{"schemaVersion":"1.0","updatedAt":"2026-09-04","project":{"id":"hermes-notebook","name":{"zh":"Hermes Notebook","en":"Hermes Notebook"},"publicStatus":"published","sourceVisibility":"private-curated","purpose":{"zh":"把散落在企业各处的文档和业务资料整理成可检索、可引用的知识，让客服机器人、内部问答和其他 AI 应用可以使用同一份企业数据。","en":"Turn documents and business information scattered across an enterprise into retrievable, citable knowledge that customer-service bots, internal Q&A, and other AI applications can share."},"originalWorkflow":[{"zh":"企业资料分散在本地文件夹、PDF、Word、Excel、PPT、图片和业务网页中，由不同部门和员工分别保管。","en":"Enterprise information was spread across local folders, PDFs, Word, Excel, PowerPoint, images, and business web pages maintained by different teams and employees."},{"zh":"员工遇到问题时依靠文件名、目录和个人经验查找资料，再手工确认答案来自哪一份文档。","en":"Employees relied on filenames, folders, and personal memory to find information, then manually checked which document supported an answer."},{"zh":"每开发一个客服机器人或内部助手，都要重新收集、清洗和导入相似资料，应用之间难以共用整理成果。","en":"Each new customer-service bot or internal assistant required similar information to be collected, cleaned, and imported again, with little reuse between applications."}],"workflowBreakpoints":[{"zh":"不同文件格式保留结构的方式不同，表格、标题、图片文字和跨页内容直接交给 AI 后容易失去原来的语义关系。","en":"Different formats preserve structure differently, so tables, headings, text in images, and cross-page content can lose their original relationships when sent directly to AI."},{"zh":"普通搜索能找到文件，却很难同时返回可以核对的原文位置、资料完整度和来源状态。","en":"Basic search may find a file but often cannot return a verifiable passage, its exact source location, completeness, and source status together."},{"zh":"每个 AI 应用单独维护一套资料，会重复清洗数据，也会让不同应用引用到不一致的版本。","en":"Maintaining a separate corpus for every AI application repeats data preparation and allows different applications to use inconsistent versions."}],"systemBoundary":{"zh":"Hermes 运行在客户控制的本地环境中，负责资料导入、结构整理、索引、检索和带引用问答。上层客服机器人、企业聊天工具和其他业务应用通过 API 使用这些能力；渠道身份和部门权限映射由企业接入层负责，Hermes 在检索前执行已配置的知识库访问范围。","en":"Hermes runs in a customer-controlled local environment and handles ingestion, structural preparation, indexing, retrieval, and citation-backed Q&A. Customer-service bots, enterprise chat tools, and other business applications use these capabilities through APIs; channel identity and department mapping belong to the enterprise integration layer, while Hermes enforces configured knowledge-base access before retrieval."},"inputs":[{"zh":"PDF、Word、Excel、PPT、图片和 OCR 文档等企业资料","en":"Enterprise information in PDF, Word, Excel, PowerPoint, image, and OCR document formats"},{"zh":"管理员指定的本地文件夹、单个文件、压缩包和业务网页 URL","en":"Administrator-selected local folders, individual files, archives, and business web page URLs"},{"zh":"知识库范围、资料完整度和外部应用需要查询的问题","en":"Knowledge-base scope, source completeness, and questions submitted by downstream applications"}],"processing":[{"zh":"按文件格式提取正文、标题、表格和图片文字，并保留每段内容与原始资料位置的关系。","en":"Extract text, headings, tables, and image text by format while preserving the relationship between each passage and its original location."},{"zh":"把整理后的内容写入对应知识库，建立可检索的知识节点、来源记录和完整度状态。","en":"Write prepared content into the relevant knowledge base and build retrievable knowledge nodes, source records, and completeness states."},{"zh":"外部应用查询时，系统先校验 Token、接口权限和知识库范围，再返回证据片段或基于这些证据生成带引用答案。","en":"When an application queries the system, Hermes first checks its token, API scope, and allowed knowledge bases, then returns evidence passages or generates a cited answer from that evidence."}],"humanReview":[{"zh":"管理员决定哪些文件夹和资料可以进入哪个知识库，并检查导入失败、重复文件和资料完整度。","en":"Administrators decide which folders and sources enter each knowledge base and review failed imports, duplicates, and source completeness."