[{"data":1,"prerenderedAt":1687},["ShallowReactive",2],{"site-content":3,"content-source":1685},{"common":4,"brand":31,"navigation":35,"footerLinks":54,"links":79,"home":99,"serviceCards":152,"publicServiceCards":182,"cases":199,"testimonials":567,"faqs":589,"about":614,"productsPage":672,"team":695,"devHub":731,"investorRelations":777,"demosPage":805,"publicServicesPage":833,"aiHelp":848,"linksPage":857,"productCards":865,"demoConcepts":947,"productDetails":957,"demos":1208,"detailPages":1526},{"skipToContent":5,"footerTitle":6,"deliveryFocus":7,"detailLabels":8,"demoLabels":13,"conversionCta":17,"theme":28},"跳到主要内容","友情链接","交付聚焦",{"overview":9,"features":10,"scenarios":11,"advantages":12},"服务概述","主要特性","应用场景","技术优势",{"stageTitle":14,"openOriginal":15,"missing":16},"阶段说明","在新窗口打开原始演示","未找到该演示资源。",{"eyebrow":18,"stages":19,"primaryLabel":23,"secondaryCasesLabel":24,"secondaryProductsLabel":25,"secondaryDemosLabel":26,"tertiaryLabel":27},"Next Step",[20,21,22],"需求沟通与场景梳理","方案确认与路径选择","进入开发或交付阶段","发起咨询","查看案例","查看产品","查看演示","邮件直连",{"light":29,"dark":30},"亮色","暗色",{"name":32,"tagline":33,"logo":34},"开发喵 AI","AI 服务与区块链解决方案","https:\u002F\u002Fcdn.sa.net\u002F2024\u002F07\u002F17\u002FyPZoEwn6uWFxihJ.jpg",[36,39,42,45,48,51],{"label":37,"to":38},"首页","\u002F",{"label":40,"to":41},"产品","\u002Fproducts",{"label":43,"to":44},"演示","\u002Faidemo",{"label":46,"to":47},"团队","\u002Fteam",{"label":49,"to":50},"开发者","\u002Fdev-hub",{"label":52,"to":53},"关于","\u002Fabout",[55,59,62,65,68,71,74,77],{"label":56,"to":57,"external":58},"OpenLens","https:\u002F\u002Fopenlens.pro",true,{"label":60,"to":61,"external":58},"HandyHub","https:\u002F\u002Fhandyhub.vip",{"label":63,"to":64,"external":58},"KFMPing","https:\u002F\u002Fkfmping.cc",{"label":66,"to":67,"external":58},"云创动力","https:\u002F\u002Fycdledu.com",{"label":69,"to":70,"external":58},"六行.神算","https:\u002F\u002Fwww.openai-ss.cn\u002F",{"label":72,"to":73},"AI喵帮助","\u002Fai-help",{"label":75,"to":76},"公共服务平台","\u002Fpublic-services",{"label":78,"to":50},"开发者中心",[80,84,89,94],{"title":81,"description":82,"icon":83,"to":38},"返回主页","回到站点总入口，查看服务能力与最新内容。","Home",{"title":85,"description":86,"icon":87,"to":88,"external":58},"GitHub","查看团队主页与开源协作入口。","Github","https:\u002F\u002Fgithub.com\u002Fkaifamiao",{"title":90,"description":91,"icon":92,"to":93},"AI 服务","API、大模型、知识库、RAG、Agent 等模块入口。","Sparkles","\u002F#services",{"title":95,"description":96,"icon":97,"to":98},"区块链服务","智能合约与 DApp 等说明见专题页。","Blocks","\u002Fblockchain",{"headTitle":33,"hero":100,"services":119,"publicServices":134,"cases":139,"faq":144,"feedback":149},{"eyebrow":101,"title":102,"description":103,"primary":104,"secondary":105,"stackLabel":107,"stackItems":108,"stats":112},"AI That Everyone Can Use","让 AI 融入业务、产品与日常工作。","开发喵 AI 聚焦 API、多模型、知识库、RAG、Agent 与区块链解决方案，帮助团队更快完成从验证到上线。",{"label":25,"to":41},{"label":106,"to":44},"体验原理演示","Capability Stack",[109,110,111],"多模型 API 与路由策略","知识库、RAG 与 Agent 工程","公共服务平台与配套文档",[113,116],{"value":114,"label":115},"6+","核心服务模块",{"value":117,"label":118},"4","公共服务能力",{"title":120,"description":121,"metrics":122},"服务能力","按模块浏览，进入详情页查看交付边界与应用场景。",[123,126,129,132],{"value":124,"label":125},"100+","服务企业",{"value":127,"label":128},"10000+","个人用户",{"value":130,"label":131},"5","覆盖行业",{"value":114,"label":133},"核心服务",{"eyebrow":135,"title":75,"description":136,"cta":137},"Public Services","统一整理开通流程、Key 使用方式与服务明细，适合研发团队集中接入。",{"label":138,"to":76},"查看开通流程",{"title":140,"description":141,"resultLabel":142,"listLabel":143},"客户案例","横向滑动查看；数字为项目摘要，非营销口号堆砌。","成果摘要","查看全部案例",{"title":145,"description":146,"moreLabel":147,"pageTitle":148},"常见问题","关于接入、交付、周期与合作方式的高频问题。","更多问题","全部常见问题",{"title":150,"description":151},"客户反馈","用行业、项目类型和交付结果去说明合作价值，而不是只放几句匿名评价。",[153,159,164,169,174,179],{"title":154,"description":155,"icon":156,"to":157,"cta":158},"AI API 服务","稳定的多模型调用与接入方案，便于现有系统快速获得推理与生成能力。","Zap","\u002Fai-api","了解详情",{"title":160,"description":161,"icon":162,"to":163,"cta":158},"大模型开发与训练","架构设计、数据与训练管线、微调与评估，覆盖从实验到上线的工程环节。","Cpu","\u002Fllm-development",{"title":165,"description":166,"icon":167,"to":168,"cta":158},"AI 知识库搭建","文档治理、索引与权限，把分散资料变成可检索、可审计的知识资产。","BookOpen","\u002Fai-knowledge-base",{"title":170,"description":171,"icon":172,"to":173,"cta":158},"RAG 应用开发","检索增强生成，更可控的问答与内容生成，贴合业务语料与合规要求。","FileSearch","\u002Frag-app",{"title":175,"description":176,"icon":177,"to":178,"cta":158},"Agent 应用开发","为复杂流程设计可协作、可执行、可观测的自主式 AI Agent 系统。","Bot","\u002Fagent-app",{"title":180,"description":181,"icon":97,"to":98,"cta":158},"区块链开发","智能合约、链上交互、钱包接入与 DApp 方案，覆盖链上产品交付。",[183,187,191,195],{"title":184,"description":185,"icon":186},"开发喵短信服务","用于验证码、通知与营销消息触达，支持回执和状态查询。","MessageSquareHeart",{"title":188,"description":189,"icon":190},"大模型接入服务","统一路由多模型能力，支持配额管理、日志与调用观测。","BrainCircuit",{"title":192,"description":193,"icon":194},"科学网络转化服务","提升跨区域网络连接质量，满足研发、抓取和数据传输场景。","Network",{"title":196,"description":197,"icon":198},"IP 质量检测服务","检测 IP 可用性、风险状态与质量评分，辅助前置筛选。","ShieldCheck",[200,240,277,314,351,388,425,460,496,532],{"id":201,"index":201,"slug":202,"title":203,"company":204,"description":205,"outcomes":206,"category":211,"industry":212,"duration":213,"metrics":214,"summary":223,"challenge":224,"solution":228,"architecture":232,"results":236},1,"enterprise-customer-support","企业智能客服系统","某大型科技公司","基于大模型的客服系统：分流、话术与质检一体化，缩短响应链路。",[207,208,209,210],"响应时间显著缩短","满意度与解决率提升","人工坐席负载下降","7×24 自动覆盖","enterprise","企业服务 \u002F 客服运营","8 