发布日期: 2025-08-09
版本号: v0.6.19

Open WebUI 0.6.19版本于2025年8月9日发布，带来了大量功能更新、安全修复和性能优化。在界面方面，侧边栏经过全新现代化设计，支持固定页头页脚；新增了引导式响应再生功能，允许用户通过提示或快捷选项调整AI回复；聊天输入现在具备格式化工具栏，并支持为多模型回复启用标签视图。功能上，新增了对Amazon S3向量、Oracle 23ai向量搜索和SCIM 2.0协议的支持；引入了SQLite数据库加密、Redis连接池与集群模式、批量响应流处理及全局配置；聊天搜索增加了状态筛选器，管理员可以更精细地控制聊天设置权限。安全与管理方面，修复了频道消息、OAuth认证和API访问权限等关键安全问题；增强了管理员对工作区内容的访问控制，支持用户数据导出、模型配置批量管理与同步；并优化了OpenTelemetry的日志与指标导出配置。性能与稳定性方面，改进了Redis和PGVector的连接处理，修复了多项工具调用、知识库更新和界面渲染问题，提升了整体响应速度和多设备兼容性。此外，更新还包含对多语言翻译的完善、无障碍访问的增强，以及将Docling集成升级至v1 API等重要变更。

更新内容 （中文）

[0.6.19] - 2025-08-09

新增

  • ✨ 现代化侧边栏与重大界面优化：主导航侧边栏已彻底重新设计，采用更现代、简洁的美学风格，并设有固定的页眉和页脚以保持关键控件始终可用。核心侧边栏逻辑，例如已固定的模型列表，也已重构为专用组件以提升性能和可维护性。
  • 🪄 引导式响应重新生成： “重新生成”按钮已变身为一个功能强大的新菜单。现在，您可以通过文本提示建议更改来引导AI的下一次尝试，或使用“重试”、“添加细节”或“更简洁”等一键选项，即时优化和重塑响应，使其更符合您的需求。
  • 🛠️ 改进GPT-OSS模型的工具调用处理：为GPT-OSS模型实现了强大的工具调用处理，确保函数执行和集成正确无误。
  • 🛑 合并响应的停止按钮：添加了专用的停止按钮，可立即中止合并AI响应的生成，为用户提供对持续输出的更多控制权。
  • 🔄 实验性SCIM 2.0支持：实现了SCIM 2.0（跨域身份管理系统）协议支持，可从Okta、Azure AD和Google Workspace等身份提供商启用企业级的自动化用户和组配置，实现无缝的用户生命周期管理。配置通过环境变量安全地进行管理。
  • 🗂️ Amazon S3向量支持：现在可以将Amazon S3 Vector作为高性能向量数据库用于您的检索增强生成工作流。这为深度集成到AWS生态系统中的用户提供了可扩展、云原生的存储选项，简化了基础设施并支持企业级知识管理。
  • 🗄️ Oracle 23ai向量搜索支持：新增了对Oracle 23ai新型向量搜索功能的支持，将其作为受支持的向量数据库，为管理大规模文档以及在数据库层面将向量搜索与现有业务数据集成提供了一个强大且可扩展的选项。
  • ⚡ Qdrant性能与配置增强：Qdrant客户端已得到显著改进，为“get”和“query”操作提供了更快的数据检索逻辑。新的环境变量（‘QDRANT_TIMEOUT’、‘QDRANT_HNSW_M’）为管理员提供了对查询超时和HNSW索引参数的更精细控制，从而能够对大规模部署进行更好的性能调优。
  • 🔐 使用SQLCipher加密SQLite数据库：现在可以使用SQLCipher对整个SQLite静态数据库进行加密。通过将“DATABASE_TYPE”设置为“sqlite+sqlcipher”并提供“DATABASE_PASSWORD”，所有数据都将被透明加密，为自托管部署中的敏感信息保护提供基本的安全层。请注意，这需要额外的系统库和’sqlcipher3-wheels’ Python包。
  • 🚀 高效的Redis连接管理：实现了共享连接池缓存以复用Redis连接，大幅减少了活跃客户端数量。这防止了连接耗尽错误，提升了性能，并确保了在高并发部署和使用Redis Sentinel的环境中具有更高的稳定性。
  • ⚡ 批量响应流式传输以提升性能：通过在将多个令牌发送给客户端之前将它们批量组合，显著提升了高速响应流式传输期间的性能和稳定性。一个新的“流式增量块大小”高级参数可以按模型或在用户/聊天设置中进行设置，显著降低了服务器、Redis和客户端的CPU负载，并防止了在高并发环境中的连接问题。
  • ⚙️ 全局批量流式传输配置：管理员现在可以使用新的“CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE”环境变量设置响应流式传输的系统范围默认值。这允许进行全局性能调优，同时仍然允许按模型或按聊天的设置覆盖默认值，以进行更精细的控制。
  • 🔎 带状态过滤器的高级聊天搜索：使用强大的新搜索过滤器快速查找任何对话。现在您可以直接在搜索栏中使用“pinned:true”、“shared:true”和“archived:true”等前缀来即时缩小聊天范围。智能下拉菜单会在您输入时建议可用的过滤器选项，简化您的工作流程，使聊天管理比以往更高效。
  • 🛂 细粒度聊天控制权限：管理员现在可以更详细地管理聊天设置。主要的“聊天控制”权限现在充当主开关，而新的“Valves”、“系统提示”和“高级参数”细粒度开关允许对用户在面板内可见的特定部分进行更具体的控制。
  • ✍️ 聊天输入的格式工具栏：为富文本聊天输入框引入了专用的格式工具栏，为用户提供更便捷的文本样式和编辑选项，可通过界面设置进行配置。
  • 📑 多模型响应的标签页视图：现在可以启用新的标签页界面来查看多个模型的响应。此紧凑视图不再使用横向滚动的卡片，而是将每个模型的响应组织到各自的标签页中，便于比较输出并节省垂直空间。此功能可以在界面设置中开启或关闭。
  • ↕️ 通过拖放重新排列固定模型：现在可以通过简单的拖放操作来组织侧边栏中固定的模型，按照您偏好的顺序排列。此自定义布局会自动保存，为您提供了更灵活的工作区控制。
  • 📌 快速取消固定模型快捷方式：现在可以通过按住Shift键并将鼠标悬停在其上来快速取消固定模型，显示即时取消固定按钮，简化您的工作区自定义。
  • ⚡ 改进的聊天输入性能：聊天输入现在响应速度显著提升，尤其是在粘贴或输入大量文本时。这是通过实现自动保存功能的去抖机制实现的，该机制可防止UI延迟并确保流畅、不间断的打字体验。
