[{"data":1,"prerenderedAt":1884},["ShallowReactive",2],{"blog-\u002Fblog\u002Fai-agent\u002Fkaiyuan-bikaiyuan-damoxing":3,"blog-related-\u002Fblog\u002Fai-agent\u002Fkaiyuan-bikaiyuan-damoxing":415},{"id":4,"title":5,"author":6,"body":7,"category":386,"cover":387,"date":388,"description":389,"draft":390,"extension":391,"faq":392,"featured":390,"image":387,"keywords":402,"meta":405,"navigation":406,"path":407,"seo":408,"sitemap":409,"stem":410,"tags":411,"updated":388,"__hash__":414},"blog\u002Fblog\u002Fai-agent\u002Fkaiyuan-bikaiyuan-damoxing.md","开源和闭源大模型怎么选","HNREIS",{"type":8,"value":9,"toc":364},"minimark",[10,19,23,26,30,46,49,63,66,69,72,86,89,103,106,195,198,201,218,221,235,238,249,252,284,287,337,340,358],[11,12,13,14,18],"p",{},"企业选大模型，第一个决策往往是：闭源商业API还是开源私有化？",[15,16,17],"strong",{},"两者各有取舍，要按数据合规、成本、能力需求权衡，不是越强越贵越好。"," 这篇讲清怎么选。",[20,21,22],"h2",{"id":22},"闭源大模型",[11,24,25],{},"代表：商业API（GPT、Claude、通义、文心等）。",[27,28,29],"h3",{"id":29},"优势",[31,32,33,37,40,43],"ul",{},[34,35,36],"li",{},"能力强，开箱即用。",[34,38,39],{},"无需算力和运维。",[34,41,42],{},"按量付费，起步成本低。",[34,44,45],{},"更新快。",[27,47,48],{"id":48},"劣势",[31,50,51,54,57,60],{},[34,52,53],{},"数据要传出去（合规风险）。",[34,55,56],{},"按量付费，量大成本高。",[34,58,59],{},"依赖厂商，受调价和停服影响。",[34,61,62],{},"定制受限。",[20,64,65],{"id":65},"开源大模型",[11,67,68],{},"代表：可私有化部署（Qwen、DeepSeek、Llama等）。",[27,70,29],{"id":71},"优势-1",[31,73,74,77,80,83],{},[34,75,76],{},"可私有化，数据不出内网。",[34,78,79],{},"可控、可定制（微调）。",[34,81,82],{},"无按量费，量大降本。",[34,84,85],{},"无厂商锁定。",[27,87,48],{"id":88},"劣势-1",[31,90,91,94,97,100],{},[34,92,93],{},"需要算力（GPU）投入。",[34,95,96],{},"需要运维能力。",[34,98,99],{},"能力可能略逊顶尖闭源。",[34,101,102],{},"要自己持续优化。",[20,104,105],{"id":105},"选择维度",[107,108,109,125],"table",{},[110,111,112],"thead",{},[113,114,115,119,122],"tr",{},[116,117,118],"th",{},"维度",[116,120,121],{},"闭源",[116,123,124],{},"开源",[126,127,128,140,151,162,173,184],"tbody",{},[113,129,130,134,137],{},[131,132,133],"td",{},"数据合规",[131,135,136],{},"弱（数据出境）",[131,138,139],{},"强（私有化）",[113,141,142,145,148],{},[131,143,144],{},"起步成本",[131,146,147],{},"低",[131,149,150],{},"高（算力）",[113,152,153,156,159],{},[131,154,155],{},"量大成本",[131,157,158],{},"高（按量）",[131,160,161],{},"低（固定）",[113,163,164,167,170],{},[131,165,166],{},"能力",[131,168,169],{},"强",[131,171,172],{},"较强（够多数场景）",[113,174,175,178,181],{},[131,176,177],{},"运维",[131,179,180],{},"无",[131,182,183],{},"需要",[113,185,186,189,192],{},[131,187,188],{},"定制",[131,190,191],{},"受限",[131,193,194],{},"可微调",[20,196,197],{"id":197},"按场景选",[27,199,200],{"id":200},"闭源适合",[31,202,203,206,209,212,215],{},[34,204,205],{},"数据不敏感。",[34,207,208],{},"追求顶尖能力。",[34,210,211],{},"不想运维。",[34,213,214],{},"用量不大。",[34,216,217],{},"快速验证。",[27,219,220],{"id":220},"开源适合",[31,222,223,226,229,232],{},[34,224,225],{},"数据敏感，要私有化。",[34,227,228],{},"用量大，想降按量成本。",[34,230,231],{},"要深度定制。",[34,233,234],{},"合规要求高（金融\u002F政务\u002F医疗）。",[27,236,237],{"id":237},"混合策略",[31,239,240,243,246],{},[34,241,242],{},"敏感数据用开源私有化。",[34,244,245],{},"通用场景用闭源API。",[34,247,248],{},"用AI网关统一管理。",[20,250,251],{"id":251},"别踩的坑",[31,253,254,260,266,272,278],{},[34,255,256,259],{},[15,257,258],{},"数据敏感还用闭源API","：合规风险。",[34,261,262,265],{},[15,263,264],{},"用量小也私有化","：算力运维不划算。",[34,267,268,271],{},[15,269,270],{},"忽视闭源按量成本","：量大时很贵。",[34,273,274,277],{},[15,275,276],{},"开源不评估能力","：顶尖场景可能不够。",[34,279,280,283],{},[15,281,282],{},"忽视运维能力","：私有化要人维护。",[20,285,286],{"id":286},"成本参考",[107,288,289,302],{},[110,290,291],{},[113,292,293,296,299],{},[116,294,295],{},"方案",[116,297,298