发布日期: 2025-06-10
版本号: v0.6.14

Open WebUI 0.6.14版本于2025年6月10日发布，引入了多项新功能并修复了若干问题。新增功能包括：在每次消息后自动生成“跟进”建议以提升工作效率；提供与OpenAI兼容的嵌入端点，便于集成；允许将常用模型固定至侧边栏以便快速访问；为模型选择器中的每个项目增加了菜单以支持固定和复制链接；在多副本集群部署中可靠地停止进行中的聊天；支持Ollama模型的“思考”参数；允许自定义Docling Loader的图像描述模式；为设置模态框的每个标签页添加了直接链接；支持在聊天中直接嵌入和播放音频；支持通过“WEBUI_SECRET_KEY_FILE”进行更安全的密钥管理；克隆的工作区提示词会清晰标注“（克隆）”和ID；更新用户角色时会弹出专用的编辑模态框；优化了代码解释器生成的图像的处理与存储；对Pinecone和向量搜索进行了多项优化以提升性能和可靠性；统一了Ollama的高级参数（如keep_alive和format）；添加了支持CUDA 12.6的Docker镜像以适配更多NVIDIA GPU；实验性地支持了pgvector表的PGVector数据加密；提升了界面的无障碍访问性，包括为按钮和控件添加标签；扩展了高对比度模式的应用范围；改善了工具提示和翻译的清晰度；大幅改进了多种语言的本地化翻译；以及通过优化后端例程提升了整体稳定性和安全性。修复方面解决了反馈分数显示溢出、管理员设置中模型编辑未立即生效、新聊天中已分配工具未更新、文档设置在用户操作前意外保存、旧版iOS设备语音录制失效、受信任邮件头会话安全验证、邮件不匹配时用户会话注销、通用错误处理和内容验证，以及编辑聊天标题时的错误反馈等问题。该版本还特别感谢了众多贡献者和支持的赞助商。