  • ✍️ 可自定义的浮动快捷操作与工具支持：通过新的可自定义浮动快捷操作完全控制您的文本交互工作流程。在设置中，您可以创建、编辑或禁用这些操作，甚至可以使用“{{TOOL:tool_id}}”语法在提示中集成工具，从而实现对所选文本的强大一键自动化。这额外支持使用“{{CONTENT}}”和“{{INPUT_CONTENT}}”等占位符进行自定义文本转换。
  • 🔒 管理员工作区隐私控制：引入了’ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS’环境变量（默认为’True’），允许管理员控制其对工作区项目（知识库、模型、提示词、工具）的访问权限。禁用后，管理员将遵守与普通用户相同的访问控制规则，增强了多租户部署的数据隔离性。
  • 🗄️ 全面的模型配置管理：管理员现在可以将整个模型配置导出到文件，并使用新的声明式同步端点来批量管理模型。这个强大的功能支持无缝备份、迁移以及跨多个实例的状态复制。
  • 📦 原生Redis集群模式支持：添加了对以集群模式连接Redis的完整支持，允许超越Sentinel管理的设置，实现可扩展和高可用的Redis部署。新的环境变量’REDIS_CLUSTER’和’WEBSOCKET_REDIS_CLUSTER’允许使用’redis.cluster.RedisCluster’客户端。
  • 📊 细粒度OpenTelemetry指标配置：引入了专门的环境变量和增强的OpenTelemetry指标配置选项，允许为指标导出单独设置OTLP端点、基本认证凭据和协议（HTTP/gRPC），独立于跟踪设置。这为集成到各种可观测性栈中提供了更大的灵活性。
  • 🪵 细粒度OpenTelemetry日志配置：通过引入专用的日志环境变量，增强了OpenTelemetry日志集成，允许为日志导出单独设置OTLP端点、基本认证凭据和协议（HTTP/gRPC），独立于常规OTel设置。应用程序的默认Python记录器现在利用此配置，在通过’ENABLE_OTEL_LOGS’启用时自动将日志发送到您的OTel端点。
  • 📁 增强的文件夹聊天管理，支持排序和时间分组：文件夹内的聊天列表现在支持按标题和更新时间进行全面排序，并支持智能的时间分组（例如，“今天”、“昨天”），类似于主聊天视图，使项目特定对话的导航和组织更加轻松。
  • ⚙️ 可配置的Datalab Marker API与高级处理选项：增强了Datalab Marker API集成，允许管理员为自托管配置自定义API基础URL，并通过新的’additional_config’ JSON参数指定全面的处理选项。这取代了已弃用的语言选择功能，并提供了对文档提取的细粒度控制，简化了API端点解析，使自托管部署更加稳健。
  • 🧑‍💼 将所有用户导出为CSV：管理员现在可以直接从管理员面板的数据库设置将所有用户的完整列表导出为CSV文件。这提供了一种简单的一键式方式，用于生成审计、报告或管理目的的用户数据。
  • 🛂 可自定义的OAuth ‘sub’声明：管理员现在可以使用’OAUTH_SUB_CLAIM_OVERRIDE’环境变量指定应使用身份提供商的哪个声明作为唯一用户标识符（‘sub’）。这为复杂的企业身份验证设置提供了更大的灵活性和控制力，尤其是在无法修改IDP默认声明的情况下。
  • 👁️ 输入框的密码可见性切换：应用程序中的密码字段（登录、注册、用户管理和账户设置）现在使用新的’SensitiveInput’组件，提供一致的切换按钮以显示/隐藏密码，从而提升可用性和安全性。
  • 🛂 注册时可选的“确认密码”：为了帮助防止在创建账户时输入密码错误，管理员现在可以在注册页面启用“确认密码”字段。此功能默认禁用，可通过环境变量激活以增强用户体验。
  • 💬 从用户反馈中查看完整聊天：管理员现在可以从反馈模态框直接导航到与用户反馈条目关联的完整对话，简化审查和故障排除流程。
  • 🎚️ 直观的混合搜索BM25-权重滑块：混合搜索中BM25-权重参数的数字输入已替换为交互式滑块，提供了一种更直观的方式来调整词汇搜索和语义搜索之间的平衡。“默认/自定义”切换和更清晰的标签增强了对这个关键参数的可用性和理解。
  • ⚙️ 增强的批量函数同步：用于同步函数的API端点已得到显著改进，能够可靠地处理批量更新。这确保了管理员导入和管理大型函数库时更加稳健和无错误。
  • 🖼️ 禁用频道中图片压缩的选项：在界面选项下引入了新设置，允许用户强制禁用专门用于频道中发布图片的图片压缩，确保重要内容的图片保持更高分辨率。
  • 🔗 自定义CORS协议支持：引入了新的环境变量’CORS_ALLOW_CUSTOM_SCHEME’，允许管理员为CORS源定义自定义URL协议（例如，‘app://’），为本地开发或桌面客户端集成提供更大的灵活性。
  • ♿ 可翻译且无障碍的横幅：增强了横幅元素，提供可翻译的徽章文本和适当的ARIA属性（aria-label, aria-hidden）用于SVG图标，显著提升了无障碍性和屏幕阅读器兼容性。
  • ⚠️ 缺少OPENID_PROVIDER_URL时的OAuth配置警告：添加了主动启动警告，当配置了OAuth提供商（Google、Microsoft或GitHub）但缺少必需的’OPENID_PROVIDER_URL’环境变量时通知管理员。这可以防止静默的OAuth注销失败，并引导管理员正确完成设置。
  • ♿ 重大无障碍性增强：界面的关键部分已得到显著的无障碍性改进。用户资料菜单现在可通过键盘完全导航，Playground中的基本控件现在包含适当的ARIA标签供屏幕阅读器使用，装饰性图片已对辅助技术隐藏以减少音频干扰。菜单按钮还通过’aria-label’、SVG的’aria-hidden’和切换按钮的’aria-pressed’增强了可访问性。
  • ⚙️ 常规后端重构：实施了各种后端改进以增强性能、稳定性和安全性，确保为所有用户提供更具弹性和可靠性的平台，包括通过有条件地包含’extra_json’字段来使日志输出更清晰高效，改进向量数据库操作中一致的元数据处理，以及为未来的分析功能奠定初步基础。
  • 🌐 本地化与国际化改进：完善并扩展了加泰罗尼亚语、丹麦语、韩语、波斯语、波兰语、简体中文和西班牙语的翻译，确保所有支持的语言为全球用户提供更流畅和原生的体验。