],{},"说明",[116,300,301],{},"成本量级",[126,303,304,315,326],{},[113,305,306,309,312],{},[131,307,308],{},"闭源API",[131,310,311],{},"按量计费",[131,313,314],{},"用量决定",[113,316,317,320,323],{},[131,318,319],{},"开源私有化",[131,321,322],{},"算力+运维",[131,324,325],{},"一次性+持续",[113,327,328,331,334],{},[131,329,330],{},"混合（网关）",[131,332,333],{},"敏感私有+通用API",[131,335,336],{},"中",[20,338,339],{"id":339},"怎么开始",[341,342,343,346,349,352,355],"ol",{},[34,344,345],{},"评估数据合规要求。",[34,347,348],{},"评估用量和成本。",[34,350,351],{},"评估能力需求和运维能力。",[34,353,354],{},"按场景选闭源\u002F开源\u002F混合。",[34,356,357],{},"用AI网关统一管理多模型。",[359,360,361],"blockquote",{},[11,362,363],{},"广州市汉诺雷斯（HNREIS）帮企业做大模型选型和部署，从闭源API集成到开源私有化部署和混合架构。把你的场景和合规要求告诉我们，我们给出务实的选型方案。",{"title":365,"searchDepth":366,"depth":366,"links":367},"",2,[368,373,377,378,383,384,385],{"id":22,"depth":366,"text":22,"children":369},[370,372],{"id":29,"depth":371,"text":29},3,{"id":48,"depth":371,"text":48},{"id":65,"depth":366,"text":65,"children":374},[375,376],{"id":71,"depth":371,"text":29},{"id":88,"depth":371,"text":48},{"id":105,"depth":366,"text":105},{"id":197,"depth":366,"text":197,"children":379},[380,381,382],{"id":200,"depth":371,"text":200},{"id":220,"depth":371,"text":220},{"id":237,"depth":371,"text":237},{"id":251,"depth":366,"text":251},{"id":286,"depth":366,"text":286},{"id":339,"depth":366,"text":339},"ai-agent",null,"2026-01-20","闭源大模型（GPT\u002FClaude\u002F通义）能力强开箱即用，开源（Qwen\u002FDeepSeek\u002FLlama）可私有化可控。本文讲清开闭源的选择权衡和企业决策。",false,"md",[393,396,399],{"q":394,"a":395},"开源和闭源大模型怎么选？","看场景和诉求。闭源（商业API）能力强、开箱即用、按量付费，适合追求效果、不想运维、数据不敏感的场景；开源（可私有化）可控、数据不出内网、无按量费但要算力和运维，适合数据敏感、要私有化、用量大想降本的场景。按数据合规、成本、能力需求权衡。",{"q":397,"a":398},"开源大模型能力够用吗？","主流开源模型（如Qwen、DeepSeek等）能力已相当强，很多企业场景够用，尤其配合RAG和微调。但在最复杂的推理、创意等顶尖能力上，闭源模型仍有优势。建议按场景评估——够用就开源（还能私有化），需要顶尖能力且不敏感就用闭源。",{"q":400,"a":401},"私有化部署要什么条件？","主要是算力（GPU服务器或云算力，按模型规模）、运维能力（部署、监控、更新）、以及持续优化投入。开源大模型私有化不是\"免费\"——算力和运维是真金白银。建议数据敏感或用量大的场景才私有化，否则用API更省心。",[65,22,403,404],"大模型选型","私有化部署",{},true,"\u002Fblog\u002Fai-agent\u002Fkaiyuan-bikaiyuan-damoxing",{"title":5,"description":389},{"loc":407},"blog\u002Fai-agent\u002Fkaiyuan-bikaiyuan-damoxing",[412,413,124],"大模型","选型","_DJbdg3r6b1ethLeREhe3MFgbIYpsLte2SAN0722r6U",[416,803,1153,1521],{"id":417,"title":418,"author":6,"body":419,"category":386,"cover":387,"date":779,"description":780,"draft":390,"extension":391,"faq":781,"featured":390,"image":387,"keywords":791,"meta":795,"navigation":406,"path":796,"seo":797,"sitemap":798,"stem":799,"tags":800,"updated":779,"__hash__":802},"blog\u002Fblog\u002Fai-agent\u002Fagent-kuangjia.md","主流Agent框架怎么选",{"type":8,"value":420,"toc":766},[421,427,431,434,454,461,464,601,606,609,613,625,629,636,640,643,647,650,652,684,686,742,744,761],[11,422,423,424,18],{},"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。",[15,425,426],{},"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。",[20,428,430],{"id":429},"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",[11,432,433],{},"很多人一上来就要\"上框架\"，其实很多场景不需要：",[31,435,436,442,448],{},[34,437,438,441],{},[15,439,440],{},"简单问答","：直接调大模型 API，几十行代码搞定，不用框架。",[34,443,444,447],{},[15,445,446],{},"RAG 知识库问答","：用轻量库或框架的检索能力，中等复杂度。",[34,449,450,453],{},[15,451,452],{},"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。",[11,455,456,457,460],{},"需求复杂度决定框架价值。",[15,458,459],{},"简单需求硬上重框架，是典型的过度设计","。",[20,462,463],{"id":463},"主流框架对比",[107,465,466,485],{},[110,467,468],{},[113,469,470,473,476,479,482],{},[116,471,472],{},"框架",[116,474,475],{},"类型",[116,477,478],{},"核心定位",[116,480,481],{},"适合谁",[116,483,484],{},"主要局限",[126,486,487,504,520,536,552,568,585],{},[113,488,489,492,495,498,501],{},[131,490,491],{},"LangChain",[131,493,494],{},"代码框架",[131,496,497],{},"通用 