[{"data":1,"prerenderedAt":1866},["ShallowReactive",2],{"blog-\u002Fblog\u002Fai-agent\u002Ffunction-calling":3,"blog-related-\u002Fblog\u002Fai-agent\u002Ffunction-calling":397},{"id":4,"title":5,"author":6,"body":7,"category":366,"cover":367,"date":368,"description":369,"draft":370,"extension":371,"faq":372,"featured":370,"image":367,"keywords":382,"meta":387,"navigation":388,"path":389,"seo":390,"sitemap":391,"stem":392,"tags":393,"updated":368,"__hash__":396},"blog\u002Fblog\u002Fai-agent\u002Ffunction-calling.md","Function Calling怎么用来接业务系统","HNREIS",{"type":8,"value":9,"toc":344},"minimark",[10,19,24,27,38,41,44,49,62,66,77,81,89,93,101,105,109,117,121,129,133,141,145,153,157,165,169,225,230,233,265,268,317,320,338],[11,12,13,14,18],"p",{},"AI如果只会聊天，价值有限。",[15,16,17],"strong",{},"Function Calling让AI能调用业务系统的API——查订单、查库存、甚至下单，从\"会说\"变成\"能做事\"。但执行真实操作，安全是关键。"," 这篇讲清怎么用。",[20,21,23],"h2",{"id":22},"function-calling是什么","Function Calling是什么",[11,25,26],{},"让AI根据意图调用预定义函数：",[28,29,34],"pre",{"className":30,"code":32,"language":33},[31],"language-text","用户：\"查下订单123的状态\"\n→ AI识别意图 → 调用 getOrderStatus(orderId=123)\n→ 业务系统返回 → AI把结果告诉用户\n","text",[35,36,32],"code",{"__ignoreMap":37},"",[11,39,40],{},"AI成了业务系统的\"智能接口\"。",[20,42,43],{"id":43},"企业能做什么",[45,46,48],"h3",{"id":47},"_1-信息查询","1. 信息查询",[50,51,52,56,59],"ul",{},[53,54,55],"li",{},"查订单、查库存、查物流。",[53,57,58],{},"查客户信息、查产品。",[53,60,61],{},"替代繁琐的系统操作。",[45,63,65],{"id":64},"_2-操作执行","2. 操作执行",[50,67,68,71,74],{},[53,69,70],{},"下单、改地址、退款。",[53,72,73],{},"创建工单、分配任务。",[53,75,76],{},"风险高，需控权。",[45,78,80],{"id":79},"_3-多步骤流程","3. 多步骤流程",[50,82,83,86],{},[53,84,85],{},"用户一句话触发多步操作。",[53,87,88],{},"AI编排多个函数。",[45,90,92],{"id":91},"_4-数据分析","4. 数据分析",[50,94,95,98],{},[53,96,97],{},"调用查询函数获取数据。",[53,99,100],{},"AI分析解读。",[20,102,104],{"id":103},"安全要点关键","安全要点（关键）",[45,106,108],{"id":107},"_1-函数与参数校验","1. 函数与参数校验",[50,110,111,114],{},[53,112,113],{},"严格校验函数名和参数。",[53,115,116],{},"防止AI传错或被构造恶意参数。",[45,118,120],{"id":119},"_2-权限控制","2. 权限控制",[50,122,123,126],{},[53,124,125],{},"按角色限制能调用的函数。",[53,127,128],{},"用户只能调自己权限内的。",[45,130,132],{"id":131},"_3-敏感操作人工确认","3. 敏感操作人工确认",[50,134,135,138],{},[53,136,137],{},"写操作（下单、支付、删除）人工确认。",[53,139,140],{},"不能AI自动执行高风险操作。",[45,142,144],{"id":143},"_4-操作审计","4. 操作审计",[50,146,147,150],{},[53,148,149],{},"所有调用记录可追溯。",[53,151,152],{},"异常监控。",[45,154,156],{"id":155},"_5-限制范围","5. 限制范围",[50,158,159,162],{},[53,160,161],{},"只暴露必要的函数。",[53,163,164],{},"危险函数不暴露。",[20,166,168],{"id":167},"查询-vs-写操作","查询 vs 写操作",[170,171,172,188],"table",{},[173,174,175],"thead",{},[176,177,178,182,185],"tr",{},[179,180,181],"th",{},"类型",[179,183,184],{},"风险",[179,186,187],{},"策略",[189,190,191,203,214],"tbody",{},[176,192,193,197,200],{},[194,195,196],"td",{},"查询类（只读）",[194,198,199],{},"低",[194,201,202],{},"可自动",[176,204,205,208,211],{},[194,206,207],{},"写操作（下单\u002F修改）",[194,209,210],{},"高",[194,212,213],{},"人工确认",[176,215,216,219,222],{},[194,217,218],{},"敏感操作（支付\u002F删除）",[194,220,221],{},"极高",[194,223,224],{},"严格确认+审计",[11,226,227],{},[15,228,229],{},"查询自动，写操作人工确认。",[20,231,232],{"id":232},"别踩的坑",[50,234,235,241,247,253,259],{},[53,236,237,240],{},[15,238,239],{},"无约束调用业务系统","：误操作和数据风险。",[53,242,243,246],{},[15,244,245],{},"写操作不确认","：AI误下单造成损失。",[53,248,249,252],{},[15,250,251],{},"不校验参数","：错误或恶意参数。",[53,254,255,258],{},[15,256,257],{},"权限不管控","：越权调用。",[53,260,261,264],{},[15,262,263],{},"不审计","：出问题无法追溯。",[20,266,267],{"id":267},"成本参考",[170,269,270,283],{},[173,271,272],{},[176,273,274,277,280],{},[179,275,276],{},"方案",[179,278,279],{},"说明",[179,281,282],{},"成本量级",[189,284,285,296,306],{},[176,286,287,290,293],{},[194,288,289],{},"单场景函数调用",[194,291,292],{},"接几个业务API",[194,294,295],{},"中",[176,297,298,301,304],{},[194,299,300],{},"多函数+流程编排",[194,302,303],{},"多API+权限+审计",[194,305,295],{},[176,307,308,311,314],{},[194,309,310],{},"企业级AI业务集成",[194,312,313],{},"全场景+安全+审计+集成",[194,315,316],{},"中高，定制",[20,318,319],{"id":319},"怎么开始",[321,322,323,326,329,332,335],"ol",{},[53,324,325],{},"选高频查询场景切入。",[53,327,328],{},"定义清晰的函数和参数。",[53,330,331],{},"严格校验和权限控制。",[53,333,334],{},"写操作人工确认。",[53,336,337],{},"全程审计。",[339,340,341],"blockquote",{},[11,342,343],{},"广州市汉诺雷斯（HNREIS）帮企业用Function Calling把AI接入业务系统，从查询到操作，带权限控制、人工确认和审计。