周",[215,217,220],{"label":216,"value":213},"交付周期",{"label":218,"value":219},"服务覆盖","7×24",{"label":221,"value":222},"核心结果","响应链路缩短","围绕海量咨询入口、复杂业务口径和服务质检压力，建设一套结合大模型问答、知识检索和人工兜底的客服系统。",[225,226,227],"客服入口分散，用户问题类型多，人工分流效率低。","业务知识更新频繁，旧知识库维护成本高，答案一致性差。","客服质检依赖人工抽检，覆盖率低，无法形成稳定优化闭环。",[229,230,231],"搭建统一的意图识别与分流入口，把高频问题优先交给 AI 处理。","把产品文档、流程说明和常见问题纳入知识库，并通过 RAG 提升回答可控性。","增加质检规则与会话摘要能力，让运营团队快速定位异常对话。",[233,234,235],"多渠道接入层","知识库与检索增强问答","人工转接与质检分析模块",[237,238,239],"高频重复咨询由 AI 稳定承接，人工坐席只处理复杂个案。","客服响应链路明显缩短，业务口径更统一。","质检从抽样走向高覆盖，后续培训和优化更有依据。",{"id":241,"index":241,"slug":242,"title":243,"company":244,"description":245,"outcomes":246,"category":251,"industry":252,"duration":253,"metrics":254,"summary":260,"challenge":261,"solution":265,"architecture":269,"results":273},2,"intelligent-knowledge-platform","智能知识库平台","知名教育机构","教学资源入库、版本与权限管理，支持教研场景下的精准检索。",[247,248,249,250],"检索效率提升","多格式文档统一治理","问答与引用可追溯","使用率持续增长","knowledge","教育 \u002F 教研管理","6 周",[255,256,259],{"label":216,"value":253},{"label":257,"value":258},"内容类型","多格式文档",{"label":221,"value":247},"为教研团队构建统一知识底座，解决教学资料分散、版本混乱和跨角色检索效率低的问题。",[262,263,264],"历史资料分散在网盘、群文件和本地目录，统一治理难度高。","教师、教研、管理人员对资料的权限要求不同。","传统关键词搜索难以满足复杂问题查询。",[266,267,268],"对教案、课件、制度和培训材料建立统一入库规范。","引入文档切片、向量检索和权限过滤机制，保证可查也可控。","问答结果附带引用片段与来源，提升教研使用信任度。",[270,271,272],"文档治理与解析","向量索引与权限控制","问答与引用展示层",[274,275,276],"资料检索入口统一，教研团队不再反复问人找文件。","权限边界清晰，敏感内容不会误暴露给无关角色。","问答引用可回溯，知识使用更规范。",{"id":278,"index":278,"slug":279,"title":280,"company":281,"description":282,"outcomes":283,"category":288,"industry":289,"duration":290,"metrics":291,"summary":297,"challenge":298,"solution":302,"architecture":306,"results":310},3,"ai-agent-automation","AI Agent 自动化","互联网企业","运营与内部流程的 Agent 编排，关键步骤保留人工审核节点。",[284,285,286,287],"重复劳动自动化","人为差错减少","复杂分支可配置","ROI 可量化复盘","automation","互联网 \u002F 运营自动化","10 周",[292,293,296],{"label":216,"value":290},{"label":294,"value":295},"流程形态","多步骤编排",{"label":221,"value":284},"为运营和内部协作流程建设 Agent 执行闭环，让系统能拆任务、调工具、回读结果并在关键节点交给人工确认。",[299,300,301],"内部流程跨系统、跨角色，人工执行步骤多且容易漏。","一次性脚本难以应对复杂分支和异常状态。","完全自动化风险高，需要保留人工审核能力。",[303,304,305],"将任务拆成若干明确节点，由 Agent 逐步执行并记录状态。","为不同任务接入查询、通知、表单和内部数据工具。","在人审节点引入回退和继续机制，避免误操作。",[307,308,309],"任务编排引擎","工具调用层","状态持久化与人审流程",[311,312,313],"重复性运营流程大幅减少人工介入。","异常路径可观测、可追溯，便于复盘与优化。","业务团队更愿意接受 Agent 协助，而不是担心黑盒执行。",{"id":315,"index":315,"slug":316,"title":317,"company":318,"description":319,"outcomes":320,"category":251,"industry":325,"duration":326,"metrics":327,"summary":334,"challenge":335,"solution":339,"architecture":343,"results":347},4,"enterprise-rag-qa","企业 RAG 问答","金融机构","内规、研报与制度库接入 RAG，回答带出处，满足合规查阅习惯。",[321,322,323,324],"答案准确率提升","延迟与吞吐优化","多语言资料支持","知识覆盖扩展","金融 \u002F 合规与研究","7 周",[328,329,332],{"label":216,"value":326},{"label":330,"value":331},"关键能力","带出处回答",{"label":221,"value":333},"准确率提升","把分散的制度、内规和研究资料连接到可追溯问答系统，重点解决回答可信度和出处要求。",[336,337,338],"用户需要的不只是答案，还要能看到来源与依据。","文档体量大且更新快，纯人工维护索引困难。","对准确性和可控性的要求高于一般内容问答场景。",[340,341,342],"设计适配金融资料的切片和召回策略，减少无关片段噪音。","在回答中显示引用来源、标题与片段，满足合规查阅习惯。","配合重排和上下文治理，稳定高价值资料的命中率。",[344,345,346],"文档解析与向量检索","重排与上下文拼接","带出处的回答生成",[348,349,350],"业务用户能更快找到制度依据和研究结论。","答案可解释性明显增强，内部使用阻力降低。","多类资料被统一纳入可维护的知识问答体系。",{"id":352,"index":352,"slug":353,"title":354,"company":355,"description":356,"outcomes":357,"category":362,"industry":363,"duration":364,"metrics":365,"summary":371,"challenge":372,"solution":376,"architecture":380,"results":384},5,"smart-contract-platform","智能合约平台","区块链公司","合约模板、审计协作与多环境部署，面向业务方的可读抽象层。",[358,359,360,361],"执行路径优化","Gas 与成本评估","安全基线对齐","多链策略支持","platform","区块链 \u002F 链上产品","9 周",[366,367,370],{"label":216,"value":364},{"label":368,"value":369},"部署范围","多环境 \u002F 多链",{"label":221,"value":360},"围绕合约模板管理、部署流程和审计协作，降低业务团队接触底层链上实现的复杂度。",[373,374,375],"业务方理解合约逻辑成本高，需求沟通效率低。","多环境部署和多链兼容带来版本管理压力。","安全和审计要求使上线流程更复杂。",[377,378,379],"抽象常用合约能力，形成模板化交付方式。","建立测试、部署和审计协作流程，减少线上风险。","为业务方提供更可读的配置和状态展示。",[381,382,383],"合约模板层","测试与部署流水线","审计协作与监控面板",[385,386,387],"业务方对链上逻辑的理解成本明显下降。","多环境与多链部署流程更规范。","上线前风险暴露更早，协作效率更高。",{"id":389,"index":389,"slug":390,"title":391,"company":392,"description":393,"outcomes":394,"category":399,"industry":400,"duration":401,"metrics":402,"summary":408,"challenge":409,"solution":413,"architecture":417,"results":421},6,"custom-model-training","大模型定制训练","AI 创新公司","领域数据清洗、训练预算与评测指标对齐，避免“训完不可用”。",[395,396,397,398],"任务指标提升","训练周期压缩","算力成本可控","持续迭代机制","training","AI \u002F 行业模型","12 周",[403,404,407],{"label":216,"value":401},{"label":405,"value":406},"工作重点","训练与评测闭环",{"label":221,"value":395},"围绕领域任务构建训练数据、评估标准和推理上线方案，让训练结果真正服务业务指标。",[410,411,412],"业务目标和训练指标之间存在脱节。","领域数据质量参差不齐，直接训练效果不稳定。","算力预算有限，需要在性能和成本间做平衡。",[414,415,416],"从业务任务反推训练目标，明确数据与评测口径。","清洗指令数据并建立分阶段评估机制。","结合推理优化方案，避免训练成果难以上线。",[418,419,420],"数据治理与标注","训练与评测流水线","推理优化与上线方案",[422,423,424],"模型在目标任务上的有效指标得到提升。","训练和评测形成可复用流程，后续迭代更稳。","算力投入和产出关系更清晰。",{"id":426,"index":426,"slug":427,"title":428,"company":429,"description":430,"outcomes":431,"category":362,"industry":436,"duration":213,"metrics":437,"summary":443,"challenge":444,"solution":448,"architecture":452,"results":456},7,"ai-api-platform","AI API 服务平台","SaaS 服务商","统一网关、配额与计费观测，多模型路由对下游产品透明。",[432,433,434,435],"调用量增长","P99 延迟下降","多模型统一接入","高可用架构","SaaS \u002F 平台能力",[438,439,442],{"label":216,"value":213},{"label":440,"value":441},"核心能力","统一模型网关",{"label":221,"value":433},"为多产品线统一建设 AI API 平台，解决供应商切换、配额管理和计费观测问题。",[445,446,447],"多个产品分别对接模型，维护成本高。","不同模型供应商接口差异大，切换困难。","调用观测和成本控制缺乏统一平台。",[449,450,451],"建立统一 API 