LLM 应用编排，生态最全",[131,499,500],{},"有研发团队、要深度定制",[131,502,503],{},"抽象较重，早期 API 频繁变动",[113,505,506,509,511,514,517],{},[131,507,508],{},"LangGraph",[131,510,494],{},[131,512,513],{},"基于图的状态机，做可控多步 Agent",[131,515,516],{},"复杂流程、需要稳定状态控制",[131,518,519],{},"学习曲线陡",[113,521,522,525,527,530,533],{},[131,523,524],{},"LlamaIndex",[131,526,494],{},[131,528,529],{},"数据连接与检索（RAG 见长）",[131,531,532],{},"以知识库\u002F文档检索为核心",[131,534,535],{},"Agent 编排不如 LangChain 全",[113,537,538,541,543,546,549],{},[131,539,540],{},"AutoGen",[131,542,494],{},[131,544,545],{},"多智能体对话协作",[131,547,548],{},"多 Agent 角色分工场景",[131,550,551],{},"偏研究，工程化要自己补",[113,553,554,557,559,562,565],{},[131,555,556],{},"CrewAI",[131,558,494],{},[131,560,561],{},"角色化多 Agent 协作，上手快",[131,563,564],{},"快速搭多角色协作原型",[131,566,567],{},"复杂控制力有限",[113,569,570,573,576,579,582],{},[131,571,572],{},"Dify",[131,574,575],{},"低代码平台",[131,577,578],{},"可视化编排 + 知识库 + 应用托管",[131,580,581],{},"快速落地、业务人员参与",[131,583,584],{},"深度定制受平台能力限制",[113,586,587,590,592,595,598],{},[131,588,589],{},"Coze",[131,591,575],{},[131,593,594],{},"字节出品，插件生态丰富",[131,596,597],{},"快速搭建对话型应用",[131,599,600],{},"偏消费侧，企业私有化受限",[11,602,603],{},[15,604,605],{},"没有\"最好\"的框架，只有\"最匹配\"的框架。",[20,607,608],{"id":608},"选型要看的维度",[27,610,612],{"id":611},"_1-编程式还是低代码","1. 编程式还是低代码",[31,614,615,620],{},[34,616,617,619],{},[15,618,494],{},"（LangChain\u002FLlamaIndex 等）：灵活、可控、能深度集成业务系统，但要有研发投入。",[34,621,622,624],{},[15,623,575],{},"（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。",[27,626,628],{"id":627},"_2-单-agent-还是多-agent","2. 单 Agent 还是多 Agent",[11,630,631,632,635],{},"大部分企业场景，",[15,633,634],{},"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。",[27,637,639],{"id":638},"_3-能不能私有化部署","3. 能不能私有化部署",[11,641,642],{},"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。",[27,644,646],{"id":645},"_4-社区活跃度与稳定性","4. 社区活跃度与稳定性",[11,648,649],{},"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。",[20,651,251],{"id":251},[31,653,654,660,666,672,678],{},[34,655,656,659],{},[15,657,658],{},"过度设计","：简单需求上重框架，开发和维护成本翻倍。",[34,661,662,665],{},[15,663,664],{},"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。",[34,667,668,671],{},[15,669,670],{},"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。",[34,673,674,677],{},[15,675,676],{},"盲目追新","：新框架未必成熟，企业项目要稳。",[34,679,680,683],{},[15,681,682],{},"忽视可观测","：Agent 行为不确定性高，没有日志和追踪很难调试。",[20,685,286],{"id":286},[107,687,688,698],{},[110,689,690],{},[113,691,692,694,696],{},[116,693,295],{},[116,695,298],{},[116,697,301],{},[126,699,700,710,720,731],{},[113,701,702,705,708],{},[131,703,704],{},"自研（直接调 API）",[131,706,707],{},"简单场景，研发 1-2 周",[131,709,147],{},[113,711,712,715,718],{},[131,713,714],{},"基于代码框架开发",[131,716,717],{},"中等复杂度，研发 3-8 周",[131,719,336],{},[113,721,722,725,728],{},[131,723,724],{},"低代码平台搭建",[131,726,727],{},"快速原型到中等应用",[131,729,730],{},"平台费 + 少量定制",[113,732,733,736,739],{},[131,734,735],{},"复杂多 Agent 系统",[131,737,738],{},"多角色协作 + 工具集成 + 私有化",[131,740,741],{},"较高，定制开发",[20,743,339],{"id":339},[341,745,746,749,752,755,758],{},[34,747,748],{},"明确场景和复杂度（是否真需要框架）。",[34,750,751],{},"评估团队能力（能否驾驭代码框架）。",[34,753,754],{},"对比 2-3 