把你的业务集成需求告诉我们，我们给出安全的方案。",{"title":37,"searchDepth":345,"depth":345,"links":346},2,[347,348,355,362,363,364,365],{"id":22,"depth":345,"text":23},{"id":43,"depth":345,"text":43,"children":349},[350,352,353,354],{"id":47,"depth":351,"text":48},3,{"id":64,"depth":351,"text":65},{"id":79,"depth":351,"text":80},{"id":91,"depth":351,"text":92},{"id":103,"depth":345,"text":104,"children":356},[357,358,359,360,361],{"id":107,"depth":351,"text":108},{"id":119,"depth":351,"text":120},{"id":131,"depth":351,"text":132},{"id":143,"depth":351,"text":144},{"id":155,"depth":351,"text":156},{"id":167,"depth":345,"text":168},{"id":232,"depth":345,"text":232},{"id":267,"depth":345,"text":267},{"id":319,"depth":345,"text":319},"ai-agent",null,"2025-12-24","Function Calling让AI能调用业务系统的API，实现查询下单等真实操作。本文讲清Function Calling的原理、企业用法和安全要点。",false,"md",[373,376,379],{"q":374,"a":375},"Function Calling是什么？","Function Calling（函数调用）是让AI根据用户意图，调用预定义的函数（对应业务系统的API），获取结果或执行操作。比如用户问\"查下我订单状态\"，AI识别意图调用订单查询函数返回结果。让AI从\"只会说\"变成\"能做事\"，是AI接业务系统的关键技术。",{"q":377,"a":378},"Function Calling能执行下单等操作吗？","能，但要严格控权。涉及查询类的（查订单、查库存）风险低，可自动执行。涉及写操作（下单、修改、支付）风险高，要严格权限控制、人工确认、操作审计。建议查询类自动，写操作和敏感操作人工确认后再执行，避免AI误操作造成损失。",{"q":380,"a":381},"Function Calling安全吗？","有风险要点。AI可能误判意图调错函数、参数错误、被诱导调用危险操作。安全要点：函数和参数严格校验、权限控制（按角色限制能调的函数）、敏感操作人工确认、操作审计可追溯、限制能调用的函数范围。不能让AI无约束地调用业务系统。",[383,384,385,386],"Function Calling","函数调用","AI工具集成","AI接业务系统",{},true,"\u002Fblog\u002Fai-agent\u002Ffunction-calling",{"title":5,"description":369},{"loc":389},"blog\u002Fai-agent\u002Ffunction-calling",[383,394,395],"集成","AI","tAytMQt_4lvH49YoEDCbVHM9O7mAhGzKrm2ZDjVA6p8",[398,786,1136,1504],{"id":399,"title":400,"author":6,"body":401,"category":366,"cover":367,"date":761,"description":762,"draft":370,"extension":371,"faq":763,"featured":370,"image":367,"keywords":773,"meta":777,"navigation":388,"path":778,"seo":779,"sitemap":780,"stem":781,"tags":782,"updated":761,"__hash__":785},"blog\u002Fblog\u002Fai-agent\u002Fagent-kuangjia.md","主流Agent框架怎么选",{"type":8,"value":402,"toc":748},[403,410,414,417,437,444,447,583,588,591,595,607,611,618,622,625,629,632,634,666,668,724,726,743],[11,404,405,406,409],{},"企业要落地一个AI智能体，市面框架多得让人发懵：LangChain、LlamaIndex、LangGraph、AutoGen、CrewAI、Dify、Coze……选错了要么过度复杂，要么撑不住业务。",[15,407,408],{},"框架不是越新越热门越好，而是要匹配你的场景和团队能力。"," 这篇讲清怎么选。",[20,411,413],{"id":412},"先想清楚你到底需不需要框架","先想清楚：你到底需不需要框架",[11,415,416],{},"很多人一上来就要\"上框架\"，其实很多场景不需要：",[50,418,419,425,431],{},[53,420,421,424],{},[15,422,423],{},"简单问答","：直接调大模型 API，几十行代码搞定，不用框架。",[53,426,427,430],{},[15,428,429],{},"RAG 知识库问答","：用轻量库或框架的检索能力，中等复杂度。",[53,432,433,436],{},[15,434,435],{},"复杂智能体","：多步推理、调用多个工具、多轮对话有记忆、多个智能体协作——这才是框架真正发挥作用的地方。",[11,438,439,440,443],{},"需求复杂度决定框架价值。",[15,441,442],{},"简单需求硬上重框架，是典型的过度设计","。",[20,445,446],{"id":446},"主流框架对比",[170,448,449,467],{},[173,450,451],{},[176,452,453,456,458,461,464],{},[179,454,455],{},"框架",[179,457,181],{},[179,459,460],{},"核心定位",[179,462,463],{},"适合谁",[179,465,466],{},"主要局限",[189,468,469,486,502,518,534,550,567],{},[176,470,471,474,477,480,483],{},[194,472,473],{},"LangChain",[194,475,476],{},"代码框架",[194,478,479],{},"通用 LLM 