网关和模型路由层。","对接日志、配额、计费与监控，向下游提供稳定接口。","通过策略路由提升高峰期可用性和响应表现。",[453,454,455],"统一鉴权网关","模型路由策略层","日志计费与监控看板",[457,458,459],"新业务接入速度明显提升。","模型切换与扩展不再影响下游产品接口。","成本、延迟和调用量都能统一观测。",{"id":461,"index":461,"slug":462,"title":463,"company":464,"description":465,"outcomes":466,"category":251,"industry":471,"duration":290,"metrics":472,"summary":479,"challenge":480,"solution":484,"architecture":488,"results":492},8,"university-knowledge-base-system","某高校知识库系统","某高校","围绕校内制度、科研资料与教学文档建设统一知识库系统，支持权限检索与来源追溯。",[467,468,469,470],"校内资料统一入库","师生检索效率提升","知识权限分层管理","答案引用可追溯","高校 \u002F 信息化建设",[473,474,477],{"label":216,"value":290},{"label":475,"value":476},"服务对象","师生与管理部门",{"label":221,"value":478},"资料统一入库","围绕高校制度文件、科研资料和教学文档，建设面向多角色的统一知识库系统。",[481,482,483],"校内资料类型多，来源分散，历史沉淀缺乏统一入口。","教师、学生和管理部门对资料访问权限差异明显。","传统搜索只能找文件名，难以回答复杂问题。",[485,486,487],"统一接入制度、科研、教学三类核心资料，并做分类治理。","按角色与部门划分权限，检索结果自动过滤无权限内容。","通过知识问答与来源引用，让师生能直接定位到相关原文。",[489,490,491],"多源资料接入与治理","权限检索与向量索引","问答与引用展示",[493,494,495],"校内知识资产不再分散在多个系统和个人目录中。","教师和学生可以更快找到制度依据与教学资料。","问答结果可追溯到来源，信息可信度更高。",{"id":497,"index":497,"slug":498,"title":499,"company":500,"description":501,"outcomes":502,"category":362,"industry":506,"duration":507,"metrics":508,"summary":515,"challenge":516,"solution":520,"architecture":524,"results":528},9,"listed-company-multi-model-platform","某上市公司大模型聚合平台","某上市公司","整合多家模型供应商能力，统一提供调用路由、监控、配额与业务接入界面。",[434,503,504,505],"调用成本更可控","平台监控能力增强","业务团队接入更快","企业平台 \u002F AI 中台","11 周",[509,510,513],{"label":216,"value":507},{"label":511,"value":512},"平台角色","企业 AI 中台",{"label":221,"value":514},"接入速度提升","建设面向内部业务团队的大模型聚合平台，让模型供应商能力统一收敛在一个平台层中。",[517,518,519],"不同业务团队重复对接模型，成本和风险都高。","模型供应商能力和计费方式不统一，平台治理难度大。","缺少统一监控与策略配置，无法快速做供应商切换。",[521,522,523],"抽象统一调用协议，把底层多家模型供应商做成平台能力。","建设统一监控、配额、日志和成本分析能力。","通过路由策略与配置管理提升平台的灵活性和可控性。",[525,526,527],"统一调用网关","模型供应商适配层","监控配额与成本分析",[529,530,531],"业务团队接入模型的门槛和时间显著下降。","平台具备成本观察和模型切换能力。","多模型能力开始从“分散接入”变成“平台资产”。",{"id":533,"index":533,"slug":534,"title":535,"company":536,"description":537,"outcomes":538,"category":288,"industry":543,"duration":364,"metrics":544,"summary":550,"challenge":551,"solution":555,"architecture":559,"results":563},10,"discipline-research-report-system","某纪律组织调研报告系统","某纪律组织","面向调研与报告撰写场景，打通资料沉淀、检索问答、报告生成与版本管理流程。",[539,540,541,542],"调研资料集中治理","报告草拟效率提升","版本流转更规范","过程留痕可审计","政务 \u002F 调研与报告",[545,546,548],{"label":216,"value":364},{"label":11,"value":547},"调研与报告流转",{"label":221,"value":549},"初稿效率提升","围绕调研资料沉淀、报告草拟和版本流转建设一套可留痕、可审计的报告系统。",[552,553,554],"调研资料积累分散，后续复用效率低。","报告撰写依赖人工整理资料，初稿效率受限。","版本流转和修改记录不够规范，追踪成本高。",[556,557,558],"统一管理调研资料、历史报告和参考素材。","引入知识问答与报告生成辅助，提升初稿整理效率。","把版本管理和流转记录纳入系统，满足审计与留痕要求。",[560,561,562],"资料治理与知识检索","报告生成辅助","版本流转与审计记录",[564,565,566],"调研资料和历史成果能被持续复用。","报告初稿生成效率明显提升。","版本流转更加规范，可审计性更强。",[568,575,582],{"industry":569,"projectType":570,"result":571,"quote":572,"author":573,"meta":574},"上市公司","大模型聚合平台","多模型统一接入，业务团队接入效率提升","需求理解很快，能把技术方案翻译成业务团队能直接判断的交付路径。","某上市公司产品负责人","平台建设项目",{"industry":576,"projectType":577,"result":578,"quote":579,"author":580,"meta":581},"高校","知识库系统","资料检索效率提升，引用来源更清晰","知识库系统上线后，老师和学生检索资料的效率明显提升，引用来源也更清晰。","某高校信息化负责人","校内知识治理项目",{"industry":583,"projectType":584,"result":585,"quote":586,"author":587,"meta":588},"纪律组织","调研报告系统","报告草拟效率提升，流程留痕更规范","项目推进节奏稳，文档和沟通都比较规范，后续扩展也预留了空间。","某纪律组织项目负责人","调研与报告项目",[590,593,596,599,602,605,608,611],{"q":591,"a":592},"你们支持哪些交付方式？","支持咨询+开发、联合研发、按模块交付，以及长期运维协作。",{"q":594,"a":595},"从需求到上线一般要多久？","PoC 通常 1-3 周，正式版本视范围通常 4-12 周可上线首版。",{"q":597,"a":598},"可以私有化部署吗？","可以，支持云端与本地私有化部署，并可对接企业 IAM 与审计体系。",{"q":600,"a":601},"如何申请公共服务 Key？","提交需求后完成审核即可发放 Key，详情可查看",{"q":603,"a":604},"你们安全可靠吗？","我们长期深耕 AI 领域，已服务多类行业客户，重视稳定交付与长期合作。公司合规经营，严格按照业务规范与法律要求开展服务，确保合作可持续、可追溯。",{"q":606,"a":607},"你们在海外有业务吗？","有。我们支持海外客户与跨区域项目协作，具备多时区沟通与全球化交付经验，可按场景提供云端、网络与模型接入方案。",{"q":609,"a":610},"你们公司在哪里？","公司位于天府之国（成都），并支持远程协作与跨城市项目交付。",{"q":612,"a":613},"你们主要的客户是谁？","我们主要服务政府、央企\u002F国企及事业单位，同时也支持具备明确业务目标的企业客户，强调安全合规、稳定交付与长期维护。",{"headTitle":615,"hero":616,"missionTitle":623,"mission":624,"whyTitle":627,"whyDescription":628,"whyNeeds":629,"capabilitiesTitle":649,"capabilities":650,"visionTitle":668,"vision":669},"关于我们",{"eyebrow":617,"title":618,"description":619,"primary":620,"secondary":621},"About","关于开发喵 AI","我们希望让 AI 融入生活，让更多人不受限制地使用 AI，把复杂能力做成真正可用的产品与服务。",{"label":25,"to":41},{"label":622,"to":50},"联系团队","使命",[625,626],"让 AI 融入生活，让更多人不受限制地使用 AI。","“AI that everyone can use” 是我们的口号，也是站点和产品设计的核心方向。","为什么需要我们？","很多团队不是不知道 AI 有价值，而是不确定怎么在合规、安全、成本和本地化限制下真正落地。我们的价值，在于把这些限制条件变成可交付方案。",[630,633,637,641,645],{"title":631,"description":632,"icon":198},"合规需求","很多行业在接入 AI 时首先遇到的不是模型效果，而是合规边界。我们会结合业务场景梳理数据流向、权限边界、供应商选择和落地方式，帮助团队在合规框架内推进，而不是停留在“想做但不敢做”的阶段。",{"title":634,"description":635,"icon":636},"本地化需求","不同团队面向的语言、业务习惯、部署环境和组织流程都不一样。我们不是简单复制通用产品，而是把模型能力、本地业务流程和已有系统接起来，让方案真正适配你的团队，而不是要求团队反过来适应工具。","MapPinned",{"title":638,"description":639,"icon":640},"安全审计需求","只让模型“自觉遵守规则”是不够的，真正可上线的系统必须考虑日志留痕、权限控制、调用审计和风险隔离。我们更重视工程化安全边界，帮助客户建立可追踪、可解释、可审查的 AI 使用链路。","LockKeyhole",{"title":642,"description":643,"icon":644},"\"翻译\"需求","很多跨语言场景不是把内容翻出来就结束了，还涉及术语统一、语气控制、上下文保持和多语言协作效率。我们可以把翻译能力和知识库、工作流、审核机制结合起来，做成真正服务业务的语言系统。","Languages",{"title":646,"description":647,"icon":648},"成本需求","模型越强不代表越适合所有请求。