更新内容 （中文）

[0.6.14] - 2025-06-10

新增

  • 🤖 自动“后续问题”建议：Open WebUI 现在能在每次发送消息后智能地生成可操作的“后续问题”建议，帮助您在不打断思路的情况下保持高效和获得灵感；如果您偏好无干扰的体验，始终可以在设置中禁用此功能。
  • 🧩 OpenAI 兼容嵌入端点：引入了完全符合 OpenAI 风格的 /api/embeddings 端点——现在您可以轻松接入 OpenAI 风格的嵌入工作流，使与外部工具和平台的集成变得无缝且熟悉。
  • ↗️ 模型快速固定访问：将您最喜欢或最常用的模型固定到侧边栏，以便即时选择——无需再滚动浏览冗长的模型列表；您常用的模型始终可见，随时可快速访问。
  • 📌 选择器模型项目菜单：选择器中的每个模型现在都有一个菜单，您可以轻松地固定/取消固定到侧边栏并复制直接链接——简化协作，即使在最繁忙的环境中也能保持有序。
  • 🛑 在多副本设置中可靠地停止进行中的聊天：在集群部署中，停止或取消正在进行的聊天现在也能可靠工作——确保每个用户都能随时中断 AI 输出，无论您的规模如何。
  • 🧠 Ollama 模型的 ’think’ 参数：利用新的 Ollama ’think’ 参数支持——让您能更高级地控制 AI 推理过程，并针对您的特定用例进一步调整模型行为。
  • 💬 Docling 的图片描述模式：自定义 Docling Loader 如何描述/提取图片，为您的文档管道提供更智能、更详细且符合工作流需求的图片理解能力。
  • 🛠 设置模态窗口深度链接：设置中的每个选项卡现在都有自己的路由——使直接导航和分享精确设置变得更快、更直观。
  • 🎤 音频 HTML 组件令牌：轻松在聊天中嵌入和播放音频，改进基于语音的工作流，使音频内容从任何对话中都能即时访问和管理。
  • 🔑 支持密钥文件：现在您可以指定 WEBUI_SECRET_KEY_FILE 以实现更安全、更灵活的密钥管理——适用于高级部署和更严格的安全标准。
  • 💡 克隆提示词时的明确标识：克隆的工作区提示词现在会清晰标记为“（克隆）”，ID 会附加“-clone”，保持您的提示词库井然有序，防止意外覆盖。
  • 📝 专用的用户角色编辑模态窗口：更新用户角色现在会可靠地打开一个专用的用户编辑模态窗口，而不是循环切换角色——使管理团队权限更安全、更清晰。
  • 🏞️ 更好地处理和存储解释器生成的图片：代码解释器生成的图片现在集中存储，并可从数据库或云存储可靠加载，确保您的工件始终可用。
  • 🚀 Pinecone 与向量搜索优化：应用了来自 Pinecone 的最新最佳实践，实现更智能的超时、智能重试控制、改进的连接池、更快的 DNS 和并发批处理——为您提供更可靠、更快的文档搜索和 RAG 性能，无需手动调整。
  • ⚙️ Ollama 高级参数统一：keep_alive 和 format 选项现在集成到高级参数部分——从模型编辑器中即可编辑所有内容，实现灵活的模型控制。
  • 🛠 CUDA 12.6 Docker 镜像支持：通过新的 cuda126 镜像，支持部署到计算能力为 7.0 及以下的 NVIDIA GPU（例如 V100、GTX1080）——拓宽了您可扩展 AI 工作负载的硬件选择。
  • 🔒 实验性表级 PGVector 数据加密：激活 pgcrypto 加密支持以保护 pgvector 向量搜索表的内容，为组织提供增强的合规性和数据保护——非常适合企业或受监管环境。
  • 👁 界面无障碍功能升级：聊天按钮和关闭控件现在经过标签化和结构优化，以获得最佳的无障碍支持，确保使用辅助技术时操作更流畅。
  • 🎨 高对比度模式扩展：高对比度无障碍模式现在也适用于菜单项、选项卡和搜索输入框，为所有用户提供更易读的体验。
  • 🛠 工具提示和翻译清晰度：改进了翻译和工具提示的清晰度，特别是在单选按钮上，使界面更容易被所有用户理解。
  • 🔠 全局本地化和翻译改进：对繁体中文、简体中文、希伯来语、俄语、爱尔兰语、德语和丹麦语翻译包进行了大量升级——使平台对全球更多用户感觉更原生、更直观。
  • ⚡ 通用后端稳定性和安全增强：改进了众多后端程序，以减少内存使用、提高性能并简化与外部 API 的集成——使整个平台在日常工作中更健壮、更安全。

修复

  • 🏷 改进反馈分数显示：解决了反馈分数的溢出和可见性问题，使评估更易读、更无障碍。
  • 🗂 管理员设置中的模型编辑立即生效：在管理员设置的模型编辑器中所做的更改现在会立即生效，消除了模型管理中的混淆。
  • 🛠 新聊天中分配的工具即时更新：分配了特定工具的模型现在能在每个新聊天中一致地更新和可用——使工具工作流更可预测、更健壮。
  • 🛠 文档设置仅在用户操作时保存：文档设置现在仅在您按下保存按钮时保存，减少意外更改，确保更大的控制权。
  • 🔊 恢复在旧款 iOS 设备上的语音录制：语音输入现在在旧款 iOS 设备上完全可用，让所有用户都能使用语音工作流。
  • 🔒 可信邮件头会话安全：用户会话现在严格验证可信邮件头是否与登录用户的电子邮件匹配，确保安全认证并防止意外的会话切换。
  • 🔒 邮件不匹配时用户一致注销：当邮件头中的可信电子邮件发生变化时，您现在会被正确注销并重定向，保障您会话的完整性。
  • 🛠 通用错误和内容验证改进：更智能的错误处理意味着更清晰的消息和更少的不必要重试——使批量上传、文档处理和知识索引更具弹性。
  • 🕵️ 改进聊天标题编辑的反馈：编辑聊天标题时如果出现问题，现在会清晰显示错误消息。