个候选框架，做小范围验证。",[34,756,757],{},"关注私有化、可观测、可迁移。",[34,759,760],{},"把核心资产与框架解耦。",[359,762,763],{},[11,764,765],{},"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":365,"searchDepth":366,"depth":366,"links":767},[768,769,770,776,777,778],{"id":429,"depth":366,"text":430},{"id":463,"depth":366,"text":463},{"id":608,"depth":366,"text":608,"children":771},[772,773,774,775],{"id":611,"depth":371,"text":612},{"id":627,"depth":371,"text":628},{"id":638,"depth":371,"text":639},{"id":645,"depth":371,"text":646},{"id":251,"depth":366,"text":251},{"id":286,"depth":366,"text":286},{"id":339,"depth":366,"text":339},"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。",[782,785,788],{"q":783,"a":784},"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":786,"a":787},"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":789,"a":790},"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。",[792,491,572,793,794],"Agent框架","AI智能体开发","框架选型",{},"\u002Fblog\u002Fai-agent\u002Fagent-kuangjia",{"title":418,"description":780},{"loc":796},"blog\u002Fai-agent\u002Fagent-kuangjia",[801,794,412],"Agent","yYDRGbvFeaR227ajWv5qiaN844ebZL6NwOSs4iOxEcE",{"id":804,"title":805,"author":6,"body":806,"category":386,"cover":387,"date":1125,"description":1126,"draft":390,"extension":391,"faq":1127,"featured":390,"image":387,"keywords":1137,"meta":1143,"navigation":406,"path":1144,"seo":1145,"sitemap":1146,"stem":1147,"tags":1148,"updated":1125,"__hash__":1152},"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":807,"toc":1116},[808,815,818,832,838,841,943,946,951,964,967,970,973,987,990,994,997,1029,1032,1078,1082,1085,1111],[11,809,810,811,814],{},"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。",[15,812,813],{},"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。",[20,816,817],{"id":817},"一句话区分",[31,819,820,826],{},[34,821,822,825],{},[15,823,824],{},"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。",[34,827,828,831],{},[15,829,830],{},"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。",[11,833,834,835],{},"本质区别：",[15,836,837],{},"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\"。",[20,839,840],{"id":840},"六维对比表",[107,842,843,853],{},[110,844,845],{},[113,846,847,849,851],{},[116,848,166],{},[116,850,824],{},[116,852,830],{},[126,854,855,866,877,888,899,910,921,932],{},[113,856,857,860,863],{},[131,858,859],{},"对话方式",[131,861,862],{},"关键词 + 流程图",[131,864,865],{},"大模型理解自然语言",[113,867,868,871,874],{},[131,869,870],{},"理解能力",[131,872,873],{},"死板，换个说法就懵",[131,875,876],{},"灵活，懂口语、省略、上下文",[113,878,879,882,885],{},[131,880,881],{},"自主性",[131,883,884],{},"被动应答",[131,886,887],{},"主动规划、多步执行",[113,889,890,893,896],{},[131,891,892],{},"工具调用",[131,894,895],{},"几乎不能",[131,897,898],{},"能查库、调 API、发通知、生成文件",[113,900,901,904,907],{},[131,902,903],{},"记忆",[131,905,906],{},"无或当次会话",[131,908,909],{},"长期记忆（记住用户历史）",[113,911,912,915,918],{},[131,913,914],{},"复杂任务",[131,916,917],{},"只能引导转人工",[131,919,920],{},"可独立完成跨系统任务",[113,922,923,926,929],{},[131,924,925],{},"知识更新",[131,927,928],{},"改流程图\u002F规则",[131,930,931],{},"喂新文档即可",[113,933,934,937,940],{},[131,935,936],{},"用户体验",[131,938,939],{},"机械、易激怒用户",[131,941,942],{},"接近真人，体验好",[20,944,945],{"id":945},"真实场景对比",[11,947,948],{},[15,949,950],{},"场景：用户问\"我上周买的那个能退吗\"",[31,952,953,959],{},[34,954,955,958],{},[15,956,957],{},"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。",[34,960,961,963],{},[15,962,830],{},"：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。",[11,965,966],{},"整个流程 AI Agent 自己做完，用户感觉\"这个客服真懂\"。",[20,968,969],{"id":969},"为什么传统机器人体验差",[11,971,972],{},"传统机器人的工作机制是\"if 用户说 X，then 回 Y\"。