessageId}).ToList();aIChatMessageStoreRp.AddRange(adddata);awaitaIChatMessageStoreRp.SaveChangesAsync();}
  2. 读取示例

    publicTask<List<ChatHistoryItemDto>>GetMessagesAsync(stringthreadId,CancellationTokencancellationToken){returnaIChatMessageStoreRp.Query().Where(t=>t.ThreadId==threadId&&t.IsDelete==false).Select(t=>newChatHistoryItemDto{Key=t.Key,ThreadId=t.ThreadId,Timestamp=t.Timestamp,SerializedMessage=t.SerializedMessage,MessageText=t.MessageText,Role=t.Role,MessageId=t.MessageId}).ToListAsync();}
性能优化建议
  • 为高频查询字段(如UserIdChannelId)添加索引
  • 实现数据分片策略应对大规模历史记录
  • 考虑使用 Redis 缓存热点对话数据
安全注意事项
  • 加密存储敏感对话内容
  • 实现数据保留策略定期清理旧记录
  • 遵守 GDPR 等数据隐私法规
测试验证方法
  1. 编写单元测试验证存储接口实现
  2. 使用 Bot Framework Emulator 进行端到端测试
  3. 进行负载测试验证性能表现
扩展可能性
  1. 添加全文检索支持(如 Azure Cognitive Search)
  2. 实现跨渠道对话历史同步
  3. 开发分析模块生成对话洞察报告

这种实现方式允许完全控制数据存储位置和格式,同时保持与 Bot Framework 的兼容性。根据具体需求可选择 SQL Server、Cosmos DB 或其他数据库解决方案。

版权声明: 本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若内容造成侵权/违法违规/事实不符,请联系邮箱:809451989@qq.com进行投诉反馈,一经查实,立即删除!
网站建设 2026/5/4 20:06:17

【稀缺资料】资深架构师亲授:高并发下多模态Agent的Docker存储优化策略

第一章&#xff1a;多模态 Agent 的 Docker 存储配置在构建支持图像、文本、语音等多模态数据处理的 Agent 系统时&#xff0c;Docker 容器化部署成为提升环境一致性与可扩展性的关键手段。其中&#xff0c;持久化存储的合理配置直接影响数据读取效率与模型训练稳定性。通过挂载…

作者头像 李华
网站建设 2026/5/4 23:49:23

27、Vim自动缩进与关键字补全功能全解析

Vim自动缩进与关键字补全功能全解析 1. Vim自动缩进方法概述 Vim 提供了四种功能逐渐增强且复杂的自动缩进文本的方法。你可以通过 :set 命令轻松选择缩进方法,例如 :set cindent 。以下是这四种方法,按复杂度递增排序: - autoindent :此方法与 vi 的 autoinde…

作者头像 李华
网站建设 2026/4/28 14:41:15

Puerts性能调优实战:从引擎配置到代码优化的完整指南

Puerts性能调优实战&#xff1a;从引擎配置到代码优化的完整指南 【免费下载链接】puerts PUER(普洱) Typescript. Lets write your game in UE or Unity with TypeScript. 项目地址: https://gitcode.com/GitHub_Trending/pu/puerts 当你的Unity或Unreal项目使用Puerts…

作者头像 李华
网站建设 2026/5/3 17:16:01

Jellyfin Bangumi插件终极指南:轻松管理中文番剧库

还在为Jellyfin中的动画资源缺少中文元数据而困扰吗&#xff1f;Jellyfin Bangumi插件为您提供完美解决方案&#xff01;这款免费开源插件专门针对中文用户设计&#xff0c;能够自动从Bangumi番组计划获取高质量的中文番剧元数据&#xff0c;并实现播放进度与Bangumi账号的实时…

作者头像 李华
网站建设 2026/5/1 2:44:33

云原生Agent资源调度避坑手册,99%工程师都会犯的3个致命错误

第一章&#xff1a;云原生Agent资源调度的现状与挑战随着云原生技术的广泛应用&#xff0c;越来越多的分布式系统开始引入智能Agent来实现自动化运维、弹性扩缩容和故障自愈。这些Agent通常以容器化形式部署在Kubernetes等编排平台之上&#xff0c;依赖底层调度器完成资源分配与…

作者头像 李华