修复

  • 🛡️ 强化频道消息安全：修复了一个关键权限缺陷，该缺陷允许具有频道访问权限的用户编辑或删除他人的消息。系统现在正确强制执行用户只能修改自己的消息，保护共享频道中的数据完整性。
  • 🛡️ 通过从URL中移除JWT强化OAuth安全性：修复了一个关键安全漏洞，该漏洞在成功的OAuth登录后会在URL中暴露身份验证令牌。现在通过浏览器Cookie传输令牌，防止通过浏览器历史记录或服务器日志泄露，保护用户会话。
  • 🛡️ 强化聊天完成API安全性：聊天完成API端点现在包含明确的所有权检查，确保非管理员用户无法访问不属于他们的聊天，并防止潜在的未经授权访问。
  • 🛠️ 弹性模型加载：修复了当加载模型列表失败（例如，由于提供商配置错误）会阻止整个用户界面（包括管理员面板）加载的问题。应用程序现在可以优雅地处理这些错误，确保UI保持可访问。
  • 🔒 解决FIPS自测失败：修复了一个关键问题，该问题阻止了Open WebUI在符合FIPS的系统上运行，具体解决了与OpenSSL和SentenceTransformers相关的“FATAL FIPS SELFTEST FAILURE”错误，恢复了与安全环境的兼容性。
  • 📦 Redis集群连接恢复：修复了后端无法连接到集群模式下的Redis的问题，现在确保与可扩展的Redis集群部署无缝集成。
  • 📦 PGVector连接稳定性：修复了只读操作可能导致数据库事务空闲的问题，防止潜在的连接错误，提高整体数据库稳定性和资源管理。
  • 🛠️ OpenAPI工具集成对数组参数的修复：解决了一个关键错误，即使用数组参数（例如，用于标签）的外部工具在与OpenAI模型一起使用时会失败。系统现在正确地在函数模式中生成必需的’items’属性，恢复了功能并防止了'400 Bad Request’错误。
  • 🛠️ 恢复用户创建工具的功能：修复了代码编辑器中的一个错误，该错误导致状态消息被错误地添加到工具脚本开头，导致保存时出现语法错误。所有授权用户现在可以可靠地创建和保存新工具。
  • 📁 恢复文件夹知识处理：修复了一个错误，即上传到文件夹和模型知识库的文件在’Max Upload Count’设置为空时未被提取或分析用于检索增强生成，确保了文档处理和知识增强的无缝进行。
  • 🧠 识别自定义模型知识库更新：修复了一个错误，即链接到知识库的自定义模型无法自动识别添加到这些知识库的新文件。模型现在正确地整合来自更新知识集合的最新信息。
  • 📦 全面的Redis键前缀：更正了硬编码前缀，确保REDIS_KEY_PREFIX现在在所有WebSocket和任务管理键中得到遵守。这防止了多实例部署中的数据冲突，并提高了与Redis集群模式的兼容性。
  • ✨ 更具描述性的OpenAI路由器错误：OpenAI兼容的API路由器现在传播详细的上游错误消息，而不是返回通用的“Bad Request”。这为开发人员和API用户提供了清晰、可操作的反馈，使调试和解决模型请求问题变得显著更容易。
  • 🔐 强化OIDC注销流程：OpenID Connect注销流程现在在尝试与其通信之前验证’OPENID_PROVIDER_URL’是否已配置，防止潜在错误并确保更可靠的注销体验。
  • 🍓 恢复树莓派兼容性：将pyarrow库固定在20.0.0版本，解决了基于ARM的设备（如树莓派）上的“非法指令”崩溃问题，确保在此硬件上稳定运行。
  • 📁 恢复文件夹系统提示变量：修复了一个错误，即提示变量（例如，’{{CURRENT_DATETIME}}’）未在文件夹级别的系统提示中渲染。这恢复了一个重要功能，用于为项目文件夹中的所有聊天创建动态、上下文相关的指令。
  • 📝 修复知识检索中的笔记访问：更正了知识检索中的权限疏忽，确保用户始终可以将自己的笔记用作RAG的来源，而无需显式共享权限。
  • 🤖 GPT-5模型的标题生成兼容性：在有效负载处理程序中添加了对’gpt-5’模型的支持，它正确地将已弃用的’max_tokens’参数转换为’max_completion_tokens’。这解决了标题生成失败问题，并确保与最新一代模型无缝运行。
  • ⚃ 流式响应中的正确API ‘finish_reason’：修复了一个问题，即流式API响应中的中间’reasoning_content’块错误地报告了’finish_reason’为’stop’。这些块的’finish_reason’现在正确设置为’null’，确保与依赖此字段的第三方应用程序兼容。
  • 📈 评估页面稳定性：解决了在处理缺少’rating’字段的旧反馈条目时，排行榜和反馈页面崩溃的问题。系统现在可以优雅地处理这些旧数据，确保所有用户都能可靠地加载这两个页面。
  • 🤝 可靠的协作会话清理：修复了实时协作引擎中的一个异步错误，该错误阻止了在所有用户离开后正确清理文档会话。这确保了协作笔记等功能具有更高的稳定性和资源管理。
  • 🧠 增强的记忆稳定性与安全性：重构了记忆更新和删除操作，以严格执行用户所有权，防止潜在的数据完整性问题。此外，改进了记忆查询的错误处理，当记忆不存在时提供更清晰的反馈。
  • 🧑‍⚖️ 恢复管理员访问用户反馈的权限：修复了一个阻止管理员查看或编辑他们未创建的反馈的权限问题，确保他们可以正确管理整个平台上的所有评估。
  • 🔐 PGVector加密元数据的修复：更正了实验性’PGVECTOR_PGCRYPTO’功能中的一个SQL语法错误，该错误阻止了加密元数据的保存。现在可以按预期将文档上传到加密的PGVector集合中。
  • 🔍 恢复Serply Web搜索集成：修复了传递给Serply网页搜索引擎的参数不正确的问题，恢复了其在RAG和网页搜索工作流中的功能。
  • 🔍 弹性网页搜索处理：网页搜索检索现在可以优雅地处理缺少’snippet’的搜索结果，防止崩溃，并确保即使在搜索引擎返回不完整数据时，RAG工作流也能成功完成。
  • 🖼️ 富文本输入中表格粘贴正确显示：修复了将表格文本粘贴到富文本输入框中时错误地将其显示为代码的问题。表格现在按预期正确渲染，改进了内容格式和用户体验。
  • ✍️ 富文本输入TypeError解决：通过重构从’RichTextInput.svelte’访问和引用’getWordAtDocPos’函数的方式，解决了’MessageInput.svelte’中潜在的’TypeError: ue.getWordAtDocPos is not a function’，确保了稳定的富文本输入行为，尤其是在生产环境重启后。
  • ✏️ 恢复聊天中手动创建代码块的功能：修复了一个问题，即在启用“按回车键发送”模式时，输入三个反引号然后按Shift+Enter会错误地移除反引号。这确保用户可以可靠地手动创建多行代码块。
  • 🎨 一致的深色模式背景：修复了一个问题，即在深色模式下页面加载和刷新时，应用程序背景可能错误地闪烁或保持白色，确保无缝且一致的视觉体验。
  • 🎨 “Her”主题渲染修复：修复了一个导致“Her”主题在某些情况下错误地渲染为深色主题的错误。该主题现在可靠地在所有会话中应用其预期的浅色外观。
  • 📜 修正Markdown表格换行渲染：修复了一个问题，即Markdown表格中的换行符（’
    ’）显示为原始HTML，而不是正确渲染。这确保包含多行单元格内容的表格现在按预期显示。
  • 🚦 修正待批准用户的配置：修复了一个问题，即等待批准的用户可能错误地加载完整应用程序界面，导致UI混乱或损坏。这确保只有完全批准的用户才能获得标准的app ‘config’，从而提供更顺畅、更可靠的入门体验。
  • 🔄 聊天克隆现在包含标签、文件夹状态和固定状态：克隆聊天或共享聊天时，其关联的标签、文件夹组织和固定状态现在被正确复制，确保一致的聊天管理。
  • ⚙️ 增强后端可靠性：解决了引用已删除笔记时知识库检索的潜在崩溃问题。此外，重构了聊天处理，确保模型信息更可靠地保存，增强了整体系统稳定性。
  • ⚃ 浮动“提问/解释”模态框稳定性：修复了一个问题，即在模型正在生成响应时导航离开浮动“提问”或“解释”模态框，会向控制台发送大量错误。进行中的请求现在被正确取消，提高了应用程序稳定性。
  • ⚡ 优化用户计数检查：通过将资源密集型的“COUNT”查询替换为更高效的“EXISTS”查询，改进了整个应用程序中用户计数和存在性检查的性能，降低了数据库负载。
  • 🔐 强化OpenTelemetry导出器配置：OTLP HTTP导出器不再使用可能不安全的显式标志，通过依赖连接URL的协议（HTTP/HTTPS）来确保传输安全，提高了安全性。
  • 📱 修复移动端用户菜单关闭行为：解决了在移动设备上选择选项后用户菜单保持打开的问题，确保菜单正确关闭并将焦点返回主界面，提供更顺畅的移动体验。
  • 📱 修复移动端入门页面显示：解决了入门页面上的按钮在某些移动浏览器上不一致可见的问题，确保在所有设备上提供功能完整、一致的用户体验。
  • ↕️ 改进的固定模型拖放行为：重新排列固定模型的拖放功能现在在移动设备上被显式禁用，确保更好的可用性并防止潜在的UI冲突或意外行为。
  • 📱 PWA旋转行为已纠正：渐进式Web应用现在正确遵守设备的屏幕方向锁定，防止不必要的旋转，确保更原生的移动体验。
  • ✏️ 改进的聊天标题编辑行为：当用户点击别处或按Enter键时，对聊天标题的更改现在会可靠地保存，取代了可能意外丢弃编辑的、不那么直观的行为。这使重命名聊天变得更顺畅、更可预测。
  • ✏️ 提示命令允许下划线：修复了提示命令的验证，以正确允许使用下划线（’_’），与文档示例保持一致，提高了自定义提示命名的灵活性。
  • 💡 标题生成按钮行为修复：解决了一个问题，即在编辑聊天或笔记标题时点击“生成标题”按钮，会在生成开始前错误地保存标题。焦点现在被正确管理，确保流畅、可预测的用户体验。
  • ✏️ 一致的聊天输入高度：修复了一个小的视觉错误，即切换“聊天富文本输入”设置时，聊天输入框的高度会发生轻微变化，确保更稳定、一致的布局。
  • 🙈 管理员界面切换稳定性：修复了管理员设置中的一个视觉故障，即页面加载时开关可能短暂显示不正确的状态，确保UI始终准确反映保存的设置。
  • 🙈 社区共享按钮可见性：当管理员设置中禁用启用社区共享功能时，反馈页面上的“共享到社区”按钮现在会正确隐藏，确保UI遵守配置的共享策略。
  • 🙈 “帮助我们翻译”链接可见性：设置中的“帮助我们翻译”链接现在在特定许可证配置的部署中正确隐藏，为企业用户提供更简洁的界面。
  • 🔗 稳健的工具服务器URL处理：修复了一个问题，即为工具服务器的OpenAPI规范提供完整URL会导致无效路径。系统现在正确处理绝对URL和相对路径，提高了配置灵活性。
  • 🔧 改进的Azure URL检测：识别Azure OpenAI端点的逻辑已变得更加稳健，确保所有有效的Azure URL现在都被正确检测，以便更顺畅地建立连接。
  • ⚃ 修正直接连接保存逻辑：通过移除“直接连接”的冗余保存操作，修复了管理员连接设置页面中的一个错误，导致更新设置时行为更可靠、更可预测。
  • 🔗 修正“发现”链接：模型、提示词、工具和函数的“发现”链接现在指向它们在openwebui.com上的特定相关页面，改进了用户的内容发现。
  • ⏱️ 细化AI思考时间的显示：调整了AI思考（推理）时间的显示逻辑，以更准确地将非常短的思考时间显示为“不到一秒”，提高了AI过程反馈的清晰度。
  • 📜 Markdown换行渲染优化：改进了Markdown渲染中的换行处理，以获得更好的视觉一致性。
  • 🛠️ 修正OpenTelemetry Docker Compose示例：通过移除重复项、添加必要的环境变量和强化安全设置，修复并增强了docker-compose.otel.yaml文件，确保开箱即用的可观测性设置更可靠。
  • 🛠️ 开发脚本CORS修复：通过移除尾部斜杠，更正了本地开发脚本（dev.sh）中的CORS源URL，确保为开发人员提供更可靠、更一致的设置。
  • ⬆️ OpenTelemetry库更新：将所有OpenTelemetry相关库升级到最新版本，确保可观测性具有更好的性能、稳定性和兼容性。