[!TIP]
寻找企业计划？ – 立即联系我们的销售团队！

获取增强功能，包括自定义主题和品牌、服务水平协议（SLA）支持、长期支持（LTS）版本等等！

👏 感谢我们了不起的贡献者的辛勤工作和奉献，使此次发布得以实现： @aindriu80, @SadmL, @Yu-QX, @vaclcer, @Davixk, @hdnh2006, @expruc, @ayanahye, @qingchunnh, @lucyknada, @rragundez, @silentoplayz, @EntropyYue, @GewoonJaap, @SeiranMaru, @mrexodia, @ER-EPR, @PeterDaveHello, @Zyfax, @guenhter, @diwakar-s-maurya, @PVBLIC-F, @dependabot[bot], @rragundez, @ShirasawaSama, @Classic298, @RGSS3, @Bouby308, @janaki-sasidhar, @TiancongLx

赞助商 🙌

🚀 我们想向我们出色的赞助商表达衷心的感谢，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。如果您希望在此展示，请随时联系我们！）

翡翠级

  • n8n n8n
  • Warp warp
  • Tailscale tailscale

@amd, @agency-agency, @roosi-gmbh, @Unsupervisedcom, @feddersen-group, @SRKConsulting, @Poxel2, @mushmax, @rndmcnlly, @icanbwell, @blogic-cz, @Krambu-Travis, @loitragg, @GenieDev101, @PortlandKyGuy, @CubicleGangster, @GitDatamind, @TOWaB, @TheZigGroup, @Classic298, @comet-ml, @catalyst

更新内容 （原始）

[0.6.14] - 2025-06-10

Added

  • 🤖 Automatic “Follow Up” Suggestions: Open WebUI now intelligently generates actionable “Follow Up” suggestions automatically with each message you send, helping you stay productive and inspired without interrupting your flow; you can always disable this in Settings if you prefer a distraction-free experience.
  • 🧩 OpenAI-Compatible Embeddings Endpoint: Introducing a fully OpenAI-style ‘/api/embeddings’ endpoint—now you can plug in OpenAI-style embeddings workflows with zero hassle, making integrations with external tools and platforms seamless and familiar.
  • ↗️ Model Pinning for Quick Access: Pin your favorite or most-used models to the sidebar for instant selection—no more scrolling through long model lists; your go-to models are always visible and ready for fast access.
  • 📌 Selector Model Item Menu: Each model in the selector now features a menu where you can easily pin/unpin to the sidebar and copy a direct link—simplifying collaboration and staying organized in even the busiest environments.
  • 🛑 Reliable Stop for Ongoing Chats in Multi-Replica Setups: Stopping or cancelling an in-progress chat now works reliably even in clustered deployments—ensuring every user can interrupt AI output at any time, no matter your scale.
  • 🧠 ‘Think’ Parameter for Ollama Models: Leverage new ’think’ parameter support for Ollama—giving you advanced control over AI reasoning process and further tuning model behavior for your unique use cases.
  • 💬 Picture Description Modes for Docling: Customize how images are described/extracted by Docling Loader for smarter, more detailed, and workflow-tailored image understanding in your document pipelines.
  • 🛠 Settings Modal Deep Linking: Every tab in Settings now has its own route—making direct navigation and sharing of precise settings faster and more intuitive.
  • 🎤 Audio HTML Component Token: Easily embed and play audio directly in your chats, improving voice-based workflows and making audio content instantly accessible and manageable from any conversation.
  • 🔑 Support for Secret Key File: Now you can specify ‘WEBUI_SECRET_KEY_FILE’ for more secure and flexible key management—ideal for advanced deployments and tighter security standards.
  • 💡 Clarity When Cloning Prompts: Cloned workspace prompts are clearly labelled with “(Clone)” and IDs have “-clone”, keeping your prompt library organized and preventing accidental overwrites.
  • 📝 Dedicated User Role Edit Modal: Updating user roles now reliably opens a dedicated edit user modal instead of cycling through roles—making it safer and more clear to manage team permissions.
  • 🏞️ Better Handling & Storage of Interpreter-Generated Images: Code interpreter-generated images are now centrally stored and reliably loaded from the database or cloud storage, ensuring your artifacts are always available.
  • 🚀 Pinecone & Vector Search Optimizations: Applied latest best practices from Pinecone for smarter timeouts, intelligent retry control, improved connection pooling, faster DNS, and concurrent batch handling—giving you more reliable, faster document search and RAG performance without manual tweaks.
  • ⚙️ Ollama Advanced Parameters Unified: ‘keep_alive’ and ‘format’ options are now integrated into the advanced params section—edit everything from the model editor for flexible model control.
  • 🛠️ CUDA 12.6 Docker Image Support: Deploy to NVIDIA GPUs with capability 7.0 and below (e.g., V100, GTX1080) via new cuda126 image—broadening your hardware options for scalable AI workloads.
  • 🔒 Experimental Table-Level PGVector Data Encryption: Activate pgcrypto encryption support for pgvector to secure your vector search table contents, giving organizations enhanced compliance and data protection—perfect for enterprise or regulated environments.
  • 👁 Accessibility Upgrades Across Interface: Chat buttons and close controls are now labelled and structured for optimal accessibility support, ensuring smoother operation with assistive technologies.
  • 🎨 High-Contrast Mode Expansions: High-contrast accessibility mode now also applies to menu items, tabs, and search input fields, offering a more readable experience for all users.
  • 🛠️ Tooltip & Translation Clarity: Improved translation and tooltip clarity, especially over radio buttons, making the UI more understandable for all users.
  • 🔠 Global Localization & Translation Improvements: Hefty upgrades to Traditional Chinese, Simplified Chinese, Hebrew, Russian, Irish, German, and Danish translation packs—making the platform feel native and intuitive for even more users worldwide.
  • ⚡ General Backend Stability & Security Enhancements: Refined numerous backend routines to minimize memory use, improve performance, and streamline integration with external APIs—making the entire platform more robust and secure for daily work.