问题在于：",[31,974,975,978,981,984],{},[34,976,977],{},"用户的说法千变万化，关键词覆盖不全。",[34,979,980],{},"业务一复杂，流程图爆炸，维护噩梦。",[34,982,983],{},"没法理解上下文，每句都从头开始。",[34,985,986],{},"没法真正\"做事\"，只能\"指路\"。",[11,988,989],{},"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\"。",[20,991,993],{"id":992},"ai-agent-为什么能真正解决问题","AI Agent 为什么能真正解决问题",[11,995,996],{},"AI Agent 的核心能力：",[31,998,999,1005,1011,1017,1023],{},[34,1000,1001,1004],{},[15,1002,1003],{},"理解自然语言","：基于大模型，用户怎么说都能懂。",[34,1006,1007,1010],{},[15,1008,1009],{},"调用工具","：接业务系统（CRM、订单、知识库），能真正查数据、改数据。",[34,1012,1013,1016],{},[15,1014,1015],{},"规划任务","：把复杂目标拆成步骤，自己执行。",[34,1018,1019,1022],{},[15,1020,1021],{},"长期记忆","：记住用户历史，体验连贯。",[34,1024,1025,1028],{},[15,1026,1027],{},"知识可更新","：喂新文档即学习，不用改代码。",[20,1030,1031],{"id":1031},"企业该上哪种",[107,1033,1034,1044],{},[110,1035,1036],{},[113,1037,1038,1041],{},[116,1039,1040],{},"场景",[116,1042,1043],{},"建议",[126,1045,1046,1054,1062,1070],{},[113,1047,1048,1051],{},[131,1049,1050],{},"客服量大、问题多样",[131,1052,1053],{},"AI Agent（体验好、省人工）",[113,1055,1056,1059],{},[131,1057,1058],{},"问题高度标准化、简单",[131,1060,1061],{},"传统机器人够用，或 AI Agent 低配版",[113,1063,1064,1067],{},[131,1065,1066],{},"需要跨系统办事（查单、改单、退款）",[131,1068,1069],{},"AI Agent（能调工具）",[113,1071,1072,1075],{},[131,1073,1074],{},"预算极有限、想先试水",[131,1076,1077],{},"AI Agent MVP（先接知识库做问答）",[20,1079,1081],{"id":1080},"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",[11,1083,1084],{},"不用一次性推翻：",[341,1086,1087,1093,1099,1105],{},[34,1088,1089,1092],{},[15,1090,1091],{},"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。",[34,1094,1095,1098],{},[15,1096,1097],{},"再接业务系统","：让 AI 能查订单、查库存、发通知，真正\"办事\"。",[34,1100,1101,1104],{},[15,1102,1103],{},"加人工兜底","：AI 把握不大时转人工，持续优化。",[34,1106,1107,1110],{},[15,1108,1109],{},"逐步替换","：AI 能力稳定后，逐步下线老机器人的场景。",[359,1112,1113],{},[11,1114,1115],{},"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":365,"searchDepth":366,"depth":366,"links":1117},[1118,1119,1120,1121,1122,1123,1124],{"id":817,"depth":366,"text":817},{"id":840,"depth":366,"text":840},{"id":945,"depth":366,"text":945},{"id":969,"depth":366,"text":969},{"id":992,"depth":366,"text":993},{"id":1031,"depth":366,"text":1031},{"id":1080,"depth":366,"text":1081},"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。",[1128,1131,1134],{"q":1129,"a":1130},"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":1132,"a":1133},"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":1135,"a":1136},"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。",[1138,1139,1140,1141,1142],"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":805,"description":1126},{"loc":1144},"blog\u002Fai-agent\u002Fai-agent-vs-chatbot",[1149,1150,1151],"对比","AI","概念","8qKIdcO08ytoPMy5eHwEUZP4TpzwR1mih2PoU1dQnm0",{"id":1154,"title":1155,"author":6,"body":1156,"category":386,"cover":387,"date":1494,"description":1495,"draft":390,"extension":391,"faq":1496,"featured":390,"image":387,"keywords":1506,"meta":1511,"navigation":406,"path":1512,"seo":1513,"sitemap":1514,"stem":1515,"tags":1516,"updated":1494,"__hash__":1520},"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":1157,"toc":1475},[1158,1165,1169,1173,1184,1188,1199,1203,1214,1218,1226,1230,1256,1261,1265,1268,1314,1319,1322,1326,1337,1341,1349,1353,1361,1365,1373,1375,1422,1425,1451,1453,1470],[11,1159,1160,1161,1164],{},"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。",[15,1162,1163],{},"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。",[20,1166,1168],{"id":1167},"ai在财务分析里能做什么","AI在财务分析里能做什么",[27,1170,1172],{"id":1171},"_1-报表解读与可视化分析","1. 