应用编排，生态最全",[194,481,482],{},"有研发团队、要深度定制",[194,484,485],{},"抽象较重，早期 API 频繁变动",[176,487,488,491,493,496,499],{},[194,489,490],{},"LangGraph",[194,492,476],{},[194,494,495],{},"基于图的状态机，做可控多步 Agent",[194,497,498],{},"复杂流程、需要稳定状态控制",[194,500,501],{},"学习曲线陡",[176,503,504,507,509,512,515],{},[194,505,506],{},"LlamaIndex",[194,508,476],{},[194,510,511],{},"数据连接与检索（RAG 见长）",[194,513,514],{},"以知识库\u002F文档检索为核心",[194,516,517],{},"Agent 编排不如 LangChain 全",[176,519,520,523,525,528,531],{},[194,521,522],{},"AutoGen",[194,524,476],{},[194,526,527],{},"多智能体对话协作",[194,529,530],{},"多 Agent 角色分工场景",[194,532,533],{},"偏研究，工程化要自己补",[176,535,536,539,541,544,547],{},[194,537,538],{},"CrewAI",[194,540,476],{},[194,542,543],{},"角色化多 Agent 协作，上手快",[194,545,546],{},"快速搭多角色协作原型",[194,548,549],{},"复杂控制力有限",[176,551,552,555,558,561,564],{},[194,553,554],{},"Dify",[194,556,557],{},"低代码平台",[194,559,560],{},"可视化编排 + 知识库 + 应用托管",[194,562,563],{},"快速落地、业务人员参与",[194,565,566],{},"深度定制受平台能力限制",[176,568,569,572,574,577,580],{},[194,570,571],{},"Coze",[194,573,557],{},[194,575,576],{},"字节出品，插件生态丰富",[194,578,579],{},"快速搭建对话型应用",[194,581,582],{},"偏消费侧，企业私有化受限",[11,584,585],{},[15,586,587],{},"没有\"最好\"的框架，只有\"最匹配\"的框架。",[20,589,590],{"id":590},"选型要看的维度",[45,592,594],{"id":593},"_1-编程式还是低代码","1. 编程式还是低代码",[50,596,597,602],{},[53,598,599,601],{},[15,600,476],{},"（LangChain\u002FLlamaIndex 等）：灵活、可控、能深度集成业务系统，但要有研发投入。",[53,603,604,606],{},[15,605,557],{},"（Dify\u002FCoze）：拖拽配置、上手快、适合原型和非纯技术团队，但定制边界明显。",[45,608,610],{"id":609},"_2-单-agent-还是多-agent","2. 单 Agent 还是多 Agent",[11,612,613,614,617],{},"大部分企业场景，",[15,615,616],{},"单个能调用工具的 Agent 就够用","。多智能体协作（AutoGen\u002FCrewAI）适合确实需要多角色分工的复杂流程，比如\"研究员+执行者+审核者\"。别为了炫技上多 Agent。",[45,619,621],{"id":620},"_3-能不能私有化部署","3. 能不能私有化部署",[11,623,624],{},"涉及企业内部数据，通常要求私有化。Dify 支持私有部署，Coze 偏公有云。代码框架本身可私有化，但要自己搭运维。",[45,626,628],{"id":627},"_4-社区活跃度与稳定性","4. 社区活跃度与稳定性",[11,630,631],{},"框架迭代快是双刃剑。优先选社区活跃、有企业背书、更新稳定的，避免踩\"作者弃坑\"的雷。",[20,633,232],{"id":232},[50,635,636,642,648,654,660],{},[53,637,638,641],{},[15,639,640],{},"过度设计","：简单需求上重框架，开发和维护成本翻倍。",[53,643,644,647],{},[15,645,646],{},"被框架锁死","：把提示词、工具、业务逻辑和框架深度耦合，迁移不动。",[53,649,650,653],{},[15,651,652],{},"忽视核心","：框架只是壳，真正决定效果的是提示词、数据质量和知识库。",[53,655,656,659],{},[15,657,658],{},"盲目追新","：新框架未必成熟，企业项目要稳。",[53,661,662,665],{},[15,663,664],{},"忽视可观测","：Agent 行为不确定性高，没有日志和追踪很难调试。",[20,667,267],{"id":267},[170,669,670,680],{},[173,671,672],{},[176,673,674,676,678],{},[179,675,276],{},[179,677,279],{},[179,679,282],{},[189,681,682,692,702,713],{},[176,683,684,687,690],{},[194,685,686],{},"自研（直接调 API）",[194,688,689],{},"简单场景，研发 1-2 周",[194,691,199],{},[176,693,694,697,700],{},[194,695,696],{},"基于代码框架开发",[194,698,699],{},"中等复杂度，研发 3-8 周",[194,701,295],{},[176,703,704,707,710],{},[194,705,706],{},"低代码平台搭建",[194,708,709],{},"快速原型到中等应用",[194,711,712],{},"平台费 + 少量定制",[176,714,715,718,721],{},[194,716,717],{},"复杂多 Agent 系统",[194,719,720],{},"多角色协作 + 工具集成 + 私有化",[194,722,723],{},"较高，定制开发",[20,725,319],{"id":319},[321,727,728,731,734,737,740],{},[53,729,730],{},"明确场景和复杂度（是否真需要框架）。",[53,732,733],{},"评估团队能力（能否驾驭代码框架）。",[53,735,736],{},"对比 2-3 个候选框架，做小范围验证。",[53,738,739],{},"关注私有化、可观测、可迁移。",[53,741,742],{},"把核心资产与框架解耦。",[339,744,745],{},[11,746,747],{},"广州市汉诺雷斯（HNREIS）帮企业做AI智能体落地，从场景评估、框架选型到私有化部署和工具集成。把你的业务场景告诉我们，我们给出务实的选型建议与方案。",{"title":37,"searchDepth":345,"depth":345,"links":749},[750,751,752,758,759,760],{"id":412,"depth":345,"text":413},{"id":446,"depth":345,"text":446},{"id":590,"depth":345,"text":590,"children":753},[754,755,756,757],{"id":593,"depth":351,"text":594},{"id":609,"depth":351,"text":610},{"id":620,"depth":351,"text":621},{"id":627,"depth":351,"text":628},{"id":232,"depth":345,"text":232},{"id":267,"depth":345,"text":267},{"id":319,"depth":345,"text":319},"2024-05-12","LangChain、LlamaIndex、LangGraph、AutoGen、CrewAI、Dify、Coze 等Agent框架各有侧重。