变更

  • ❗ Docling集成升级至v1 API（破坏性更改）：与Docling文档处理引擎的集成已更新到其新的稳定版’/v1’ API。这是为了与Docling 1.0.0及更高版本兼容所必需的。因此，不再支持旧版本的Docling。依赖Docling进行文档摄入的用户必须升级其docling-serve实例以确保功能持续可用。
  • 🗣️ 管理员优先的Whisper语言设置：全局WHISPER_LANGUAGE设置现在作为音频转录的严格覆盖。如果设置，它将用于所有语音转录任务，忽略用户在每次请求中指定的任何语言。这使管理员能更好地控制转录一致性。
  • ✂️ 移除Datalab Marker API语言选择：已移除Datalab Marker API的单独语言选择选项，因为其功能现在已集成并被更全面的’additional_config’参数取代。用户应过渡使用’additional_config’进行相关的语言和处理设置。
  • 📄 文档和发布链接可见性：用户菜单中的“文档”和“发布”链接现在仅对管理员用户可见，为非管理员角色简化了用户界面。

[!TIP] 寻找企业版计划？ – 立即与我们的销售团队联系！

获取增强功能，包括自定义主题和品牌、服务水平协议支持、长期支持版本以及更多！

👏 非常感谢我们了不起的贡献者们的辛勤工作和奉献精神，使这个版本得以实现： @silentoplayz, @andrewbbaek, @psy42a, @expruc, @Ithanil, @17jmumford, @dependabot[bot], @Classic298, @Rain6435, @Jakobu5, @Ithanil, @rndmcnlly, @hiwylee, @ahangarha, @gaby, @hadadrjt, @leebeomhun