报表解读与可视化分析",[31,1174,1175,1178,1181],{},[34,1176,1177],{},"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。",[34,1179,1180],{},"自动生成同比、环比、趋势的文字说明。",[34,1182,1183],{},"辅助管理层快速理解\"这份报表说明了什么\"。",[27,1185,1187],{"id":1186},"_2-异常检测","2. 异常检测",[31,1189,1190,1193,1196],{},[34,1191,1192],{},"识别费用、成本、应收应付的异常波动。",[34,1194,1195],{},"标记可能的问题（如某类支出突增、账期异常）。",[34,1197,1198],{},"帮财务聚焦风险点，而不是大海捞针。",[27,1200,1202],{"id":1201},"_3-辅助预测","3. 辅助预测",[31,1204,1205,1208,1211],{},[34,1206,1207],{},"基于历史数据做现金流、收入、成本的短期预测。",[34,1209,1210],{},"结合业务假设做情景分析。",[34,1212,1213],{},"预测是辅助参考，不是承诺。",[27,1215,1217],{"id":1216},"_4-数据问答","4. 数据问答",[31,1219,1220,1223],{},[34,1221,1222],{},"用自然语言问数据：\"上季度哪个产品线毛利率最高\"。",[34,1224,1225],{},"AI转为查询，返回结果——类似 BI 工具的自然语言版。",[20,1227,1229],{"id":1228},"ai不能替代的","AI不能替代的",[31,1231,1232,1238,1244,1250],{},[34,1233,1234,1237],{},[15,1235,1236],{},"记账与出表","：强规则、强合规，应以专业财务软件为准。",[34,1239,1240,1243],{},[15,1241,1242],{},"对外财报","：必须人工编制和审计，AI不可直接生成。",[34,1245,1246,1249],{},[15,1247,1248],{},"专业判断","：税务筹划、投资决策、合规判断，需要财务专业人员。",[34,1251,1252,1255],{},[15,1253,1254],{},"责任承担","：AI出错无人担责，最终决策和签字在人。",[11,1257,1258],{},[15,1259,1260],{},"AI是财务的助手，不是替代者。",[20,1262,1264],{"id":1263},"数据安全财务ai的红线","数据安全：财务AI的红线",[11,1266,1267],{},"财务数据敏感度极高，这是用AI最大的顾虑：",[107,1269,1270,1280],{},[110,1271,1272],{},[113,1273,1274,1277],{},[116,1275,1276],{},"风险点",[116,1278,1279],{},"应对做法",[126,1281,1282,1290,1298,1306],{},[113,1283,1284,1287],{},[131,1285,1286],{},"数据泄露",[131,1288,1289],{},"敏感数据不上公有云，或先脱敏",[113,1291,1292,1295],{},[131,1293,1294],{},"合规要求",[131,1296,1297],{},"涉及上市公司\u002F监管要求，必须私有化",[113,1299,1300,1303],{},[131,1301,1302],{},"模型幻觉",[131,1304,1305],{},"计算和分析结果必须人工复核",[113,1307,1308,1311],{},[131,1309,1310],{},"权限控制",[131,1312,1313],{},"AI应用也要做数据权限隔离",[11,1315,1316],{},[15,1317,1318],{},"核心原则：敏感财务数据优先私有化部署，脱敏后再分析。",[20,1320,1321],{"id":1321},"实施路径",[27,1323,1325],{"id":1324},"_1-先从低风险场景切入","1. 先从低风险场景切入",[31,1327,1328,1331,1334],{},[34,1329,1330],{},"用 AI 解读已公开\u002F已汇总的报表数据。",[34,1332,1333],{},"做趋势分析、异常提示等辅助工作。",[34,1335,1336],{},"不碰原始敏感凭证。",[27,1338,1340],{"id":1339},"_2-数据脱敏与隔离","2. 数据脱敏与隔离",[31,1342,1343,1346],{},[34,1344,1345],{},"分析前对客户名、金额等敏感字段脱敏。",[34,1347,1348],{},"AI 应用与核心财务系统数据隔离。",[27,1350,1352],{"id":1351},"_3-私有化部署","3. 私有化部署",[31,1354,1355,1358],{},[34,1356,1357],{},"有条件的企业部署私有化模型。",[34,1359,1360],{},"数据不出内网。",[27,1362,1364],{"id":1363},"_4-人机协作流程","4. 人机协作流程",[31,1366,1367,1370],{},[34,1368,1369],{},"AI 出初稿和提示，财务复核确认。",[34,1371,1372],{},"关键决策必须人工签字。",[20,1374,286],{"id":286},[107,1376,1377,1387],{},[110,1378,1379],{},[113,1380,1381,1383,1385],{},[116,1382,295],{},[116,1384,298],{},[116,1386,301],{},[126,1388,1389,1400,1411],{},[113,1390,1391,1394,1397],{},[131,1392,1393],{},"轻量分析工具",[131,1395,1396],{},"接入脱敏数据做报表解读",[131,1398,1399],{},"较低",[113,1401,1402,1405,1408],{},[131,1403,1404],{},"私有化AI分析",[131,1406,1407],{},"本地部署模型 + 财务数据集成",[131,1409,1410],{},"中到高",[113,1412,1413,1416,1419],{},[131,1414,1415],{},"完整智能财务中台",[131,1417,1418],{},"多模块 + 权限 + 审计 + 