本文从企业落地视角对比主流框架，讲清选型维度，帮你按场景选对工具而不踩坑。",[764,767,770],{"q":765,"a":766},"企业做AI智能体一定要用框架吗？","不一定。简单的单轮问答或单次工具调用，直接调大模型API就能实现。框架的价值在复杂场景：多步推理、工具编排、记忆管理、多智能体协作、可观测的流程控制。需求越复杂，框架越能省去重复造轮子。但框架也有学习成本和被锁定的风险，按需选择，不要为了用框架而用框架。",{"q":768,"a":769},"LangChain和Dify怎么选？","看团队和技术诉求。LangChain是代码框架，灵活、可控、可深度定制，适合有研发团队、要和业务系统深度集成的企业；Dify是低代码平台，拖拽编排、可视化，适合快速出原型、让业务人员参与配置。要做底层定制选代码框架，要快速落地和降低门槛选低代码平台。",{"q":771,"a":772},"用框架会被技术锁定吗？","有一定风险。深度依赖某框架的特有抽象，后续迁移成本会高。建议把核心资产——提示词、工具定义、业务数据、知识库——与框架解耦，框架只承担编排层。同时优先选开源、社区活跃、更新稳定的框架，降低被单一厂商锁定或项目停更的风险。",[774,473,554,775,776],"Agent框架","AI智能体开发","框架选型",{},"\u002Fblog\u002Fai-agent\u002Fagent-kuangjia",{"title":400,"description":762},{"loc":778},"blog\u002Fai-agent\u002Fagent-kuangjia",[783,776,784],"Agent","大模型","yYDRGbvFeaR227ajWv5qiaN844ebZL6NwOSs4iOxEcE",{"id":787,"title":788,"author":6,"body":789,"category":366,"cover":367,"date":1109,"description":1110,"draft":370,"extension":371,"faq":1111,"featured":370,"image":367,"keywords":1121,"meta":1127,"navigation":388,"path":1128,"seo":1129,"sitemap":1130,"stem":1131,"tags":1132,"updated":1109,"__hash__":1135},"blog\u002Fblog\u002Fai-agent\u002Fai-agent-vs-chatbot.md","AI Agent 和传统聊天机器人有什么区别？别再被忽悠",{"type":8,"value":790,"toc":1100},[791,798,801,815,821,824,927,930,935,948,951,954,957,971,974,978,981,1013,1016,1062,1066,1069,1095],[11,792,793,794,797],{},"很多企业几年前上过\"智能客服\"，结果发现它\"一点都不智能\"——答非所问、死板的流程树、动不动就\"转人工\"。于是对\"AI 客服\"有阴影。",[15,795,796],{},"但现在的 AI Agent 和当年的聊天机器人，是完全不同的两种东西。"," 这篇文章讲清它们的本质区别，帮你判断企业该上哪种、值不值得换。",[20,799,800],{"id":800},"一句话区分",[50,802,803,809],{},[53,804,805,808],{},[15,806,807],{},"传统聊天机器人","：基于关键词匹配和预设流程图。用户说\"退款\"，它跳到\"退款流程\"；用户换个说法\"我不要了\"，它就懵了。",[53,810,811,814],{},[15,812,813],{},"AI Agent","：基于大语言模型，能理解意图、规划任务、调用工具、记住上下文。用户说\"我上周买的那个订单想退\"，它能自己查订单、判断是否符合退款规则、发起退款流程。",[11,816,817,818],{},"本质区别：",[15,819,820],{},"前者是\"执行固定脚本的程序\"，后者是\"能理解并完成任务的助手\"。",[20,822,823],{"id":823},"六维对比表",[170,825,826,837],{},[173,827,828],{},[176,829,830,833,835],{},[179,831,832],{},"能力",[179,834,807],{},[179,836,813],{},[189,838,839,850,861,872,883,894,905,916],{},[176,840,841,844,847],{},[194,842,843],{},"对话方式",[194,845,846],{},"关键词 + 流程图",[194,848,849],{},"大模型理解自然语言",[176,851,852,855,858],{},[194,853,854],{},"理解能力",[194,856,857],{},"死板，换个说法就懵",[194,859,860],{},"灵活，懂口语、省略、上下文",[176,862,863,866,869],{},[194,864,865],{},"自主性",[194,867,868],{},"被动应答",[194,870,871],{},"主动规划、多步执行",[176,873,874,877,880],{},[194,875,876],{},"工具调用",[194,878,879],{},"几乎不能",[194,881,882],{},"能查库、调 API、发通知、生成文件",[176,884,885,888,891],{},[194,886,887],{},"记忆",[194,889,890],{},"无或当次会话",[194,892,893],{},"长期记忆（记住用户历史）",[176,895,896,899,902],{},[194,897,898],{},"复杂任务",[194,900,901],{},"只能引导转人工",[194,903,904],{},"可独立完成跨系统任务",[176,906,907,910,913],{},[194,908,909],{},"知识更新",[194,911,912],{},"改流程图\u002F规则",[194,914,915],{},"喂新文档即可",[176,917,918,921,924],{},[194,919,920],{},"用户体验",[194,922,923],{},"机械、易激怒用户",[194,925,926],{},"接近真人，体验好",[20,928,929],{"id":929},"真实场景对比",[11,931,932],{},[15,933,934],{},"场景：用户问\"我上周买的那个能退吗\"",[50,936,937,943],{},[53,938,939,942],{},[15,940,941],{},"传统机器人","：抓不到关键词，回\"请问您要咨询什么\"，或跳到通用退款说明页。用户烦躁，转人工。",[53,944,945,947],{},[15,946,813],{},"：理解意图（查订单+判断退款）→ 调用订单系统查到上周的订单 → 判断商品是否符合退款规则 → 回答\"您的订单 xxx 符合 7 天无理由退款，我帮您发起申请，确认吗？\" → 用户确认 → 调用退款 API → 完成。",[11,949,950],{},"整个流程 AI Agent 自己做完，用户感觉\"这个客服真懂\"。",[20,952,953],{"id":953},"为什么传统机器人体验差",[11,955,956],{},"传统机器人的工作机制是\"if 用户说 X，then 回 Y\"。