赞助商 🙌

🚀 我们想向我们出色的赞助商们表达衷心的感谢，感谢他们的慷慨支持（注：我们已将私人赞助商从此列表中排除。如果您希望在此出现，请随时联系我们！）

翡翠赞助商

  • Warp warp
  • Tailscale tailscale

@amd, @agency-agency, @roosi-gmbh, @feddersen-group, @SRKConsulting, @Poxel2, @mushmax, @rndmcnlly, @icanbwell, @Krambu-Travis, @loitragg, @GenieDev101, @CubicleGangster, @GitDatamind, @TOWaB, @TheZigGroup, @Classic298, @FominVO, @catalyst-net-nz, @JrdnHans, @hkosm, @VeroFess

更新内容 （原始）

[0.6.19] - 2025-08-09

Added

  • ✨ Modernized Sidebar and Major UI Refinements: The main navigation sidebar has been completely redesigned with a modern, cleaner aesthetic, featuring a sticky header and footer to keep key controls accessible. Core sidebar logic, like the pinned models list, was also refactored into dedicated components for better performance and maintainability.
  • 🪄 Guided Response Regeneration: The “Regenerate” button has been transformed into a powerful new menu. You can now guide the AI’s next attempt by suggesting changes in a text prompt, or use one-click options like “Try Again,” “Add Details,” or “More Concise” to instantly refine and reshape the response to better fit your needs.
  • 🛠️ Improved Tool Call Handling for GPT-OSS Models: Implemented robust handling for tool calls specifically for GPT-OSS models, ensuring proper function execution and integration.
  • 🛑 Stop Button for Merge Responses: Added a dedicated stop button to immediately halt the generation of merged AI responses, providing users with more control over ongoing outputs.
  • 🔄 Experimental SCIM 2.0 Support: Implemented SCIM 2.0 (System for Cross-domain Identity Management) protocol support, enabling enterprise-grade automated user and group provisioning from identity providers like Okta, Azure AD, and Google Workspace for seamless user lifecycle management. Configuration is managed securely via environment variables.
  • 🗂️ Amazon S3 Vector Support: You can now use Amazon S3 Vector as a high-performance vector database for your Retrieval-Augmented Generation (RAG) workflows. This provides a scalable, cloud-native storage option for users deeply integrated into the AWS ecosystem, simplifying infrastructure and enabling enterprise-scale knowledge management.
  • 🗄️ Oracle 23ai Vector Search Support: Added support for Oracle 23ai’s new vector search capabilities as a supported vector database, providing a robust and scalable option for managing large-scale documents and integrating vector search with existing business data at the database level.
  • ⚡ Qdrant Performance and Configuration Enhancements: The Qdrant client has been significantly improved with faster data retrieval logic for ‘get’ and ‘query’ operations. New environment variables (‘QDRANT_TIMEOUT’, ‘QDRANT_HNSW_M’) provide administrators with finer control over query timeouts and HNSW index parameters, enabling better performance tuning for large-scale deployments.
  • 🔐 Encrypted SQLite Database with SQLCipher: You can now encrypt your entire SQLite database at rest using SQLCipher. By setting the ‘DATABASE_TYPE’ to ‘sqlite+sqlcipher’ and providing a ‘DATABASE_PASSWORD’, all data is transparently encrypted, providing an essential security layer for protecting sensitive information in self-hosted deployments. Note that this requires additional system libraries and the ‘sqlcipher3-wheels’ Python package.
  • 🚀 Efficient Redis Connection Management: Implemented a shared connection pool cache to reuse Redis connections, dramatically reducing the number of active clients. This prevents connection exhaustion errors, improves performance, and ensures greater stability in high-concurrency deployments and those using Redis Sentinel.
  • ⚡ Batched Response Streaming for High Performance: Dramatically improve performance and stability during high-speed response streaming by batching multiple tokens together before sending them to the client. A new ‘Stream Delta Chunk Size’ advanced parameter can be set per-model or in user/chat settings, significantly reducing CPU load on the server, Redis, and client, and preventing connection issues in high-concurrency environments.
  • ⚙️ Global Batched Streaming Configuration: Administrators can now set a system-wide default for response streaming using the new ‘CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE’ environment variable. This allows for global performance tuning, while still letting per-model or per-chat settings override the default for more granular control.
  • 🔎 Advanced Chat Search with Status Filters: Quickly find any conversation with powerful new search filters. You can now instantly narrow down your chats using prefixes like ‘pinned:true’, ‘shared:true’, and ‘archived:true’ directly in the search bar. An intelligent dropdown menu assists you by suggesting available filter options as you type, streamlining your workflow and making chat management more efficient than ever.
  • 🛂 Granular Chat Controls Permissions: Administrators can now manage chat settings with greater detail. The main “Chat Controls” permission now acts as a master switch, while new granular toggles for “Valves”, “System Prompts”, and “Advanced Parameters” allow for more specific control over which sections are visible to users inside the panel.
  • ✍️ Formatting Toolbar for Chat Input: Introduced a dedicated formatting toolbar for the rich text chat input field, providing users with more accessible options for text styling and editing, configurable via interface settings.
  • 📑 Tabbed View for Multi-Model Responses: You can now enable a new tabbed interface to view responses from multiple models. Instead of side-scrolling cards, this compact view organizes each model’s response into its own tab, making it easier to compare outputs and saving vertical space. This feature can be toggled on or off in Interface settings.
  • ↕️ Reorder Pinned Models via Drag-and-Drop: You can now organize your pinned models in the sidebar by simply dragging and dropping them into your preferred order. This custom layout is saved automatically, giving you more flexible control over your workspace.
  • 📌 Quick Model Unpin Shortcut: You can now quickly unpin a model by holding the Shift key and hovering over it to reveal an instant unpin button, streamlining your workspace customization.
  • ⚡ Improved Chat Input Performance: The chat input is now significantly more responsive, especially when pasting or typing large amounts of text. This was achieved by implementing a debounce mechanism for the auto-save feature, which prevents UI lag and ensures a smooth, uninterrupted typing experience.
  • ✍️ Customizable Floating Quick Actions with Tool Support: Take full control of your text interaction workflow with new customizable floating quick actions. In Settings, you can create, edit, or disable these actions and even integrate tools using the ‘{{TOOL:tool_id}}’ syntax in your prompts, enabling powerful one-click automations on selected text. This is in addition to using placeholders like ‘{{CONTENT}}’ and ‘{{INPUT_CONTENT}}’ for custom text transformations.
  • 🔒 Admin Workspace Privacy Control: Introduced the ‘ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS’ environment variable (defaults to ‘True’) allowing administrators to control their access privileges to workspace items (Knowledge, Models, Prompts, Tools). When disabled, administrators adhere to the same access control rules as regular users, enhancing data separation for multi-tenant deployments.
  • 🗄️ Comprehensive Model Configuration Management: Administrators can now export the entire model configuration to a file and use a new declarative sync endpoint to manage models in bulk. This powerful feature enables seamless backups, migrations, and state replication across multiple instances.
  • 📦 Native Redis Cluster Mode Support: Added full support for connecting to Redis in cluster mode, allowing for scalable and highly available Redis deployments beyond Sentinel-managed setups. New environment variables ‘REDIS_CLUSTER’ and ‘WEBSOCKET_REDIS_CLUSTER’ enable the use of ‘redis.cluster.RedisCluster’ clients.
  • 📊 Granular OpenTelemetry Metrics Configuration: Introduced dedicated environment variables and enhanced configuration options for OpenTelemetry metrics, allowing for separate OTLP endpoints, basic authentication credentials, and protocol (HTTP/gRPC) specifically for metrics export, independent of trace settings. This provides greater flexibility for integrating with diverse observability stacks.
  • 🪵 Granular OpenTelemetry Logging Configuration: Enhanced the OpenTelemetry logging integration by introducing dedicated environment variables for logs, allowing separate OTLP endpoints, basic authentication credentials, and protocol (HTTP/gRPC) specifically for log export, independent of general OTel settings. The application’s default Python logger now leverages this configuration to automatically send logs to your OTel endpoint when enabled via ‘ENABLE_OTEL_LOGS’.
  • 📁 Enhanced Folder Chat Management with Sorting and Time Blocks: The chat list within folders now supports comprehensive sorting options by title and updated time, along with intelligent time-based grouping (e.g., “Today,” “Yesterday”) similar to the main chat view, making navigation and organization of project-specific conversations significantly easier.
  • ⚙️ Configurable Datalab Marker API & Advanced Processing Options: Enhanced Datalab Marker API integration, allowing administrators to configure custom API base URLs for self-hosting and to specify comprehensive processing options via a new ‘additional_config’ JSON parameter. This replaces the deprecated language selection feature and provides granular control over document extraction, with streamlined API endpoint resolution for more robust self-hosted deployments.
  • 🧑‍💼 Export All Users to CSV: Administrators can now export a complete list of all users to a CSV file directly from the Admin Panel’s database settings. This provides a simple, one-click way to generate user data for auditing, reporting, or management purposes.
  • 🛂 Customizable OAuth ‘sub’ Claim: Administrators can now use the ‘OAUTH_SUB_CLAIM_OVERRIDE’ environment variable to specify which claim from the identity provider should be used as the unique user identifier (‘sub’). This provides greater flexibility and control for complex enterprise authentication setups where modifying the IDP’s default claims is not possible.
  • 👁️ Password Visibility Toggle for Input Fields: Password fields across the application (login, registration, user management, and account settings) now utilize a new ‘SensitiveInput’ component, providing a consistent toggle to reveal/hide passwords for improved usability and security.
  • 🛂 Optional “Confirm Password” on Sign-Up: To help prevent password typos during account creation, administrators can now enable a “Confirm Password” field on the sign-up page. This feature is disabled by default and can be activated via an environment variable for enhanced user experience.
  • 💬 View Full Chat from User Feedback: Administrators can now easily navigate to the full conversation associated with a user feedback entry directly from the feedback modal, streamlining the review and troubleshooting process.
  • 🎚️ Intuitive Hybrid Search BM25-Weight Slider: The numerical input for the BM25-Weight parameter in Hybrid Search has been replaced with an interactive slider, offering a more intuitive way to adjust the balance between lexical and semantic search. A “Default/Custom” toggle and clearer labels enhance usability and understanding of this key parameter.
  • ⚙️ Enhanced Bulk Function Synchronization: The API endpoint for synchronizing functions has been significantly improved to reliably handle bulk updates. This ensures that importing and managing large libraries of functions is more robust and error-free for administrators.
  • 🖼️ Option to Disable Image Compression in Channels: Introduced a new setting under Interface options to allow users to force-disable image compression specifically for images posted in channels, ensuring higher resolution for critical visual content.
  • 🔗 Custom CORS Scheme Support: Introduced a new environment variable ‘CORS_ALLOW_CUSTOM_SCHEME’ that allows administrators to define custom URL schemes (e.g., ‘app://’) for CORS origins, enabling greater flexibility for local development or desktop client integrations.
  • ♿ Translatable and Accessible Banners: Enhanced banner elements with translatable badge text and proper ARIA attributes (aria-label, aria-hidden) for SVG icons, significantly improving accessibility and screen reader compatibility.
  • ⚠️ OAuth Configuration Warning for Missing OPENID_PROVIDER_URL: Added a proactive startup warning that notifies administrators when OAuth providers (Google, Microsoft, or GitHub) are configured but the essential ‘OPENID_PROVIDER_URL’ environment variable is missing. This prevents silent OAuth logout failures and guides administrators to complete their setup correctly.
  • ♿ Major Accessibility Enhancements: Key parts of the interface have been made significantly more accessible. The user profile menu is now fully navigable via keyboard, essential controls in the Playground now include proper ARIA labels for screen readers, and decorative images have been hidden from assistive technologies to reduce audio clutter. Menu buttons also feature enhanced accessibility with ‘aria-label’, ‘aria-hidden’ for SVGs, and ‘aria-pressed’ for toggle buttons.
  • ⚙️ General Backend Refactoring: Implemented various backend improvements to enhance performance, stability, and security, ensuring a more resilient and reliable platform for all users, including refining logging output to be cleaner and more efficient by conditionally including ’extra_json’ fields and improving consistent metadata handling in vector database operations, and laying preliminary scaffolding for future analytics features.
  • 🌐 Localization & Internationalization Improvements: Refined and expanded translations for Catalan, Danish, Korean, Persian, Polish, Simplified Chinese, and Spanish, ensuring a more fluent and native experience for global users across all supported languages.