私有化",[131,1420,1421],{},"高，定制开发",[20,1423,1424],{"id":1424},"常见误区",[31,1426,1427,1433,1439,1445],{},[34,1428,1429,1432],{},[15,1430,1431],{},"让AI直接做账出表","：合规风险高，不可取。",[34,1434,1435,1438],{},[15,1436,1437],{},"原始财务数据直接传公有云","：泄露和合规风险。",[34,1440,1441,1444],{},[15,1442,1443],{},"盲目相信AI分析结论","：不复核直接用，可能出错。",[34,1446,1447,1450],{},[15,1448,1449],{},"忽视权限","：AI能看的数据范围要和岗位职责匹配。",[20,1452,339],{"id":339},[341,1454,1455,1458,1461,1464,1467],{},[34,1456,1457],{},"梳理可被AI辅助的重复分析工作。",[34,1459,1460],{},"评估数据敏感度和合规要求。",[34,1462,1463],{},"选私有化或脱敏方案。",[34,1465,1466],{},"从低风险场景试点。",[34,1468,1469],{},"建立人机协作和复核流程。",[359,1471,1472],{},[11,1473,1474],{},"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":365,"searchDepth":366,"depth":366,"links":1476},[1477,1483,1484,1485,1491,1492,1493],{"id":1167,"depth":366,"text":1168,"children":1478},[1479,1480,1481,1482],{"id":1171,"depth":371,"text":1172},{"id":1186,"depth":371,"text":1187},{"id":1201,"depth":371,"text":1202},{"id":1216,"depth":371,"text":1217},{"id":1228,"depth":366,"text":1229},{"id":1263,"depth":366,"text":1264},{"id":1321,"depth":366,"text":1321,"children":1486},[1487,1488,1489,1490],{"id":1324,"depth":371,"text":1325},{"id":1339,"depth":371,"text":1340},{"id":1351,"depth":371,"text":1352},{"id":1363,"depth":371,"text":1364},{"id":286,"depth":366,"text":286},{"id":1424,"depth":366,"text":1424},{"id":339,"depth":366,"text":339},"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。",[1497,1500,1503],{"q":1498,"a":1499},"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":1501,"a":1502},"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":1504,"a":1505},"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。",[1507,1508,1509,1510],"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":1155,"description":1495},{"loc":1512},"blog\u002Fai-agent\u002Fai-caiwu-baobiao",[1517,1518,1519],"财务","数据分析","合规","Pdri7PqftoKRgTfvW1VvvY-liWyZ8b5GUMVIU9UtDsk",{"id":1522,"title":1523,"author":6,"body":1524,"category":386,"cover":387,"date":1857,"description":1858,"draft":390,"extension":391,"faq":1859,"featured":390,"image":387,"keywords":1869,"meta":1874,"navigation":406,"path":1875,"seo":1876,"sitemap":1877,"stem":1878,"tags":1879,"updated":1857,"__hash__":1883},"blog\u002Fblog\u002Fai-agent\u002Fai-canyin.md","餐饮行业怎么用AI做营销和排班",{"type":8,"value":1525,"toc":1843},[1526,1533,1536,1553,1557,1561,1572,1579,1583,1594,1598,1609,1613,1624,1628,1639,1642,1723,1729,1731,1763,1765,1819,1821,1838],[11,1527,1528,1529,1532],{},"餐饮行业人力成本高、客流波动大、点评和营销又特别耗时。",[15,1530,1531],{},"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。",[20,1534,1535],{"id":1535},"餐饮的痛点",[31,1537,1538,1541,1544,1547,1550],{},[34,1539,1540],{},"排班靠店长经验，忙时人不够、闲时人过剩。",[34,1542,1543],{},"客流预测不准，备料和排班都跟着错。",[34,1545,1546],{},"大众点评、外卖平台的评价回复耗时。",[34,1548,1549],{},"朋友圈、社群营销文案写不出来。",[34,1551,1552],{},"会员复购靠拍脑袋，缺少数据支撑。",[20,1554,1556],{"id":1555},"ai能落地的场景","AI能落地的场景",[27,1558,1560],{"id":1559},"_1-客流预测与智能排班","1. 客流预测与智能排班",[31,1562,1563,1566,1569],{},[34,1564,1565],{},"基于历史客流、天气、节假日、周边活动预测各时段客流。",[34,1567,1568],{},"自动生成排班建议，匹配高峰人力需求。",[34,1570,1571],{},"连锁门店可统一调度。",[11,1573,1574,1575,1578],{},"这是餐饮AI",[15,1576,1577],{},"价值最直接","的场景——直接关联人力成本。",[27,1580,1582],{"id":1581},"_2-智能推荐与连带销售","2. 智能推荐与连带销售",[31,1584,1585,1588,1591],{},[34,1586,1587],{},"根据顾客点单、时段、季节推荐搭配。",[34,1589,1590],{},"收银\u002F小程序点餐页的智能推荐位。",[34,1592,1593],{},"提升客单价。",[27,1595,1597],{"id":1596},"_3-点评分析与自动回复","3. 点评分析与自动回复",[31,1599,1600,1603,1606],{},[34,1601,1602],{},"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。",[34,1604,1605],{},"辅助生成评价回复（人工把关后发送）。",[34,1607,1608],{},"发现差评共性问题，改进运营。",[27,1610,1612],{"id":1611},"_4-营销文案生成","4. 营销文案生成",[31,1614,1615,1618,1621],{},[34,1616,1617],{},"朋友圈、社群、公众号文案快速生成。",[34,1619,1620],{},"节日活动、新品上市的推广素材。",[34,1622,1623],{},"多门店内容统一又本地化。",[27,1625,1627],{"id":1626},"_5-会员与私域","5. 