问题在于：",[50,958,959,962,965,968],{},[53,960,961],{},"用户的说法千变万化，关键词覆盖不全。",[53,963,964],{},"业务一复杂，流程图爆炸，维护噩梦。",[53,966,967],{},"没法理解上下文，每句都从头开始。",[53,969,970],{},"没法真正\"做事\"，只能\"指路\"。",[11,972,973],{},"所以传统机器人最后基本都退化成\"转人工的入口\"，用户也学会了\"上来就转人工\"。",[20,975,977],{"id":976},"ai-agent-为什么能真正解决问题","AI Agent 为什么能真正解决问题",[11,979,980],{},"AI Agent 的核心能力：",[50,982,983,989,995,1001,1007],{},[53,984,985,988],{},[15,986,987],{},"理解自然语言","：基于大模型，用户怎么说都能懂。",[53,990,991,994],{},[15,992,993],{},"调用工具","：接业务系统（CRM、订单、知识库），能真正查数据、改数据。",[53,996,997,1000],{},[15,998,999],{},"规划任务","：把复杂目标拆成步骤，自己执行。",[53,1002,1003,1006],{},[15,1004,1005],{},"长期记忆","：记住用户历史，体验连贯。",[53,1008,1009,1012],{},[15,1010,1011],{},"知识可更新","：喂新文档即学习，不用改代码。",[20,1014,1015],{"id":1015},"企业该上哪种",[170,1017,1018,1028],{},[173,1019,1020],{},[176,1021,1022,1025],{},[179,1023,1024],{},"场景",[179,1026,1027],{},"建议",[189,1029,1030,1038,1046,1054],{},[176,1031,1032,1035],{},[194,1033,1034],{},"客服量大、问题多样",[194,1036,1037],{},"AI Agent（体验好、省人工）",[176,1039,1040,1043],{},[194,1041,1042],{},"问题高度标准化、简单",[194,1044,1045],{},"传统机器人够用，或 AI Agent 低配版",[176,1047,1048,1051],{},[194,1049,1050],{},"需要跨系统办事（查单、改单、退款）",[194,1052,1053],{},"AI Agent（能调工具）",[176,1055,1056,1059],{},[194,1057,1058],{},"预算极有限、想先试水",[194,1060,1061],{},"AI Agent MVP（先接知识库做问答）",[20,1063,1065],{"id":1064},"怎么从传统机器人升级到-ai-agent","怎么从传统机器人升级到 AI Agent",[11,1067,1068],{},"不用一次性推翻：",[321,1070,1071,1077,1083,1089],{},[53,1072,1073,1076],{},[15,1074,1075],{},"先接知识库","：把产品手册、FAQ、SOP 喂给 AI，让它能准确回答咨询类问题。",[53,1078,1079,1082],{},[15,1080,1081],{},"再接业务系统","：让 AI 能查订单、查库存、发通知，真正\"办事\"。",[53,1084,1085,1088],{},[15,1086,1087],{},"加人工兜底","：AI 把握不大时转人工，持续优化。",[53,1090,1091,1094],{},[15,1092,1093],{},"逐步替换","：AI 能力稳定后，逐步下线老机器人的场景。",[339,1096,1097],{},[11,1098,1099],{},"广州市汉诺雷斯（HNREIS）提供 AI 智能客服定制，支持从传统机器人渐进式升级。告诉我们你现在的客服痛点和量级，我们帮你评估 AI Agent 能解决多少、ROI 多少。",{"title":37,"searchDepth":345,"depth":345,"links":1101},[1102,1103,1104,1105,1106,1107,1108],{"id":800,"depth":345,"text":800},{"id":823,"depth":345,"text":823},{"id":929,"depth":345,"text":929},{"id":953,"depth":345,"text":953},{"id":976,"depth":345,"text":977},{"id":1015,"depth":345,"text":1015},{"id":1064,"depth":345,"text":1065},"2024-05-26","传统聊天机器人基于关键词和流程图，只能被动应答；AI Agent 基于大模型，能理解意图、规划任务、调用工具、长期记忆。本文用对比表和真实场景讲清两者的本质区别，帮你判断企业该上哪种。",[1112,1115,1118],{"q":1113,"a":1114},"AI Agent 比传统聊天机器人贵很多吗？","一次性投入是的，但要看性价比。传统聊天机器人便宜（几千到一两万），但只能回答固定问题，用户体验差、转人工率高；AI Agent 投入更高（几万起），但能理解自然语言、调用业务系统、真正解决问题，长期看能省更多人工成本。如果客服量大，AI Agent 的回报更快。",{"q":1116,"a":1117},"我们已经有客服机器人了，要全部换掉吗？","不必全部换。可以分步：先用 AI Agent 接管高频、标准的问题（产品咨询、订单查询、售后流程），复杂问题继续走老系统或人工。我们支持渐进式升级，避免一次性推翻重来。",{"q":1119,"a":1120},"AI Agent 会不会答错或者说胡话？","会，但有办法控制。关键是 RAG（检索增强生成）——让 AI 先在企业知识库里查到准确信息再回答，而不是凭空生成。再配上人工兜底机制（AI 把握不大时转人工）和回答质量回流，准确率可以做到很高。",[1122,1123,1124,1125,1126],"AI Agent和聊天机器人区别","智能客服","AI客服机器人","传统客服机器人","AI智能体",{},"\u002Fblog\u002Fai-agent\u002Fai-agent-vs-chatbot",{"title":788,"description":1110},{"loc":1128},"blog\u002Fai-agent\u002Fai-agent-vs-chatbot",[1133,395,1134],"对比","概念","8qKIdcO08ytoPMy5eHwEUZP4TpzwR1mih2PoU1dQnm0",{"id":1137,"title":1138,"author":6,"body":1139,"category":366,"cover":367,"date":1477,"description":1478,"draft":370,"extension":371,"faq":1479,"featured":370,"image":367,"keywords":1489,"meta":1494,"navigation":388,"path":1495,"seo":1496,"sitemap":1497,"stem":1498,"tags":1499,"updated":1477,"__hash__":1503},"blog\u002Fblog\u002Fai-agent\u002Fai-caiwu-baobiao.md","AI辅助财务报表和数据分析怎么做",{"type":8,"value":1140,"toc":1458},[1141,1148,1152,1156,1167,1171,1182,1186,1197,1201,1209,1213,1239,1244,1248,1251,1297,1302,1305,1309,1320,1324,1332,1336,1344,1348,1356,1358,1405,1408,1434,1436,1453],[11,1142,1143,1144,1147],{},"财务部门每天和报表、数据打交道，重复分析多、对数据敏感度要求极高。",[15,1145,1146],{},"AI能帮财务提效，但财务数据的特殊性决定了\"怎么用\"比\"用不用\"更重要。"