Fixed

  • 🛡️ Hardened Channel Message Security: Fixed a key permission flaw that allowed users with channel access to edit or delete messages belonging to others. The system now correctly enforces that users can only modify their own messages, protecting data integrity in shared channels.
  • 🛡️ Hardened OAuth Security by Removing JWT from URL: Fixed a critical security vulnerability where the authentication token was exposed in the URL after a successful OAuth login. The token is now transferred via a browser cookie, preventing potential leaks through browser history or server logs and protecting user sessions.
  • 🛡️ Hardened Chat Completion API Security: The chat completion API endpoint now includes an explicit ownership check, ensuring non-admin users cannot access chats that do not belong to them and preventing potential unauthorized access.
  • 🛠️ Resilient Model Loading: Fixed an issue where a failure in loading the model list (e.g., from a misconfigured provider) would prevent the entire user interface, including the admin panel, from loading. The application now gracefully handles these errors, ensuring the UI remains accessible.
  • 🔒 Resolved FIPS Self-Test Failure: Fixed a critical issue that prevented Open WebUI from running on FIPS-compliant systems, specifically resolving the “FATAL FIPS SELFTEST FAILURE” error related to OpenSSL and SentenceTransformers, restoring compatibility with secure environments.
  • 📦 Redis Cluster Connection Restored: Fixed an issue where the backend was unable to connect to Redis in cluster mode, now ensuring seamless integration with scalable Redis cluster deployments.
  • 📦 PGVector Connection Stability: Fixed an issue where read-only operations could leave database transactions idle, preventing potential connection errors and improving overall database stability and resource management.
  • 🛠️ OpenAPI Tool Integration for Array Parameters Fixed: Resolved a critical bug where external tools using array parameters (e.g., for tags) would fail when used with OpenAI models. The system now correctly generates the required ‘items’ property in the function schema, restoring functionality and preventing ‘400 Bad Request’ errors.
  • 🛠️ Tool Creation for Users Restored: Fixed a bug in the code editor where status messages were incorrectly prepended to tool scripts, causing a syntax error upon saving. All authorized users can now reliably create and save new tools.
  • 📁 Folder Knowledge Processing Restored: Fixed a bug where files uploaded to folder and model knowledge bases were not being extracted or analyzed for Retrieval-Augmented Generation (RAG) when the ‘Max Upload Count’ setting was empty, ensuring seamless document processing and knowledge augmentation.
  • 🧠 Custom Model Knowledge Base Updates Recognized: Fixed a bug where custom models linked to to knowledge bases did not automatically recognize newly added files to those knowledge bases. Models now correctly incorporate the latest information from updated knowledge collections.
  • 📦 Comprehensive Redis Key Prefixing: Corrected hardcoded prefixes to ensure the REDIS_KEY_PREFIX is now respected across all WebSocket and task management keys. This prevents data collisions in multi-instance deployments and improves compatibility with Redis cluster mode.
  • ✨ More Descriptive OpenAI Router Errors: The OpenAI-compatible API router now propagates detailed upstream error messages instead of returning a generic ‘Bad Request’. This provides clear, actionable feedback for developers and API users, making it significantly easier to debug and resolve issues with model requests.
  • 🔐 Hardened OIDC Signout Flow: The OpenID Connect signout process now verifies that the ‘OPENID_PROVIDER_URL’ is configured before attempting to communicate with it, preventing potential errors and ensuring a more reliable logout experience.
  • 🍓 Raspberry Pi Compatibility Restored: Pinned the pyarrow library to version 20.0.0, resolving an “Illegal Instruction” crash on ARM-based devices like the Raspberry Pi and ensuring stable operation on this hardware.
  • 📁 Folder System Prompt Variables Restored: Fixed a bug where prompt variables (e.g., ‘{{CURRENT_DATETIME}}’) were not being rendered in Folder-level System Prompts. This restores an important capability for creating dynamic, context-aware instructions for all chats within a project folder.
  • 📝 Note Access in Knowledge Retrieval Fixed: Corrected a permission oversight in knowledge retrieval, ensuring users can always use their own notes as a source for RAG without needing explicit sharing permissions.
  • 🤖 Title Generation Compatibility for GPT-5 Models: Added support for ‘gpt-5’ models in the payload handler, which correctly converts the deprecated ‘max_tokens’ parameter to ‘max_completion_tokens’. This resolves title generation failures and ensures seamless operation with the latest generation of models.
  • ⚙️ Correct API ‘finish_reason’ in Streaming Responses: Fixed an issue where intermediate ‘reasoning_content’ chunks in streaming API responses incorrectly reported a ‘finish_reason’ of ‘stop’. The ‘finish_reason’ is now correctly set to ’null’ for these chunks, ensuring compatibility with third-party applications that rely on this field.
  • 📈 Evaluation Pages Stability: Resolved a crash on the Leaderboard and Feedbacks pages when processing legacy feedback entries that were missing a ‘rating’ field. The system now gracefully handles this older data, ensuring both pages load reliably for all users.
  • 🤝 Reliable Collaborative Session Cleanup: Fixed an asynchronous bug in the real-time collaboration engine that prevented document sessions from being properly cleaned up after all users had left. This ensures greater stability and resource management for features like Collaborative Notes.
  • 🧠 Enhanced Memory Stability and Security: Refactored memory update and delete operations to strictly enforce user ownership, preventing potential data integrity issues. Additionally, improved error handling for memory queries now provides clearer feedback when no memories exists.
  • 🧑‍⚖️ Restored Admin Access to User Feedback: Fixed a permission issue that blocked administrators from viewing or editing user feedback they didn’t create, ensuring they can properly manage all evaluations across the platform.
  • 🔐 PGVector Encryption Fix for Metadata: Corrected a SQL syntax error in the experimental ‘PGVECTOR_PGCRYPTO’ feature that prevented encrypted metadata from being saved. Document uploads to encrypted PGVector collections now work as intended.
  • 🔍 Serply Web Search Integration Restored: Fixed an issue where incorrect parameters were passed to the Serply web search engine, restoring its functionality for RAG and web search workflows.
  • 🔍 Resilient Web Search Processing: Web search retrieval now gracefully handles search results that are missing a ‘snippet’, preventing crashes and ensuring that RAG workflows complete successfully even with incomplete data from search engines.
  • 🖼️ Table Pasting in Rich Text Input Displayed Correctly: Fixed an issue where pasting table text into the rich text input would incorrectly display it as code. Tables are now properly rendered as expected, improving content formatting and user experience.