Fixed

  • 🏷 Feedback Score Display Improved: Addressed overflow and visibility issues with feedback scores for more readable and accessible evaluations.
  • 🗂 Admin Settings Model Edits Apply Immediately: Changes made in the Model Editor within Admin Settings now take effect instantly, eliminating confusion during model management.
  • 🔄 Assigned Tools Update Instantly on New Chats: Models assigned with specific tools now consistently update and are available in every new chat—making tool workflows more predictable and robust.
  • 🛠 Document Settings Saved Only on User Action: Document settings now save only when you press the Save button, reducing accidental changes and ensuring greater control.
  • 🔊 Voice Recording on Older iOS Devices Restored: Voice input is now fully functional on older iOS devices, keeping voice workflows accessible to all users.
  • 🔒 Trusted Email Header Session Security: User sessions now strictly verify the trusted email header matches the logged-in user’s email, ensuring secure authentication and preventing accidental session switching.
  • 🔒 Consistent User Signout on Email Mismatch: When the trusted email in the header changes, you will now be properly signed out and redirected, safeguarding your session’s integrity.
  • 🛠 General Error & Content Validation Improvements: Smarter error handling means clearer messages and fewer unnecessary retries—making batch uploads, document handling, and knowledge indexing more resilient.
  • 🕵️ Better Feedback on Chat Title Edits: Error messages now show clearly if problems occur while editing chat titles.

[!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: @aindriu80, @SadmL, @Yu-QX, @vaclcer, @Davixk, @hdnh2006, @expruc, @ayanahye, @qingchunnh, @lucyknada, @rragundez, @silentoplayz, @EntropyYue, @GewoonJaap, @SeiranMaru, @mrexodia, @ER-EPR, @PeterDaveHello, @Zyfax, @guenhter, @diwakar-s-maurya, @PVBLIC-F, @dependabot[bot], @rragundez, @ShirasawaSama, @Classic298, @RGSS3, @Bouby308, @janaki-sasidhar, @TiancongLx

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

  • n8n n8n
  • Warp warp
  • Tailscale tailscale

@amd, @agency-agency, @roosi-gmbh, @Unsupervisedcom, @feddersen-group, @SRKConsulting, @Poxel2, @mushmax, @rndmcnlly, @icanbwell, @blogic-cz, @Krambu-Travis, @loitragg, @GenieDev101, @PortlandKyGuy, @CubicleGangster, @GitDatamind, @TOWaB, @TheZigGroup, @Classic298, @comet-ml, @catalyst

下载链接