," 这篇讲清AI辅助财务分析的边界、做法和安全要点。",[20,1149,1151],{"id":1150},"ai在财务分析里能做什么","AI在财务分析里能做什么",[45,1153,1155],{"id":1154},"_1-报表解读与可视化分析","1. 报表解读与可视化分析",[50,1157,1158,1161,1164],{},[53,1159,1160],{},"把利润表、资产负债表、现金流量表的关键指标提炼成通俗解读。",[53,1162,1163],{},"自动生成同比、环比、趋势的文字说明。",[53,1165,1166],{},"辅助管理层快速理解\"这份报表说明了什么\"。",[45,1168,1170],{"id":1169},"_2-异常检测","2. 异常检测",[50,1172,1173,1176,1179],{},[53,1174,1175],{},"识别费用、成本、应收应付的异常波动。",[53,1177,1178],{},"标记可能的问题（如某类支出突增、账期异常）。",[53,1180,1181],{},"帮财务聚焦风险点，而不是大海捞针。",[45,1183,1185],{"id":1184},"_3-辅助预测","3. 辅助预测",[50,1187,1188,1191,1194],{},[53,1189,1190],{},"基于历史数据做现金流、收入、成本的短期预测。",[53,1192,1193],{},"结合业务假设做情景分析。",[53,1195,1196],{},"预测是辅助参考，不是承诺。",[45,1198,1200],{"id":1199},"_4-数据问答","4. 数据问答",[50,1202,1203,1206],{},[53,1204,1205],{},"用自然语言问数据：\"上季度哪个产品线毛利率最高\"。",[53,1207,1208],{},"AI转为查询，返回结果——类似 BI 工具的自然语言版。",[20,1210,1212],{"id":1211},"ai不能替代的","AI不能替代的",[50,1214,1215,1221,1227,1233],{},[53,1216,1217,1220],{},[15,1218,1219],{},"记账与出表","：强规则、强合规，应以专业财务软件为准。",[53,1222,1223,1226],{},[15,1224,1225],{},"对外财报","：必须人工编制和审计，AI不可直接生成。",[53,1228,1229,1232],{},[15,1230,1231],{},"专业判断","：税务筹划、投资决策、合规判断，需要财务专业人员。",[53,1234,1235,1238],{},[15,1236,1237],{},"责任承担","：AI出错无人担责，最终决策和签字在人。",[11,1240,1241],{},[15,1242,1243],{},"AI是财务的助手，不是替代者。",[20,1245,1247],{"id":1246},"数据安全财务ai的红线","数据安全：财务AI的红线",[11,1249,1250],{},"财务数据敏感度极高，这是用AI最大的顾虑：",[170,1252,1253,1263],{},[173,1254,1255],{},[176,1256,1257,1260],{},[179,1258,1259],{},"风险点",[179,1261,1262],{},"应对做法",[189,1264,1265,1273,1281,1289],{},[176,1266,1267,1270],{},[194,1268,1269],{},"数据泄露",[194,1271,1272],{},"敏感数据不上公有云，或先脱敏",[176,1274,1275,1278],{},[194,1276,1277],{},"合规要求",[194,1279,1280],{},"涉及上市公司\u002F监管要求，必须私有化",[176,1282,1283,1286],{},[194,1284,1285],{},"模型幻觉",[194,1287,1288],{},"计算和分析结果必须人工复核",[176,1290,1291,1294],{},[194,1292,1293],{},"权限控制",[194,1295,1296],{},"AI应用也要做数据权限隔离",[11,1298,1299],{},[15,1300,1301],{},"核心原则：敏感财务数据优先私有化部署，脱敏后再分析。",[20,1303,1304],{"id":1304},"实施路径",[45,1306,1308],{"id":1307},"_1-先从低风险场景切入","1. 先从低风险场景切入",[50,1310,1311,1314,1317],{},[53,1312,1313],{},"用 AI 解读已公开\u002F已汇总的报表数据。",[53,1315,1316],{},"做趋势分析、异常提示等辅助工作。",[53,1318,1319],{},"不碰原始敏感凭证。",[45,1321,1323],{"id":1322},"_2-数据脱敏与隔离","2. 数据脱敏与隔离",[50,1325,1326,1329],{},[53,1327,1328],{},"分析前对客户名、金额等敏感字段脱敏。",[53,1330,1331],{},"AI 应用与核心财务系统数据隔离。",[45,1333,1335],{"id":1334},"_3-私有化部署","3. 私有化部署",[50,1337,1338,1341],{},[53,1339,1340],{},"有条件的企业部署私有化模型。",[53,1342,1343],{},"数据不出内网。",[45,1345,1347],{"id":1346},"_4-人机协作流程","4. 人机协作流程",[50,1349,1350,1353],{},[53,1351,1352],{},"AI 出初稿和提示，财务复核确认。",[53,1354,1355],{},"关键决策必须人工签字。",[20,1357,267],{"id":267},[170,1359,1360,1370],{},[173,1361,1362],{},[176,1363,1364,1366,1368],{},[179,1365,276],{},[179,1367,279],{},[179,1369,282],{},[189,1371,1372,1383,1394],{},[176,1373,1374,1377,1380],{},[194,1375,1376],{},"轻量分析工具",[194,1378,1379],{},"接入脱敏数据做报表解读",[194,1381,1382],{},"较低",[176,1384,1385,1388,1391],{},[194,1386,1387],{},"私有化AI分析",[194,1389,1390],{},"本地部署模型 + 财务数据集成",[194,1392,1393],{},"中到高",[176,1395,1396,1399,1402],{},[194,1397,1398],{},"完整智能财务中台",[194,1400,1401],{},"多模块 + 权限 + 审计 + 私有化",[194,1403,1404],{},"高，定制开发",[20,1406,1407],{"id":1407},"常见误区",[50,1409,1410,1416,1422,1428],{},[53,1411,1412,1415],{},[15,1413,1414],{},"让AI直接做账出表","：合规风险高，不可取。",[53,1417,1418,1421],{},[15,1419,1420],{},"原始财务数据直接传公有云","：泄露和合规风险。",[53,1423,1424,1427],{},[15,1425,1426],{},"盲目相信AI分析结论","：不复核直接用，可能出错。",[53,1429,1430,1433],{},[15,1431,1432],{},"忽视权限","：AI能看的数据范围要和岗位职责匹配。",[20,1435,319],{"id":319},[321,1437,1438,1441,1444,1447,1450],{},[53,1439,1440],{},"梳理可被AI辅助的重复分析工作。",[53,1442,1443],{},"评估数据敏感度和合规要求。",[53,1445,1446],{},"选私有化或脱敏方案。",[53,1448,1449],{},"从低风险场景试点。",[53,1451,1452],{},"建立人机协作和复核流程。",[339,1454,1455],{},[11,1456,1457],{},"广州市汉诺雷斯（HNREIS）帮企业搭建私有化的AI数据分析应用，在数据安全和合规前提下，让财务和业务团队用自然语言分析数据。