我们会从路由策略、调用频率、任务分层、缓存和交付形态上一起设计，帮助客户把成本花在真正有价值的地方，而不是用最高配置去处理所有问题。","BadgeDollarSign","我们在做什么",[651,655,658,661,664],{"title":652,"description":653,"icon":654},"AI 教学","培训与科普，帮助用户理解与运用 AI。","GraduationCap",{"title":656,"description":657,"icon":92},"AIGC 与 GPT","整合生成式能力，提供更智能的支持。",{"title":659,"description":660,"icon":156},"企业级 AI API","定制化 API，满足企业与产品集成需求。",{"title":662,"description":663,"icon":167},"知识库搭建","构建可检索、可维护的知识体系。",{"title":665,"description":666,"icon":667},"AI 软件研发","定制开发与交付 AI 相关软件。","Code2","愿景",[670,671],"无论你要提效、解决问题还是创造新价值，开发喵 AI 都希望成为可靠助手。","我们努力降低技术门槛，让 AI 服务于生活与工作，并与合作伙伴建立长期可迭代的关系。",{"headTitle":673,"hero":674,"categoriesTitle":681,"categories":682},"我们的产品",{"eyebrow":675,"title":673,"description":676,"primary":677,"secondary":679},"Products","从 AI 应用开发到公共服务平台、知识库、训练和聚合能力，围绕实际交付组织产品矩阵。",{"label":678,"to":50},"联系合作",{"label":680,"to":38},"返回首页","产品分类",{"enterprise":683,"personal":686,"self":689,"global":692},{"title":684,"description":685},"企业服务","面向企业系统、交付项目和业务协同场景的服务能力。",{"title":687,"description":688},"个人服务","围绕账号、设备、数字内容和个人使用场景提供的服务入口。",{"title":690,"description":691},"自研产品","团队自主打磨的产品、平台和工程模板。",{"title":693,"description":694},"全球用户","面向跨区域访问、全球模型接入与国际化业务需求的产品能力。",{"headTitle":696,"hero":697,"introTitle":696,"intro":704,"valuesTitle":707,"values":708,"cooperationTitle":725,"cooperation":726},"团队介绍",{"eyebrow":698,"title":699,"description":700,"primary":701,"secondary":703},"Team","开发喵团队","围绕 AI 产品、开发工具链与业务落地协作，强调长期主义与工程质量。",{"label":702,"to":50},"访问开发者中心",{"label":25,"to":41},[705,706],"团队由一群长期关注 AI 产品、工程效率与交付质量的成员组成。","支持远程协作，重视主动沟通、明确边界与责任交付。","价值观",[709,713,717,721],{"title":710,"description":711,"icon":712},"主动学习","持续关注模型、工具链与交付方式的变化。","Lightbulb",{"title":714,"description":715,"icon":716},"交付负责","重视需求澄清、过程沟通与结果稳定性。","BadgeCheck",{"title":718,"description":719,"icon":720},"长期合作","倾向与客户建立长期、可迭代的协作关系。","Handshake",{"title":722,"description":723,"icon":724},"用户价值","把复杂技术转成真正可用的产品体验。","Heart","能力与合作方式",[727,728,729,730],"定制化开发与交付","AI 接入咨询与方案设计","知识库、RAG、Agent 模板沉淀","持续维护与版本迭代",{"headTitle":78,"hero":732,"collaborationTitle":741,"collaboration":742,"roadmapTitle":765,"roadmap":766,"joinTitle":771,"joinText":772,"githubCta":773,"sponsorTitle":774,"sponsorText":775,"sponsorCta":776},{"eyebrow":733,"title":78,"description":734,"primary":735,"secondary":738},"Developer Hub","面向 API 接入、功能共建、定制开发和技术合作，整理协作方式、计划与需求提交入口。",{"label":736,"to":737,"external":58},"邮件提交需求","mailto:hello@kaifamiao.com?subject=%E5%BC%80%E5%8F%91%E9%9C%80%E6%B1%82%E6%8F%90%E4%BA%A4",{"label":739,"to":740},"查看联系方式","\u002Flink","协作方式",[743,747,751,755,758,761],{"title":744,"description":745,"icon":746},"开放协作","欢迎面向产品、工具链、内容和基础设施共同建设。","Globe2",{"title":748,"description":749,"icon":750},"异步优先","支持远程协作，强调信息留痕与明确边界。","Clock3",{"title":752,"description":753,"icon":754},"模块化沉淀","偏好可复用组件、模板与基础设施积累。","Layers3",{"title":714,"description":756,"icon":757},"重视版本、质量、验收和后续维护。","CheckCheck",{"title":759,"description":760,"icon":198},"安全意识","涉及模型、数据和权限时强调审计与合规。",{"title":762,"description":763,"icon":764},"投资者","欢迎关注 AI 产品、基础设施与长期能力建设的投资者沟通交流，我们更看重长期主义、清晰路径和可持续增长。","LineChart","正在开发的计划",[767,768,769,770],"公共服务平台能力扩展与开放文档完善","多模型路由策略优化与成本观测看板","RAG \u002F Agent 工程模板的标准化组件沉淀","演示体系与教学资料持续迭代","加入我们","我们欢迎对 AI 产品、开发工具链、工程效率有热情的伙伴加入，重视主动沟通与责任交付。","访问 GitHub 团队主页","请喝咖啡","如果你认可我们的方向，愿意请我们喝杯咖啡，支持团队持续打磨产品、内容和基础设施，这里会保留一个直接支持入口。二维码放上来后，可以直接扫码支持。","也欢迎投资合作",{"headTitle":778,"hero":779,"narrativeTitle":649,"narrative":787,"focusTitle":790,"focus":791,"whyTitle":798,"why":799,"contactTitle":803,"contactText":804},"投资者关系",{"eyebrow":780,"title":778,"description":781,"primary":782,"secondary":785},"Investor Relations","面向关注 AI 产品、知识系统、基础设施与长期能力建设的投资者，集中说明我们的方向、阶段、合作方式与沟通入口。",{"label":783,"to":784,"external":58},"邮件联系","mailto:hello@kaifamiao.com?subject=%E6%8A%95%E8%B5%84%E5%90%88%E4%BD%9C%E6%B2%9F%E9%80%9A",{"label":786,"to":50},"返回开发者中心",[788,789],"我们围绕 AI 应用交付、知识库、模型接入、原理演示和公共服务能力，持续沉淀可复用产品与工程资产。","重点不是做短期概念包装，而是做能上线、能维护、能扩展、能形成长期积累的 AI 产品体系。","当前关注方向",[792,793,794,795,796,797],"企业级 AI 应用交付","知识库与 RAG 产品化","多模型接入与路由层","教学演示与认知型产品","公共服务平台能力沉淀","长期可维护的 AI 工程模板","为什么值得沟通",[800,801,802],"方向围绕真实交付与长期产品资产，而不是短期热点","产品、演示、知识系统和服务能力之间能够互相支撑","适合长期主义、偏工程和基础能力建设的合作逻辑","沟通方式","如果你关注 AI 产品、基础设施、知识系统和可持续增长，可以直接通过邮件联系。我们更希望以务实、透明和长期视角展开交流。",{"headTitle":806,"hero":807,"categoriesTitle":814,"categoriesDescription":815,"categories":816,"conceptsTitle":818,"conceptsDescription":829,"openLabel":830,"nextTitle":831,"nextDescription":832},"AI原理演示",{"eyebrow":808,"title":806,"description":809,"primary":810,"secondary":813},"Demos","把大模型、RAG 与向量空间的核心处理链路拆解成更容易观察和理解的可视化页面。",{"label":811,"to":812},"打开 LLM Demo","\u002Fdemo\u002Fllm",{"label":673,"to":41},"学习路径","按主题理解这些演示，会比按时间顺序逐个点开更容易形成整体认识。",{"fundamentals":817,"retrieval":820,"orchestration":823,"platform":826},{"title":818,"description":819},"基础概念","先建立对 token、上下文、提示词和采样的直觉，再看工程演示会更顺。",{"title":821,"description":822},"RAG \u002F 检索","理解向量、召回、分块、重排和阈值，才能看懂知识库为什么有时答得准、有时会偏。",{"title":824,"description":825},"Agent \u002F 编排","关注技能路由、工具调用、结构化输出和安全边界，理解模型如何接近真实系统。",{"title":827,"description":828},"工程平台","看平台如何处理模型路由、资源分配和质量成本平衡。","先理解这些概念，再去看动画，会更容易把模型处理过程和工程限制对应起来。","打开演示","推荐下一步","继续沿着相关主题看下去，会更容易把概念和工程实现串起来。",{"headTitle":75,"hero":834,"stepsTitle":841,"steps":842,"servicesTitle":847},{"eyebrow":835,"title":75,"description":836,"primary":837,"secondary":839},"Key Access","统一整理短信、大模型、网络与 IP 检测服务，便于研发团队集中查看开通流程、Key 使用方式与服务明细。",{"label":838,"to":157},"返回 AI API 服务",{"label":840,"to":41},"查看全部产品","开通流程",[843,844,845,846],"提交需求与业务场景","获取评估与接入建议","开通 Key 与权限","对接文档并进入联调","四项公共服务",{"headTitle":849,"hero":850},"AI 喵帮助",{"eyebrow":851,"title":849,"description":852,"primary":853,"secondary":856},"Help Center","文档、常见问题、接入说明与使用指引已统一整理在帮助中心，建议优先从帮助中心入口查看最新内容。",{"label":854,"to":855,"external":58},"进入帮助中心","https:\u002F\u002Fhelp.1155665ai.com",{"label":680,"to":38},{"headTitle":858,"hero":859,"openLabel":864},"相关链接",{"eyebrow":860,"title":858,"description":861,"primary":862,"secondary":863},"Links","团队主页、服务入口与联系方式集中在这里，方便快速跳转。",{"label":680,"to":38},{"label":78,"to":50},"打开",[866,872,878,884,889,894,900,905,911,918,922,926,932,937,942],{"badge":867,"category":211,"title":868,"description":869,"icon":667,"to":870,"cta":871},"01","AI 