},{"zh":"业务负责人确认检索结果和引用是否足以支持回答，再决定哪些知识库可以提供给具体应用。","en":"Business owners confirm whether retrieved evidence and citations support the answer and decide which knowledge bases each application may use."},{"zh":"客服话术、对外回复和需要执行的业务动作继续由对应应用和人员审核。","en":"Customer-service wording, external responses, and business actions remain subject to review in the relevant application and by responsible staff."}],"outputs":[{"zh":"按业务主题整理的企业知识库，包含来源、原文位置、完整度和处理记录","en":"Enterprise knowledge bases organized by business topic with sources, original locations, completeness, and processing records"},{"zh":"供应用直接调用的证据检索结果，以及带来源引用的知识问答","en":"Evidence retrieval results ready for application use and knowledge answers with source citations"},{"zh":"面向客服机器人、内部知识问答和其他业务助手的检索与问答 API","en":"Retrieval and Q&A APIs for customer-service bots, internal knowledge Q&A, and other business assistants"}],"failureRecovery":[{"zh":"文件导入保留成功、失败和跳过记录；重复内容通过来源与文件指纹识别，避免无提示地重复写入。","en":"Ingestion preserves success, failure, and skip records, while source and file fingerprints identify duplicates instead of silently importing them again."},{"zh":"无法完整提取的资料会保留 partial 等完整度状态，检索和回答可以据此降低承诺，不把缺失内容当成完整事实。","en":"Sources that cannot be fully extracted retain completeness states such as partial, allowing retrieval and answers to reduce confidence instead of treating missing content as complete fact."}],"currentStatus":{"zh":"项目目前支持 Docker Compose 本地部署，已具备文件、文件夹和 URL 导入，多格式资料整理，可追溯检索，带引用问答，以及受 Token 和知识库范围限制的检索与问答 API。","en":"The project currently supports local Docker Compose deployment, file, folder, and URL ingestion, multi-format preparation, traceable retrieval, citation-backed Q&A, and retrieval and query APIs restricted by tokens and knowledge-base scope."},"transferableCapabilities":[{"zh":"把多个部门和多种格式的企业资料整理成一套可共用的知识来源","en":"Prepare multi-department, multi-format enterprise information as one reusable knowledge source"},{"zh":"让每条检索证据和回答可以回到原始文档位置核对","en":"Make every retrieved passage and answer traceable to its original document location"},{"zh":"让客服机器人、内部问答和后续业务助手复用同一份企业知识，减少重复导入和清洗","en":"Let customer-service bots, internal Q&A, and future business assistants reuse the same enterprise knowledge and reduce repeated ingestion and preparation"},{"zh":"通过知识库范围和接口权限控制不同应用能够检索的资料","en":"Control which information each application may retrieve through knowledge-base scope and API permissions"}],"evidenceArtifacts":[{"id":"knowledge-flow-film","label":{"zh":"企业资料整理与问答流程演示","en":"Enterprise knowledge ingestion and Q&A demo"},"description":{"zh":"展示企业资料从导入、整理和检索，到向上层应用返回证据与引用答案的过程。","en":"Shows enterprise information moving from ingestion and preparation through retrieval to evidence and cited answers for downstream applications."},"kind":"demo","href":"/animations/projects/hermes/index.html"}],"limitations":[{"zh":"Hermes 提供企业知识整理和检索内核；客服机器人、企业微信或飞书接入、界面和业务动作需要由上层应用完成。","en":"Hermes provides the enterprise knowledge preparation and retrieval engine; customer-service bots, WeCom or Feishu integration, interfaces, and business actions belong to downstream applications."},{"zh":"原始资料缺失、过期或扫描质量过低时，系统无法凭空补齐事实，需要企业先修正资料来源。","en":"When source information is missing, outdated, or poorly scanned, the system cannot reconstruct facts and the enterprise must correct the source material."},{"zh":"企业群聊和部门身份不能只依赖模型判断，必须由接入层明确映射到允许访问的知识库。","en":"Enterprise group and department identity cannot rely on model judgment and must be explicitly mapped to allowed knowledge bases by the integration layer."}],"reviewedAt":"2026-08-15","review":{"approvedForPublic":true,"approvedBy":"site-owner"},"featured":true,"sortOrder":15,"visual":{"animationSrc":"/animations/projects/hermes/index.html","alt":{"zh":"Hermes Notebook 企业资料导入、检索和引用问答流程","en":"Hermes Notebook enterprise ingestion, retrieval, and cited Q&A flow"}}},"canonicalUrl":"/projects/hermes-notebook"}