把你的数据分析需求告诉我们，我们给出安全合规的方案。",{"title":37,"searchDepth":345,"depth":345,"links":1459},[1460,1466,1467,1468,1474,1475,1476],{"id":1150,"depth":345,"text":1151,"children":1461},[1462,1463,1464,1465],{"id":1154,"depth":351,"text":1155},{"id":1169,"depth":351,"text":1170},{"id":1184,"depth":351,"text":1185},{"id":1199,"depth":351,"text":1200},{"id":1211,"depth":345,"text":1212},{"id":1246,"depth":345,"text":1247},{"id":1304,"depth":345,"text":1304,"children":1469},[1470,1471,1472,1473],{"id":1307,"depth":351,"text":1308},{"id":1322,"depth":351,"text":1323},{"id":1334,"depth":351,"text":1335},{"id":1346,"depth":351,"text":1347},{"id":267,"depth":345,"text":267},{"id":1407,"depth":345,"text":1407},{"id":319,"depth":345,"text":319},"2024-06-03","AI能帮财务做报表解读、异常检测、趋势预测，但财务数据高度敏感。本文讲清AI辅助财务分析能做什么、不能做什么，以及数据安全和私有化的关键考量。",[1480,1483,1486],{"q":1481,"a":1482},"AI能直接帮我做账出报表吗？","目前不适合让AI直接生成对外财报。AI擅长的是辅助：解读已有报表、做数据透视、发现异常、辅助预测分析。记账、出表这类强规则、强合规的工作，仍应以专业财务软件和会计为准，AI作为辅助分析工具。涉及对外披露的报表必须人工复核把关。",{"q":1484,"a":1485},"财务数据这么敏感，能用云端AI吗？","要谨慎。财务数据涉及商业机密和合规要求，一般不建议直接传公有云大模型。可行做法：敏感字段脱敏后再分析、用私有化部署的模型、或只让AI处理汇总后的非敏感数据。涉及上市公司或有严格保密要求的，必须私有化部署。",{"q":1487,"a":1488},"AI做财务分析的准确率可靠吗？","对结构化数据的计算和趋势识别较可靠，但对复杂业务判断和因果分析可能出错（幻觉）。正确用法是AI出初稿和提示，财务专业人员复核确认。AI不能替代财务判断，尤其涉及投资、税务、合规决策时，必须以人为准。",[1490,1491,1492,1493],"AI财务分析","财务报表AI","AI数据分析","智能财务",{},"\u002Fblog\u002Fai-agent\u002Fai-caiwu-baobiao",{"title":1138,"description":1478},{"loc":1495},"blog\u002Fai-agent\u002Fai-caiwu-baobiao",[1500,1501,1502],"财务","数据分析","合规","Pdri7PqftoKRgTfvW1VvvY-liWyZ8b5GUMVIU9UtDsk",{"id":1505,"title":1506,"author":6,"body":1507,"category":366,"cover":367,"date":1839,"description":1840,"draft":370,"extension":371,"faq":1841,"featured":370,"image":367,"keywords":1851,"meta":1856,"navigation":388,"path":1857,"seo":1858,"sitemap":1859,"stem":1860,"tags":1861,"updated":1839,"__hash__":1865},"blog\u002Fblog\u002Fai-agent\u002Fai-canyin.md","餐饮行业怎么用AI做营销和排班",{"type":8,"value":1508,"toc":1825},[1509,1516,1519,1536,1540,1544,1555,1562,1566,1577,1581,1592,1596,1607,1611,1622,1625,1705,1711,1713,1745,1747,1801,1803,1820],[11,1510,1511,1512,1515],{},"餐饮行业人力成本高、客流波动大、点评和营销又特别耗时。",[15,1513,1514],{},"AI帮餐饮解决的不是\"做饭\"，而是那些重复、依赖经验、占用店长精力的环节。"," 这篇讲清餐饮AI的落地场景。",[20,1517,1518],{"id":1518},"餐饮的痛点",[50,1520,1521,1524,1527,1530,1533],{},[53,1522,1523],{},"排班靠店长经验，忙时人不够、闲时人过剩。",[53,1525,1526],{},"客流预测不准，备料和排班都跟着错。",[53,1528,1529],{},"大众点评、外卖平台的评价回复耗时。",[53,1531,1532],{},"朋友圈、社群营销文案写不出来。",[53,1534,1535],{},"会员复购靠拍脑袋，缺少数据支撑。",[20,1537,1539],{"id":1538},"ai能落地的场景","AI能落地的场景",[45,1541,1543],{"id":1542},"_1-客流预测与智能排班","1. 客流预测与智能排班",[50,1545,1546,1549,1552],{},[53,1547,1548],{},"基于历史客流、天气、节假日、周边活动预测各时段客流。",[53,1550,1551],{},"自动生成排班建议，匹配高峰人力需求。",[53,1553,1554],{},"连锁门店可统一调度。",[11,1556,1557,1558,1561],{},"这是餐饮AI",[15,1559,1560],{},"价值最直接","的场景——直接关联人力成本。",[45,1563,1565],{"id":1564},"_2-智能推荐与连带销售","2. 智能推荐与连带销售",[50,1567,1568,1571,1574],{},[53,1569,1570],{},"根据顾客点单、时段、季节推荐搭配。",[53,1572,1573],{},"收银\u002F小程序点餐页的智能推荐位。",[53,1575,1576],{},"提升客单价。",[45,1578,1580],{"id":1579},"_3-点评分析与自动回复","3. 点评分析与自动回复",[50,1582,1583,1586,1589],{},[53,1584,1585],{},"批量分析大众点评\u002F外卖评价，提取菜品、服务、卫生的口碑趋势。",[53,1587,1588],{},"辅助生成评价回复（人工把关后发送）。",[53,1590,1591],{},"发现差评共性问题，改进运营。",[45,1593,1595],{"id":1594},"_4-营销文案生成","4. 营销文案生成",[50,1597,1598,1601,1604],{},[53,1599,1600],{},"朋友圈、社群、公众号文案快速生成。",[53,1602,1603],{},"节日活动、新品上市的推广素材。",[53,1605,1606],{},"多门店内容统一又本地化。",[45,1608,1610],{"id":1609},"_5-会员与私域","5. 