应用软件开发","面向业务流程打造可交付 AI 应用，覆盖前端、后端与模型集成。","\u002Fproducts\u002Fai-application-development","查看详情",{"badge":873,"category":874,"title":875,"description":876,"icon":167,"to":877,"cta":871},"02","self","everyone-llm（知识库）","沉淀文档检索、问答和企业知识管理能力。","\u002Fproducts\u002Feveryone-llm",{"badge":879,"category":880,"title":881,"description":882,"icon":883},"03","global","海外家宽","提供稳定跨区域网络能力，支持业务拓展与测试场景。","Globe",{"badge":885,"category":880,"title":886,"description":887,"icon":156,"to":888,"cta":871},"04","全球大模型 API 🔥","统一多模型接入、鉴权与配额管理。","\u002Fproducts\u002Fglobal-llm-api",{"badge":890,"category":211,"title":891,"description":892,"icon":162,"to":893,"cta":871},"05","大模型训练","从数据、训练到评测与上线的全流程支持。","\u002Fproducts\u002Fllm-training",{"badge":895,"category":211,"title":896,"description":897,"icon":898,"to":899,"cta":871},"06","AI 产品聚合服务","聚合文本、图像、语音、视频等多类 AI 能力。","Layers","\u002Fproducts\u002Fai-product-aggregation",{"badge":901,"category":211,"title":902,"description":903,"icon":904},"07","自研 DeepSeek 本地服务器","提供面向企业内部部署、推理调用和模型运行环境管理的 DeepSeek 服务器支持。","ServerCog",{"badge":906,"category":874,"title":907,"description":908,"icon":909,"to":910,"cta":871},"08","AI 多模型决策平台 OpenLens","多模型对话决策系统，支持比较、对话与策略辅助。","ScanEye","\u002Fproducts\u002Fopenlens",{"badge":912,"category":913,"title":914,"description":915,"icon":916,"to":61,"external":58,"cta":917},"09","personal","手边小卖部 HandyHub","账号与数字商品服务入口。","Store","访问官网",{"badge":919,"category":913,"title":920,"description":921,"icon":167},"10","亚马逊 Kindle 服务","围绕 Kindle 账号、设备与内容使用场景提供配套服务支持。",{"badge":923,"category":913,"title":924,"description":925,"icon":167},"11","全球电子书AI增强","面向全球电子书阅读、整理、问答与辅助理解场景的 AI 服务能力。",{"badge":927,"category":874,"title":928,"description":929,"icon":930,"to":931,"cta":871},"12","kfm_nuxtchat","面向业务系统的 AI 聊天与工作流集成模板。","MessageSquareCode","\u002Fproducts\u002Fkfm-nuxtchat",{"badge":933,"category":211,"title":934,"description":935,"icon":936,"to":76,"cta":138},"13","开发喵公共服务平台","统一提供短信、模型、网络与 IP 检测服务。","KeyRound",{"badge":938,"category":211,"title":939,"description":940,"icon":904,"to":941,"cta":871},"14","自主式 AI 部署 🔥","面向 Agent、RAG 与多服务协同的工程化部署方案。","\u002Fproducts\u002Fautonomous-ai-deployment",{"badge":943,"category":880,"title":944,"description":945,"icon":764,"to":946,"cta":871},"15","全球金融市场数据接口 🔥","提供高频与结构化市场数据接入能力。","\u002Fproducts\u002Fglobal-financial-market-data-api",[948,951,954],{"title":949,"description":950,"icon":667},"什么是 Token","Token 是模型处理文本时使用的最小计量单位，不完全等于“一个汉字”或“一个英文单词”。模型会先把输入拆成 token，再进行编码、注意力计算和生成。英文里常常一个词会拆成多个 token，中文里短句也可能按词块或子词切分。理解 token，有助于理解计费、速度和上下文限制。 ",{"title":952,"description":953,"icon":898},"LLM 上下文长度怎么理解","上下文长度可以理解为模型当前一次能“看见”的总 token 数量，里面同时包含系统提示词、历史对话、用户新输入以及模型生成内容。它不是只算你的提问文本。上下文越长，模型能参考的信息越多，但成本、延迟和注意力分配压力也会增加。",{"title":955,"description":956,"icon":92},"什么是 Skills","Skills 可以理解为给模型预先准备好的能力模块或工作说明。它不只是知识说明，更像一套可复用的操作规范，告诉模型在某类任务里该怎样判断、调用什么工具、遵循什么输出格式。Skills 做得好，模型表现会更稳定，也更接近真正可交付的系统。",{"ai-application-development":958,"llm-training":987,"ai-product-aggregation":1015,"everyone-llm":1043,"openlens":1070,"autonomous-ai-deployment":1097,"global-financial-market-data-api":1125,"global-llm-api":1153,"kfm-nuxtchat":1181},{"id":201,"slug":959,"title":960,"eyebrow":961,"summary":962,"icon":667,"intro":963,"features":966,"scenarios":973,"advantages":980,"primaryCta":984,"secondaryCta":986},"ai-application-development","AI应用软件开发","AI Applications","围绕真实业务流程设计和交付 AI 应用，把模型能力接入前端、后端、工作流与运维体系。",[964,965],"这类项目的重点不是把模型接上去，而是把角色、流程、输入输出、权限和稳定性一起设计清楚。","我们更关注交付后的可维护性，包括接口边界、日志观测、错误处理、版本迭代和业务方可用性。",[967,968,969,970,971,972],"业务流程梳理","前后端一体化交付","模型接入与工作流设计","权限与日志机制","多语言界面支持","上线与持续迭代",[974,975,976,977,978,979],"企业内部 Copilot","智能客服与售后支持","内容生产平台","运营分析助手","审批和知识协同系统","跨系统工作流集成",[981,982,983],"强调业务落地而不是单点 Demo","从产品体验到工程交付一体化推进","支持后续版本持续迭代与能力扩展",{"label":985,"to":50},"联系开发者中心",{"label":25,"to":41},{"id":241,"slug":988,"title":891,"eyebrow":989,"summary":990,"icon":162,"intro":991,"features":994,"scenarios":1001,"advantages":1008,"primaryCta":1012,"secondaryCta":1013},"llm-training","Model Training","覆盖数据整理、训练、评测、推理优化和上线交付的大模型训练服务。",[992,993],"不是所有场景都需要从零训练，但很多场景需要结合业务数据做微调、对齐和专项评测。","我们会根据预算、算力和目标效果，在训练强度、部署方案和上线节奏之间做实际权衡。",[995,996,997,998,999,1000],"训练数据清洗与构建","指令微调方案设计","评测指标与回归验证","推理性能优化","GPU 与资源规划","上线和运维方案",[1002,1003,1004,1005,1006,1007],"行业垂直模型","企业知识问答增强","内容生产能力强化","分类与审核模型","专项语义理解任务","定制化对话系统",[1009,1010,1011],"兼顾实验效率与交付质量","支持性能、成本和效果三方平衡","可与知识库、RAG 和 Agent 体系联动",{"label":678,"to":50},{"label":1014,"to":41},"返回产品页",{"id":278,"slug":1016,"title":1017,"eyebrow":1018,"summary":1019,"icon":898,"intro":1020,"features":1023,"scenarios":1030,"advantages":1037,"primaryCta":1041,"secondaryCta":1042},"ai-product-aggregation","AI产品聚合服务","AI Aggregation","统一接入文本、图像、语音、视频等多类 AI 能力，提供聚合层、路由层和运营层支撑。",[1021,1022],"当产品需要同时使用多种模型能力时，聚合层决定了接入成本、扩展速度和后续维护复杂度。","我们提供的不只是模型列表，而是统一鉴权、调用封装、路由策略和能力编排。",[1024,1025,1026,1027,1028,1029],"多能力统一接入","调用封装与鉴权","模型与能力路由","错误兜底机制","成本观测与配额管理","跨模型组合编排",[1031,1032,1033,1034,1035,1036],"AI 工具平台","企业统一 AI 能力层","多模态产品接入","内部研发平台","SaaS 能力聚合","对外 API 中台",[1038,1039,1040],"降低多模型接入复杂度","让产品团队更快试错和扩展","适合做长期能力中台和服务层",{"label":622,"to":50},{"label":1014,"to":41},{"id":315,"slug":1044,"title":875,"eyebrow":1045,"summary":1046,"icon":167,"intro":1047,"features":1050,"scenarios":1057,"advantages":1064,"primaryCta":1068,"secondaryCta":1069},"everyone-llm","Knowledge Base","面向团队知识沉淀、检索问答和内容协同的知识库产品，把文档、规则和业务经验整理成可被检索和复用的工作资产。",[1048,1049],"知识库的价值不只是“能搜到”，而是让团队能把分散的文档、流程、FAQ 和经验真正转成可引用、可维护、可更新的知识系统。","everyone-llm 更适合用在企业内部知识管理、客户支持、 SOP 