会员与私域",[31,1629,1630,1633,1636],{},[34,1631,1632],{},"会员消费数据分析，识别高价值客户和流失风险。",[34,1634,1635],{},"个性化优惠券和召回。",[34,1637,1638],{},"复购预测。",[20,1640,1641],{"id":1641},"落地的现实考量",[107,1643,1644,1659],{},[110,1645,1646],{},[113,1647,1648,1650,1653,1656],{},[116,1649,1040],{},[116,1651,1652],{},"数据依赖",[116,1654,1655],{},"难度",[116,1657,1658],{},"见效",[126,1660,1661,1673,1684,1697,1709],{},[113,1662,1663,1666,1668,1670],{},[131,1664,1665],{},"营销文案生成",[131,1667,147],{},[131,1669,147],{},[131,1671,1672],{},"快",[113,1674,1675,1678,1680,1682],{},[131,1676,1677],{},"点评分析回复",[131,1679,336],{},[131,1681,147],{},[131,1683,1672],{},[113,1685,1686,1689,1692,1694],{},[131,1687,1688],{},"客流预测排班",[131,1690,1691],{},"高（历史数据）",[131,1693,336],{},[131,1695,1696],{},"中期",[113,1698,1699,1702,1705,1707],{},[131,1700,1701],{},"智能推荐",[131,1703,1704],{},"中（点单数据）",[131,1706,336],{},[131,1708,1696],{},[113,1710,1711,1714,1717,1720],{},[131,1712,1713],{},"会员复购预测",[131,1715,1716],{},"高",[131,1718,1719],{},"中高",[131,1721,1722],{},"中长期",[11,1724,1725,1728],{},[15,1726,1727],{},"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）。",[20,1730,251],{"id":251},[31,1732,1733,1739,1745,1751,1757],{},[34,1734,1735,1738],{},[15,1736,1737],{},"数据质量差就上预测","：历史客流数据不准，预测也不准。",[34,1740,1741,1744],{},[15,1742,1743],{},"自动回复不经审核","：AI回复可能不当，引发公关问题。",[34,1746,1747,1750],{},[15,1748,1749],{},"忽视门店执行","：AI给排班建议，门店不执行等于零。",[34,1752,1753,1756],{},[15,1754,1755],{},"一上来搞大而全","：建议单点突破，验证后再扩展。",[34,1758,1759,1762],{},[15,1760,1761],{},"把AI当万能","：餐饮核心仍是产品和体验，AI是提效工具。",[20,1764,286],{"id":286},[107,1766,1767,1777],{},[110,1768,1769],{},[113,1770,1771,1773,1775],{},[116,1772,295],{},[116,1774,298],{},[116,1776,301],{},[126,1778,1779,1789,1799,1808],{},[113,1780,1781,1784,1787],{},[131,1782,1783],{},"文案\u002F点评辅助",[131,1785,1786],{},"SaaS 或 API，按量计费",[131,1788,147],{},[113,1790,1791,1794,1797],{},[131,1792,1793],{},"客流排班模块",[131,1795,1796],{},"和门店系统打通，定制",[131,1798,336],{},[113,1800,1801,1803,1806],{},[131,1802,1701],{},[131,1804,1805],{},"点餐\u002F收银集成",[131,1807,336],{},[113,1809,1810,1813,1816],{},[131,1811,1812],{},"连锁智能运营中台",[131,1814,1815],{},"多门店 + 预测 + 推荐 + 私域",[131,1817,1818],{},"高，定制",[20,1820,339],{"id":339},[341,1822,1823,1826,1829,1832,1835],{},[34,1824,1825],{},"找出最耗时的环节（排班？点评？文案？）。",[34,1827,1828],{},"从低成本场景试点（文案、点评辅助）。",[34,1830,1831],{},"积累门店数据，再做客流预测排班。",[34,1833,1834],{},"逐步扩展到推荐、会员。",[34,1836,1837],{},"始终让人把关关键输出。",[359,1839,1840],{},[11,1841,1842],{},"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":365,"searchDepth":366,"depth":366,"links":1844},[1845,1846,1853,1854,1855,1856],{"id":1535,"depth":366,"text":1535},{"id":1555,"depth":366,"text":1556,"children":1847},[1848,1849,1850,1851,1852],{"id":1559,"depth":371,"text":1560},{"id":1581,"depth":371,"text":1582},{"id":1596,"depth":371,"text":1597},{"id":1611,"depth":371,"text":1612},{"id":1626,"depth":371,"text":1627},{"id":1641,"depth":366,"text":1641},{"id":251,"depth":366,"text":251},{"id":286,"depth":366,"text":286},{"id":339,"depth":366,"text":339},"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。",[1860,1863,1866],{"q":1861,"a":1862},"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":1864,"a":1865},"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":1867,"a":1868},"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。",[1870,1871,1872,1873],"餐饮AI","餐饮排班AI","AI营销文案","餐饮数字化",{},"\u002Fblog\u002Fai-agent\u002Fai-canyin",{"title":1523,"description":1858},{"loc":1875},"blog\u002Fai-agent\u002Fai-canyin",[1880,1881,1882],"餐饮","营销","排班","wwPJmcDR-PabsSXv_ZcpsIAe0cgabBUgBGYdFdSCdPc",1781688906894]