会员与私域",[50,1612,1613,1616,1619],{},[53,1614,1615],{},"会员消费数据分析，识别高价值客户和流失风险。",[53,1617,1618],{},"个性化优惠券和召回。",[53,1620,1621],{},"复购预测。",[20,1623,1624],{"id":1624},"落地的现实考量",[170,1626,1627,1642],{},[173,1628,1629],{},[176,1630,1631,1633,1636,1639],{},[179,1632,1024],{},[179,1634,1635],{},"数据依赖",[179,1637,1638],{},"难度",[179,1640,1641],{},"见效",[189,1643,1644,1656,1667,1680,1692],{},[176,1645,1646,1649,1651,1653],{},[194,1647,1648],{},"营销文案生成",[194,1650,199],{},[194,1652,199],{},[194,1654,1655],{},"快",[176,1657,1658,1661,1663,1665],{},[194,1659,1660],{},"点评分析回复",[194,1662,295],{},[194,1664,199],{},[194,1666,1655],{},[176,1668,1669,1672,1675,1677],{},[194,1670,1671],{},"客流预测排班",[194,1673,1674],{},"高（历史数据）",[194,1676,295],{},[194,1678,1679],{},"中期",[176,1681,1682,1685,1688,1690],{},[194,1683,1684],{},"智能推荐",[194,1686,1687],{},"中（点单数据）",[194,1689,295],{},[194,1691,1679],{},[176,1693,1694,1697,1699,1702],{},[194,1695,1696],{},"会员复购预测",[194,1698,210],{},[194,1700,1701],{},"中高",[194,1703,1704],{},"中长期",[11,1706,1707,1710],{},[15,1708,1709],{},"先从低门槛、见效快的场景切入","（文案、点评），再逐步做需要数据的场景（排班、推荐）。",[20,1712,232],{"id":232},[50,1714,1715,1721,1727,1733,1739],{},[53,1716,1717,1720],{},[15,1718,1719],{},"数据质量差就上预测","：历史客流数据不准，预测也不准。",[53,1722,1723,1726],{},[15,1724,1725],{},"自动回复不经审核","：AI回复可能不当，引发公关问题。",[53,1728,1729,1732],{},[15,1730,1731],{},"忽视门店执行","：AI给排班建议，门店不执行等于零。",[53,1734,1735,1738],{},[15,1736,1737],{},"一上来搞大而全","：建议单点突破，验证后再扩展。",[53,1740,1741,1744],{},[15,1742,1743],{},"把AI当万能","：餐饮核心仍是产品和体验，AI是提效工具。",[20,1746,267],{"id":267},[170,1748,1749,1759],{},[173,1750,1751],{},[176,1752,1753,1755,1757],{},[179,1754,276],{},[179,1756,279],{},[179,1758,282],{},[189,1760,1761,1771,1781,1790],{},[176,1762,1763,1766,1769],{},[194,1764,1765],{},"文案\u002F点评辅助",[194,1767,1768],{},"SaaS 或 API，按量计费",[194,1770,199],{},[176,1772,1773,1776,1779],{},[194,1774,1775],{},"客流排班模块",[194,1777,1778],{},"和门店系统打通，定制",[194,1780,295],{},[176,1782,1783,1785,1788],{},[194,1784,1684],{},[194,1786,1787],{},"点餐\u002F收银集成",[194,1789,295],{},[176,1791,1792,1795,1798],{},[194,1793,1794],{},"连锁智能运营中台",[194,1796,1797],{},"多门店 + 预测 + 推荐 + 私域",[194,1799,1800],{},"高，定制",[20,1802,319],{"id":319},[321,1804,1805,1808,1811,1814,1817],{},[53,1806,1807],{},"找出最耗时的环节（排班？点评？文案？）。",[53,1809,1810],{},"从低成本场景试点（文案、点评辅助）。",[53,1812,1813],{},"积累门店数据，再做客流预测排班。",[53,1815,1816],{},"逐步扩展到推荐、会员。",[53,1818,1819],{},"始终让人把关关键输出。",[339,1821,1822],{},[11,1823,1824],{},"广州市汉诺雷斯（HNREIS）帮餐饮企业落地AI应用，从客流预测排班、点评分析到营销文案和会员私域，和小程序、收银系统打通。把你的门店运营痛点告诉我们，我们给出务实的AI落地方案。",{"title":37,"searchDepth":345,"depth":345,"links":1826},[1827,1828,1835,1836,1837,1838],{"id":1518,"depth":345,"text":1518},{"id":1538,"depth":345,"text":1539,"children":1829},[1830,1831,1832,1833,1834],{"id":1542,"depth":351,"text":1543},{"id":1564,"depth":351,"text":1565},{"id":1579,"depth":351,"text":1580},{"id":1594,"depth":351,"text":1595},{"id":1609,"depth":351,"text":1610},{"id":1624,"depth":345,"text":1624},{"id":232,"depth":345,"text":232},{"id":267,"depth":345,"text":267},{"id":319,"depth":345,"text":319},"2024-06-16","餐饮门店客流波动大、排班靠经验、点评回复耗时。AI能帮餐饮做客流预测排班、智能推荐、点评分析和营销文案，本文讲清落地场景与成本。",[1842,1845,1848],{"q":1843,"a":1844},"餐饮店有必要上AI吗？","看规模和痛点。单店小店，痛点不突出可以先不上；连锁或多门店，客流预测排班、点评批量回复、营销文案生成这类重复工作，AI能明显提效。建议从最耗时的环节切入，比如用AI生成朋友圈\u002F社群文案、辅助排班，先看到效果再扩展。",{"q":1846,"a":1847},"AI排班靠谱吗？","AI排班是基于历史客流、天气、节假日等数据预测各时段客流，再匹配人力需求，比纯靠店长经验更稳定。但它依赖历史数据质量和门店执行，适合作为排班参考，最终仍需店长结合实际情况调整。新店数据不足时预测效果有限。",{"q":1849,"a":1850},"餐饮用AI要花多少钱？","轻量场景（文案生成、点评回复辅助）用现有SaaS或大模型API，成本很低；客流预测排班、智能推荐这类要和门店系统打通的，属于定制开发，几万到十几万不等，看门店数量和集成深度。建议先从低成本场景验证价值。",[1852,1853,1854,1855],"餐饮AI","餐饮排班AI","AI营销文案","餐饮数字化",{},"\u002Fblog\u002Fai-agent\u002Fai-canyin",{"title":1506,"description":1840},{"loc":1857},"blog\u002Fai-agent\u002Fai-canyin",[1862,1863,1864],"餐饮","营销","排班","wwPJmcDR-PabsSXv_ZcpsIAe0cgabBUgBGYdFdSCdPc",1781688906875]