沉淀和多角色协同场景中，让问答和检索真正服务业务。",[1051,1052,1053,1054,1055,1056],"文档接入与整理","RAG 检索问答","多知识源管理","引用片段与答案溯源","知识更新与维护","团队协同使用",[1058,1059,1060,1061,1062,1063],"企业内部知识库","售前售后问答","SOP 与操作手册查询","项目资料沉淀","培训与上手支持","跨团队经验共享",[1065,1066,1067],"把知识沉淀和问答能力放到同一产品里","更适合长期维护而不是一次性导入","能与 RAG、Agent 和工作流体系继续联动",{"label":622,"to":50},{"label":1014,"to":41},{"id":352,"slug":1071,"title":907,"eyebrow":1072,"summary":1073,"icon":909,"intro":1074,"features":1077,"scenarios":1084,"advantages":1091,"primaryCta":1095,"secondaryCta":1096},"openlens","Multi-Model Platform","面向多模型比较、问答评估和决策辅助的产品平台，帮助团队更高效地理解不同模型的差异与适用场景。",[1075,1076],"OpenLens 不是简单把多个模型放在一个页面里，而是把比较、观察、试用和策略判断放到同一工作界面里。","它更适合需要做模型选型、回答对比、质量观察和策略试验的个人用户与团队。",[1078,1079,1080,1081,1082,1083],"多模型并行对话","回答差异对比","策略与效果观察","统一交互入口","多场景试验支持","辅助模型选型",[1085,1086,1087,1088,1089,1090],"模型选型对比","Prompt 测试","回答质量观察","内部评估与演示","策略辅助决策","多模型工作流入口",[1092,1093,1094],"更适合做横向比较与判断","降低多模型试验成本","把观察和决策放在同一界面里",{"label":917,"to":57,"external":58},{"label":1014,"to":41},{"id":389,"slug":1098,"title":1099,"eyebrow":1100,"summary":1101,"icon":904,"intro":1102,"features":1105,"scenarios":1112,"advantages":1119,"primaryCta":1123,"secondaryCta":1124},"autonomous-ai-deployment","自主式AI部署","AI Deployment","面向 Agent、RAG、多模型调用和私有环境的工程化部署方案，强调稳定、可观测和可扩展。",[1103,1104],"部署不是把服务启动起来就结束，还需要处理资源规划、权限边界、日志观测、版本升级和异常回滚。","我们更关注可长期运行的工程体系，而不是一次性的环境搭建。",[1106,1107,1108,1109,1110,1111],"私有化与混合部署","多服务编排","权限与访问控制","日志监控与告警","版本升级与回滚","资源与成本规划",[1113,1114,1115,1116,1117,1118],"企业内网部署","Agent 系统落地","知识库与 RAG 平台","私有模型服务","多模型路由中台","长期运行的 AI 基础设施",[1120,1121,1122],"强调上线后的可持续运维","适合复杂 AI 工程体系整合","兼顾安全、性能和可扩展性",{"label":985,"to":50},{"label":1014,"to":41},{"id":426,"slug":1126,"title":1127,"eyebrow":1128,"summary":1129,"icon":764,"intro":1130,"features":1133,"scenarios":1140,"advantages":1147,"primaryCta":1151,"secondaryCta":1152},"global-financial-market-data-api","全球金融市场数据接口","Market Data","提供面向全球市场的结构化金融数据接口，支持量化研究、监控、策略和产品接入。",[1131,1132],"数据接口的价值不只是拿到数据，而是拿到稳定、结构清晰、适合下游系统消费的数据形态。","我们会更关注字段设计、调用稳定性、频率限制、观测能力和后续扩展。",[1134,1135,1136,1137,1138,1139],"全球市场数据接入","结构化字段设计","稳定调用与限频控制","历史与实时数据支持","接口文档与接入辅助","监控与异常观测",[1141,1142,1143,1144,1145,1146],"量化研究","交易监控看板","金融资讯产品","分析助手","策略信号系统","跨市场数据聚合",[1148,1149,1150],"接口更贴近下游应用消费","支持全球市场与扩展需求","适合产品化和工程化接入",{"label":678,"to":50},{"label":1014,"to":41},{"id":461,"slug":1154,"title":1155,"eyebrow":1156,"summary":1157,"icon":156,"intro":1158,"features":1161,"scenarios":1167,"advantages":1174,"primaryCta":1178,"secondaryCta":1180},"global-llm-api","全球大模型 API","Global LLM API","统一接入全球主流大模型能力，帮助产品和企业系统快速获得稳定的推理与生成接口。",[1159,1160],"对于需要面向全球用户和多模型策略的产品来说，统一 API 层可以明显降低接入和维护成本。","我们会把鉴权、路由、限流、日志、配额和成本控制放进同一层能力里。",[434,1162,1163,1164,1165,1166],"鉴权与配额管理","路由与限流机制","调用日志与监控","多语言接入支持","成本与使用观测",[1168,1169,1170,1171,1172,1173],"国际化产品接入","企业 AI 能力中台","内容生成平台","多模型实验和比较","客服与问答产品","开发者平台",[1175,1176,1177],"更适合全球化和多模型场景","统一接入层降低维护复杂度","支持后续扩展更多模型和能力",{"label":1179,"to":76},"进入公共服务平台",{"label":1014,"to":41},{"id":497,"slug":1182,"title":928,"eyebrow":1183,"summary":1184,"icon":930,"intro":1185,"features":1188,"scenarios":1195,"advantages":1202,"primaryCta":1206,"secondaryCta":1207},"kfm-nuxtchat","Chat Template","面向业务系统的 AI 聊天与工作流集成模板，适合快速构建带权限、上下文和业务动作的对话入口。",[1186,1187],"很多业务系统都需要一个能接模型、接流程、接权限的聊天入口，而不是单独的聊天窗口。kfm_nuxtchat 更适合做这类可继续扩展的集成模板。","它强调的是页面结构、会话管理、业务动作挂接和后续定制能力，适合作为企业内外部 AI 对话入口的起点。",[1189,1190,1191,1192,1193,1194],"聊天界面模板","上下文与会话管理","业务动作集成","权限与角色适配","多模型接入扩展","适合二次开发",[1196,1197,1198,1199,1200,1201],"企业内部 AI 助手","业务系统对话入口","客服与咨询窗口","流程型问答页面","知识问答前端","定制化 AI 产品原型",[1203,1204,1205],"不是孤立聊天页，而是业务集成模板","适合快速落地并继续扩展","能够承接权限、知识库和工作流需求",{"label":985,"to":50},{"label":1014,"to":41},[1209,1224,1239,1253,1266,1278,1290,1302,1314,1326,1338,1351,1364,1377,1389,1402,1415,1427,1440,1453,1466,1478,1490,1503,1515],{"id":201,"slug":1210,"category":1211,"title":1212,"summary":1213,"focus":1214,"accent":1215,"code":1216,"embedPath":1217,"steps":1218},"llm","fundamentals","LLM 问答过程动画 🔥","把提问、编码、推理、解码和输出的链路拆成可观察节点。","重点：Transformer 处理链路","from-blue-500 via-cyan-500 to-emerald-400","Tokenize -> Embed -> Attend -> Decode -> Answer","\u002Finteractive-demos\u002Fllm.html",[1219,1220,1221,1222,1223],"用户提问","Token 编码","上下文注意力","解码生成","答案输出",{"id":241,"slug":1225,"category":1226,"title":1227,"summary":1228,"focus":1229,"accent":1230,"code":1231,"embedPath":1232,"steps":1233},"rag","retrieval","RAG 检索增强生成 🔥","演示查询改写、召回、重排、拼接上下文与最终生成的完整流程。","重点：检索链路","from-sky-500 via-cyan-500 to-teal-400","Query -> Retrieve -> Rerank -> Context -> Generate","\u002Finteractive-demos\u002Frag.html",[1234,1235,1236,1237,1238],"用户查询","向量召回","重排筛选","上下文拼接","生成回答",{"id":278,"slug":1240,"category":1226,"title":1241,"summary":1242,"focus":1243,"accent":1244,"code":1245,"embedPath":1246,"steps":1247},"embedding","Embedding 向量空间","通过二维示意和相似度说明文本如何落入向量空间。","重点：向量空间直觉","from-amber-500 via-orange-500 to-rose-400","Text -> Vector -> Similarity -> Clusters","\u002Finteractive-demos\u002Fembedding.html",[1248,1249,1250,1251,1252],"原始文本","分词切块","向量映射","相似度计算","聚类结果",{"id":315,"slug":1254,"category":1211,"title":1255,"summary":1256,"focus":1257,"accent":1258,"code":1259,"embedPath":1260,"steps":1261},"token","什么是 Token 🔥","用动画把一句话拆成模型真正处理的 token，理解 token 不是“一个字=一个 token”。","重点：切分、计量与生成单位","from-fuchsia-500 via-violet-500 to-sky-400","Input -> Tokenize -> Count -> Process",null,[1262,1263,1264,1265],"输入文本","子词切分","token 计数","进入模型处理",{"id":352,"slug":1267,"category":1211,"title":1268,"summary":1269,"focus":1270,"accent":1271,"code":1272,"embedPath":1260,"steps":1273},"context-window","LLM 上下文长度","通过滑动窗口展示模型一次真正能“看到”的 