  • ✍️ Rich Text Input TypeError Resolution: Addressed a potential ‘TypeError: ue.getWordAtDocPos is not a function’ in ‘MessageInput.svelte’ by refactoring how the ‘getWordAtDocPos’ function is accessed and referenced from ‘RichTextInput.svelte’, ensuring stable rich text input behavior, especially after production restarts.
  • ✏️ Manual Code Block Creation in Chat Restored: Fixed an issue where typing three backticks and then pressing Shift+Enter would incorrectly remove the backticks when “Enter to Send” mode was active. This ensures users can reliably create multi-line code blocks manually.
  • 🎨 Consistent Dark Mode Background: Fixed an issue where the application background could incorrectly flash or remain white during page loads and refreshes in dark mode, ensuring a seamless and consistent visual experience.
  • 🎨 ‘Her’ Theme Rendering Fixed: Corrected a bug that caused the “Her” theme to incorrectly render as a dark theme in some situations. The theme now reliably applies its intended light appearance across all sessions.
  • 📜 Corrected Markdown Table Line Break Rendering: Fixed an issue where line breaks (’
    ’) within Markdown tables were displayed as raw HTML instead of being rendered correctly. This ensures that tables with multi-line cell content are now displayed as intended.
  • 🚦 Corrected App Configuration for Pending Users: Fixed an issue where users awaiting approval could incorrectly load the full application interface, leading to a confusing or broken UI. This ensures that only fully approved users receive the standard app ‘config’, resulting in a smoother and more reliable onboarding experience.
  • 🔄 Chat Cloning Now Includes Tags, Folder Status, and Pinned Status: When cloning a chat or shared chat, its associated tags, folder organization, and pinned status are now correctly replicated, ensuring consistent chat management.
  • ⚙️ Enhanced Backend Reliability: Resolved a potential crash in knowledge base retrieval when referencing a deleted note. Additionally, chat processing was refactored to ensure model information is saved more reliably, enhancing overall system stability.
  • ⚙️ Floating ‘Ask/Explain’ Modal Stability: Fixed an issue that spammed the console with errors when navigating away while a model was generating a response in the floating ‘Ask’ or ‘Explain’ modals. In-flight requests are now properly cancelled, improving application stability.
  • ⚡ Optimized User Count Checks: Improved performance for user count and existence checks across the application by replacing resource-intensive ‘COUNT’ queries with more efficient ‘EXISTS’ queries, reducing database load.
  • 🔐 Hardened OpenTelemetry Exporter Configuration: The OTLP HTTP exporter no longer uses a potentially insecure explicit flag, improving security by relying on the connection URL’s protocol (HTTP/HTTPS) to ensure transport safety.
  • 📱 Mobile User Menu Closing Behavior Fixed: Resolved an issue where the user menu would remain open on mobile devices after selecting an option, ensuring the menu correctly closes and returns focus to the main interface for a smoother mobile experience.
  • 📱 OnBoarding Page Display Fixed on Mobile: Resolved an issue where buttons on the OnBoarding page were not consistently visible on certain mobile browsers, ensuring a functional and complete user experience across devices.
  • ↕️ Improved Pinned Models Drag-and-Drop Behavior: The drag-and-drop functionality for reordering pinned models is now explicitly disabled on mobile devices, ensuring better usability and preventing potential UI conflicts or unexpected behavior.
  • 📱 PWA Rotation Behavior Corrected: The Progressive Web App now correctly respects the device’s screen orientation lock, preventing unwanted rotation and ensuring a more native mobile experience.
  • ✏️ Improved Chat Title Editing Behavior: Changes to a chat title are now reliably saved when the user clicks away or presses Enter, replacing a less intuitive behavior that could accidentally discard edits. This makes renaming chats a smoother and more predictable experience.
  • ✏️ Underscores Allowed in Prompt Commands: Fixed the validation for prompt commands to correctly allow the use of underscores (’_’), aligning with documentation examples and improving flexibility in naming custom prompts.
  • 💡 Title Generation Button Behavior Fixed: Resolved an issue where clicking the “Generate Title” button while editing a chat or note title would incorrectly save the title before generation could start. The focus is now managed correctly, ensuring a smooth and predictable user experience.
  • ✏️ Consistent Chat Input Height: Fixed a minor visual bug where the chat input field’s height would change slightly when toggling the “Rich Text Input for Chat” setting, ensuring a more stable and consistent layout.
  • 🙈 Admin UI Toggle Stability: Fixed a visual glitch in the Admin settings where toggle switches could briefly display an incorrect state on page load, ensuring the UI always accurately reflects the saved settings.
  • 🙈 Community Sharing Button Visibility: The “Share to Community” button on the feedback page is now correctly hidden when the Enable Community Sharing feature is disabled in the admin settings, ensuring the UI respects the configured sharing policy.
  • 🙈 “Help Us Translate” Link Visibility: The “Help us translate” link in settings is now correctly hidden in deployments with specific license configurations, ensuring a cleaner interface for enterprise users.
  • 🔗 Robust Tool Server URL Handling: Fixed an issue where providing a full URL for a tool server’s OpenAPI specification resulted in an invalid path. The system now correctly handles both absolute URLs and relative paths, improving configuration flexibility.
  • 🔧 Improved Azure URL Detection: The logic for identifying Azure OpenAI endpoints has been made more robust, ensuring all valid Azure URLs are now correctly detected for a smoother connection setup.
  • ⚙️ Corrected Direct Connection Save Logic: Fixed a bug in the Admin Connections settings page by removing a redundant save action for ‘Direct Connections’, leading to more reliable and predictable behavior when updating settings.
  • 🔗 Corrected “Discover” Links: The “Discover” links for models, prompts, tools, and functions now point to their specific, relevant pages on openwebui.com, improving content discovery for users.
  • ⏱️ Refined Display of AI Thought Duration: Adjusted the display logic for AI thought (reasoning) durations to more accurately show very short thought times as “less than a second,” improving clarity in AI process feedback.
  • 📜 Markdown Line Break Rendering Refinement: Improved handling of line breaks within Markdown rendering for better visual consistency.
  • 🛠️ Corrected OpenTelemetry Docker Compose Example: The docker-compose.otel.yaml file has been fixed and enhanced by removing duplicates, adding necessary environment variables, and hardening security settings, ensuring a more reliable out-of-box observability setup.
  • 🛠️ Development Script CORS Fix: Corrected the CORS origin URL in the local development script (dev.sh) by removing the trailing slash, ensuring a more reliable and consistent setup for developers.
  • ⬆️ OpenTelemetry Libraries Updated: Upgraded all OpenTelemetry-related libraries to their latest versions, ensuring better performance, stability, and compatibility for observability.