token 范围，以及为什么旧内容会被截断。","重点：可见窗口与截断直觉","from-emerald-500 via-teal-500 to-cyan-400","System + History + User + Output \u003C= Context Window",[1274,1275,1276,1277],"系统提示词","历史对话","最新输入","生成输出",{"id":389,"slug":1279,"category":1280,"title":955,"summary":1281,"focus":1282,"accent":1283,"code":1284,"embedPath":1260,"steps":1285},"skills","orchestration","把 Skills 理解成给模型的能力模块，演示请求如何被技能路由并转成稳定执行过程。","重点：能力路由与执行规范","from-rose-500 via-orange-500 to-amber-400","Task -> Skill Match -> Tool Plan -> Structured Output",[1286,1287,1288,1289],"识别任务","匹配技能","生成执行计划","结构化输出",{"id":426,"slug":1291,"category":1211,"title":1292,"summary":1293,"focus":1294,"accent":1295,"code":1296,"embedPath":1260,"steps":1297},"prompt-structure","Prompt 结构演示","展示 system、user、assistant 示例如何被拼成最终输入，理解“提示词”不是单独一句话。","重点：消息结构与角色分工","from-indigo-500 via-sky-500 to-cyan-400","System + Few-shot + User -> Final Prompt",[1298,1299,1300,1301],"系统指令","示例消息","用户输入","拼接成最终提示",{"id":461,"slug":1303,"category":1211,"title":1304,"summary":1305,"focus":1306,"accent":1307,"code":1308,"embedPath":1260,"steps":1309},"transformer","Transformer 原理演示 🔥","用可视化方式展示 token 如何彼此关注、加权汇聚并形成新的上下文表示。","重点：自注意力与上下文建模","from-violet-500 via-indigo-500 to-cyan-400","Tokens -> Attention Scores -> Weighted Sum -> Contextual Output",[1310,1311,1312,1313],"输入 token","计算相关性","归一化权重","生成上下文表示",{"id":497,"slug":1315,"category":1211,"title":1316,"summary":1317,"focus":1318,"accent":1319,"code":1320,"embedPath":1260,"steps":1321},"temperature","Temperature 温度演示","用同一个问题对比低温和高温采样，理解模型为什么会更稳或更发散。","重点：随机性与稳定性","from-blue-500 via-violet-500 to-pink-400","Low Temp -> Stable | High Temp -> Diverse",[1322,1323,1324,1325],"同一提示词","设置温度","候选概率变化","输出风格差异",{"id":533,"slug":1327,"category":1226,"title":1328,"summary":1329,"focus":1330,"accent":1331,"code":1332,"embedPath":1260,"steps":1333},"rag-chunking","RAG 分块 Chunking 演示","对比大块、适中、小块切分对召回命中的影响，理解为什么 chunk 大小会改变答案质量。","重点：切块粒度与召回质量","from-emerald-500 via-teal-500 to-lime-400","Document -> Chunk -> Embed -> Retrieve",[1334,1335,1336,1337],"原始文档","不同粒度切块","向量化","召回命中差异",{"id":1339,"slug":1340,"category":1280,"title":1341,"summary":1342,"focus":1343,"accent":1344,"code":1345,"embedPath":1260,"steps":1346},11,"agent-tools","Agent 工具调用演示","展示 Agent 如何理解任务、挑选工具、读取结果并决定下一步，而不是一次性给答案。","重点：工具使用闭环","from-orange-500 via-amber-500 to-yellow-400","Task -> Tool -> Result -> Next Action",[1347,1348,1349,1350],"理解任务","选择工具","读取结果","继续决策",{"id":1352,"slug":1353,"category":1280,"title":1354,"summary":1355,"focus":1356,"accent":1357,"code":1358,"embedPath":1260,"steps":1359},12,"function-calling","Function Calling \u002F JSON 输出","展示模型如何把自然语言请求转成结构化参数，而不是只返回一段描述文字。","重点：结构化输出与参数映射","from-cyan-500 via-sky-500 to-indigo-400","Prompt -> Schema Match -> JSON Arguments",[1360,1361,1362,1363],"理解意图","匹配字段","生成 JSON","调用函数",{"id":1365,"slug":1366,"category":1211,"title":1367,"summary":1368,"focus":1369,"accent":1370,"code":1371,"embedPath":1260,"steps":1372},13,"chat-memory","多轮对话记忆","演示历史消息如何逐轮进入上下文，以及为什么对话越长越需要摘要和裁剪。","重点：历史消息与上下文占用","from-teal-500 via-emerald-500 to-lime-400","History + Latest Input -> Context Window",[1373,1374,1375,1376],"历史累积","上下文占用","摘要压缩","继续回答",{"id":1378,"slug":1379,"category":1226,"title":1380,"summary":1381,"focus":1382,"accent":1383,"code":1384,"embedPath":1260,"steps":1385},14,"rag-rerank","RAG 重排 Rerank","展示召回结果为什么还要重排，以及最终真正送进模型的片段通常只有少数几条。","重点：召回不等于最终采用","from-emerald-500 via-cyan-500 to-sky-400","Retrieve -> Score -> Rerank -> Keep Top Results",[1386,1387,1236,1388],"初始召回","相关性评分","送入生成",{"id":1390,"slug":1391,"category":1280,"title":1392,"summary":1393,"focus":1394,"accent":1395,"code":1396,"embedPath":1260,"steps":1397},15,"prompt-injection","Prompt 注入 \u002F 安全边界","说明为什么 system 指令、权限隔离和工具边界不能只靠模型“自觉遵守”。","重点：安全约束与越权风险","from-rose-500 via-red-500 to-orange-400","System Rules > User Injection > Guardrails",[1398,1399,1400,1401],"系统规则","恶意输入","安全检查","拒绝或隔离",{"id":1403,"slug":1404,"category":1226,"title":1405,"summary":1406,"focus":1407,"accent":1408,"code":1409,"embedPath":1260,"steps":1410},16,"embedding-threshold","Embedding 相似度阈值 🔥","通过相似度阈值控制展示为什么“有点像”不等于应该被采纳。","重点：相似度阈值与误召回","from-amber-500 via-orange-500 to-red-400","Vector Similarity >= Threshold ?",[1411,1412,1413,1414],"向量比较","计算相似度","设置阈值","决定是否采用",{"id":1416,"slug":1417,"category":1226,"title":1418,"summary":1419,"focus":1420,"accent":1357,"code":1421,"embedPath":1260,"steps":1422},17,"vector-database","向量数据库原理与存储","展示文本如何被写入向量库、建立索引，并在近邻检索时返回最相关的记录。","重点：向量写入、索引结构与存储记录","Document -> Embedding -> Vector Index -> ANN Search -> Metadata Hit",[1423,1424,1425,1426],"文本入库","生成向量","建立索引","近邻检索",{"id":1428,"slug":1429,"category":362,"title":1430,"summary":1431,"focus":1432,"accent":1433,"code":1434,"embedPath":1260,"steps":1435},18,"llm-distillation","大模型蒸馏","演示 Teacher 模型如何把能力迁移到更小的 Student 模型，以换取更低成本和更快响应。","重点：能力迁移、成本压缩与效果平衡","from-violet-500 via-fuchsia-500 to-rose-400","Teacher Output -> Distill -> Student Model",[1436,1437,1438,1439],"教师模型生成","软标签学习","对齐训练","学生模型上线",{"id":1441,"slug":1442,"category":362,"title":1443,"summary":1444,"focus":1445,"accent":1446,"code":1447,"embedPath":1260,"steps":1448},19,"llm-fine-tuning","大模型微调","展示通用模型如何通过业务数据微调，逐步适应特定领域语气、术语和输出格式。","重点：任务对齐、参数更新与效果提升","from-emerald-500 via-cyan-500 to-blue-400","Base Model + Domain Data -> Fine-tune -> Specialized Model",[1449,1450,1451,1452],"准备数据集","设定训练目标","参数更新","验证效果",{"id":1454,"slug":1455,"category":362,"title":1456,"summary":1457,"focus":1458,"accent":1459,"code":1460,"embedPath":1260,"steps":1461},20,"lora-vs-full-finetune","LoRA \u002F 全量微调对比","对比 