Changed

  • ❗ Docling Integration Upgraded to v1 API (Breaking Change): The integration with the Docling document processing engine has been updated to its new, stable ‘/v1’ API. This is required for compatibility with Docling version 1.0.0 and newer. As a result, older versions of Docling are no longer supported. Users who rely on Docling for document ingestion must upgrade their docling-serve instance to ensure continued functionality.
  • 🗣️ Admin-First Whisper Language Priority: The global WHISPER_LANGUAGE setting now acts as a strict override for audio transcriptions. If set, it will be used for all speech-to-text tasks, ignoring any language specified by the user on a per-request basis. This gives administrators more control over transcription consistency.
  • ✂️ Datalab Marker API Language Selection Removed: The separate language selection option for the Datalab Marker API has been removed, as its functionality is now integrated and superseded by the more comprehensive ‘additional_config’ parameter. Users should transition to using ‘additional_config’ for relevant language and processing settings.
  • 📄 Documentation and Releases Links Visibility: The “Documentation” and “Releases” links in the user menu are now visible only to admin users, streamlining the user interface for non-admin roles.

[!TIP]
Looking for an Enterprise Plan? – Speak with Our Sales Team Today!

Get enhanced capabilities, including custom theming and branding, Service Level Agreement (SLA) support, Long-Term Support (LTS) versions, and more!

👏 Massive thanks to our incredible contributors for their hard work and dedication to making this release possible: @silentoplayz, @andrewbbaek, @psy42a, @expruc, @Ithanil, @17jmumford, @dependabot[bot], @Classic298, @Rain6435, @Jakobu5, @Ithanil, @rndmcnlly, @hiwylee, @ahangarha, @gaby, @hadadrjt, @leebeomhun

Sponsors 🙌

🚀 We’d like to extend a heartfelt thank you to our amazing sponsors for their generous support (Note: We’ve excluded private sponsors from this list. If you’d like to get featured here, feel free to reach out to us!)

Emerald

  • Warp warp
  • Tailscale tailscale

@amd, @agency-agency, @roosi-gmbh, @feddersen-group, @SRKConsulting, @Poxel2, @mushmax, @rndmcnlly, @icanbwell, @Krambu-Travis, @loitragg, @GenieDev101, @CubicleGangster, @GitDatamind, @TOWaB, @TheZigGroup, @Classic298, @FominVO, @catalyst-net-nz, @JrdnHans, @hkosm, @VeroFess

下载链接