LoRA 和全量微调在显存占用、训练成本、上线灵活性和效果提升上的差异。","重点：参数更新范围与工程取舍","from-indigo-500 via-violet-500 to-fuchsia-400","Base Model -> LoRA Adapters | Full Parameter Update",[1462,1463,1464,1465],"选择训练方式","更新参数范围","观察资源消耗","比较上线效果",{"id":1467,"slug":1468,"category":362,"title":1469,"summary":1470,"focus":1471,"accent":1472,"code":1473,"embedPath":1260,"steps":1474},21,"model-routing","模型路由","展示同一个请求为什么会按成本、速度和质量要求被分发给不同模型。","重点：路由策略与成本质量平衡","from-violet-500 via-indigo-500 to-sky-400","Task -> Route Policy -> Best Model",[1286,1475,1476,1477],"评估约束","分发模型","返回结果",{"id":1479,"slug":1480,"category":1226,"title":1481,"summary":1482,"focus":1483,"accent":1484,"code":1485,"embedPath":1260,"steps":1486},22,"rag-citations","RAG 引用来源","展示答案为什么需要带出处、片段编号和引用范围，帮助用户判断内容可信度。","重点：答案可信度与来源追踪","from-cyan-500 via-teal-500 to-emerald-400","Question -> Retrieve -> Answer + Citations",[1487,1488,1238,1489],"提出问题","召回片段","标注引用",{"id":1491,"slug":1492,"category":362,"title":1493,"summary":1494,"focus":1495,"accent":1496,"code":1497,"embedPath":1260,"steps":1498},23,"hallucination-check","模型幻觉与校验","通过对比“直接回答”和“经过校验的回答”，直观看到为什么系统需要事实校验与置信控制。","重点：幻觉风险与事实校验","from-pink-500 via-rose-500 to-orange-400","Draft Answer -> Verify -> Confidence Gate",[1499,1500,1501,1502],"生成草稿","检查事实","评估置信","输出结果",{"id":1504,"slug":1505,"category":1211,"title":1506,"summary":1507,"focus":1508,"accent":1509,"code":1510,"embedPath":1260,"steps":1511},24,"memory-summary","多轮对话摘要压缩","演示长对话为什么不能原样一直保留，以及系统如何把历史压成可复用摘要。","重点：记忆压缩与上下文节省","from-teal-500 via-cyan-500 to-sky-400","Long History -> Summary Memory -> Continue",[1373,1512,1513,1514],"触发压缩","生成摘要","继续对话",{"id":1516,"slug":1517,"category":1280,"title":1518,"summary":1519,"focus":1520,"accent":1408,"code":1521,"embedPath":1260,"steps":1522},25,"permission-routing","权限分层调用","展示同一个请求在普通用户、审核员和管理员权限下，会进入完全不同的工具链路。","重点：权限边界与工具分流","Request + Role -> Allowed Tools -> Result",[1523,1524,1525,1477],"识别身份","权限过滤","调用可用工具",{"ai-api":1527,"llm-development":1553,"ai-knowledge-base":1578,"rag-app":1605,"agent-app":1632,"blockchain":1658},{"id":201,"slug":1528,"title":154,"eyebrow":1529,"summary":1530,"icon":156,"intro":1531,"features":1534,"scenarios":1540,"advantages":1547,"primaryCta":1551,"secondaryCta":1552},"ai-api","AI Service","稳定的多模型接入与统一调用能力，帮助现有系统快速获得推理与生成能力。",[1532,1533],"支持文本、图像、语音等能力的统一接入，适合现有业务系统平滑升级。","提供鉴权、路由、限流和日志观测，降低业务团队的模型接入复杂度。",[1535,1536,1164,1537,1538,1539],"统一鉴权与计费","多模型路由","弹性扩缩容","多语言 SDK","成本透明",[1541,1542,1543,1544,1545,1546],"智能客服","内容生成","文本分析","语音与图像处理","翻译场景","推荐与个性化",[1548,1549,1550],"微服务与分布式架构","支持高并发与弹性扩缩","配套文档、SDK 与接入支持",{"label":1179,"to":76},{"label":25,"to":41},{"id":241,"slug":1554,"title":160,"eyebrow":1555,"summary":1556,"icon":162,"intro":1557,"features":1560,"scenarios":1567,"advantages":1572,"primaryCta":1576,"secondaryCta":1577},"llm-development","Model Engineering","覆盖数据治理、训练、评估、微调与上线的完整大模型工程服务。",[1558,1559],"适用于企业知识问答、行业模型适配和专项能力强化等场景。","从训练管线设计到推理部署，兼顾实验效率与工程可交付性。",[1561,1562,1563,1564,1565,1566],"训练数据清洗","指令数据构建","微调与评测","推理优化","GPU 训练经验","上线方案",[1002,974,1568,1569,1570,1571],"客服增强","内容审核","专用生成任务","多轮对话优化",[1573,1574,1575],"兼顾实验与工程落地","支持成本和性能权衡","可与知识库和 Agent 方案联动",{"label":678,"to":50},{"label":680,"to":38},{"id":278,"slug":1579,"title":165,"eyebrow":1045,"summary":1580,"icon":167,"intro":1581,"features":1584,"scenarios":1591,"advantages":1598,"primaryCta":1602,"secondaryCta":1604},"ai-knowledge-base","把分散文档沉淀为可检索、可维护、可审计的知识系统。",[1582,1583],"围绕文档治理、切片、索引、权限和问答效果做整体设计。","适合 SOP、售后手册、制度、产品资料与项目文档场景。",[1585,1586,1587,1588,1589,1590],"文档治理","结构化切片","向量索引","权限控制","知识更新流程","效果评估",[1592,1593,1594,1595,1596,1597],"企业内部知识问答","售前资料检索","客服辅助","研发文档检索","项目交付文档管理","培训助手",[1599,1600,1601],"强调知识生命周期管理","支持权限与可审计","可与 RAG 和 Agent 方案组合",{"label":1603,"to":173},"查看 RAG 服务",{"label":622,"to":50},{"id":315,"slug":1606,"title":170,"eyebrow":1607,"summary":1608,"icon":172,"intro":1609,"features":1612,"scenarios":1618,"advantages":1625,"primaryCta":1629,"secondaryCta":1631},"rag-app","Retrieval Augmented Generation","围绕召回、重排、上下文构造与生成效果优化的 RAG 工程服务。",[1610,1611],"适合需要把业务语料接入模型、提升回答准确性与可控性的场景。","关注可解释性、召回质量、权限和落地体验，而不只是把模型接上去。",[1613,1614,1615,1616,1617,1590],"查询改写","召回与重排","上下文治理","引用与出处","权限隔离",[1619,1620,1621,1622,1623,1624],"企业问答","帮助中心","售后支持","运营知识助手","法务检索","教育资料问答",[1626,1627,1628],"更贴近业务语料","比纯生成更可控","可与知识库和 API 体系联动",{"label":26,"to":1630},"\u002Fdemo\u002Frag",{"label":680,"to":38},{"id":352,"slug":1633,"title":175,"eyebrow":1634,"summary":1635,"icon":177,"intro":1636,"features":1639,"scenarios":1645,"advantages":1652,"primaryCta":1656,"secondaryCta":1657},"agent-app","AI Agent","构建具备任务拆解、工具调用、状态管理与执行回路的 Agent 系统。",[1637,1638],"适合流程自动化、复杂问答、跨系统执行和多步骤分析任务。","重点是把 Agent 从概念演示做成可观测、可回滚、可维护的工程系统。",[1640,1641,1642,1588,1643,1644],"任务编排","工具调用","状态持久化","失败重试","执行观测",[1646,1647,1648,1649,1650,1651],"运营自动化","智能分析助手","跨系统工单流转","内部流程执行","知识与工具结合问答","多角色协作",[1653,1654,1655],"强调工程稳定性","支持与现有业务系统集成","便于沉淀模板和组件",{"label":985,"to":50},{"label":25,"to":41},{"id":389,"slug":1659,"title":180,"eyebrow":1660,"summary":1661,"icon":97,"intro":1662,"features":1665,"scenarios":1672,"advantages":1679,"primaryCta":1683,"secondaryCta":1684},"blockchain","Blockchain","覆盖智能合约、链上交互、钱包接入与 DApp 交付的区块链开发服务。",[1663,1664],"支持链上产品从原型、合约设计到前后端接入与部署上线。","适合需要资产逻辑、链上凭证、交易流程或去中心化交互的业务。",[1666,1667,1668,1669,1670,1671],"智能合约开发","钱包接入","链上交易流程","DApp 前端","测试与审计协同","部署上线",[1673,1674,1675,1676,1677,1678],"链上资产管理","会员与凭证系统","链游功能模块","支付与结算场景","交易数据看板","链上身份场景",[1680,1681,1682],"前后端与链上协同交付","可与 AI 服务结合形成复合方案","支持定制化产品开发",{"label":622,"to":50},{"label":680,"to":38},{"source":1686},"directus",1776250603267]