发布日期: 2025-05-10
版本号: v0.6.8

版本0.6.8（2025年5月10日）带来了众多新功能和问题修复。新功能包括支持外部重排序器以优化知识库搜索、提供无样式PDF导出选项以减小文件大小、为阿拉伯语和波斯语用户添加专用字体、OneDrive支持指定SharePoint租户ID、OAuth头像实时更新、改进Milvus配置与S3标签兼容性、增强按钮和模态框的无障碍访问、优化管理员用户列表的加载与分页、明确默认代码解释器和标题生成提示模板、支持非根路径的Jupyter笔记本、始终显示工具栏滚动条、进行大量后端重构以提升稳定性、显著加快网络搜索速度并扩展了语言支持。修复的问题包括解决Nginx反向代理下WebSocket会话耗尽工作线程的问题、修复OpenAI音频转录错误、修正从右到左语言的消息输入显示、解决数学公式与中文字符相邻时的渲染问题、消除重复的网络搜索网址、修复知识库中的Markdown显示、解决LDAP导入问题以及确保Pinecone异步操作的安全性。

更新内容 （中文）

[0.6.8] - 2025-05-10

新增功能

  • 🏆 知识库搜索支持外部重排序器：通过全新的外部重排序器集成，为您的检索增强生成（RAG）工作流赋能；可通过界面轻松接入先进的重排序服务，提供更精准、相关的搜索结果，加速研究与洞见发现。
  • 📤 无样式 PDF 导出选项（减小文件体积）：导出聊天记录或文档时，现在可选择无样式 PDF 导出，下载更快、文件体积最小化、数据归档更整洁——非常适合大规模存储或共享。
  • 📝 波斯语与阿拉伯语支持 Vazirmatn 字体：阿拉伯语和波斯语用户现在将使用专业的 Vazirmatn 字体美观呈现文本，提升本地化阅读体验。
  • 🏷️ OneDrive 支持 SharePoint 租户 ID：现在可在 OneDrive 设置中指定 SharePoint 租户 ID，实现无缝身份验证和精细的企业集成。
  • 👤 刷新 OAuth 头像图片：您的 OAuth 头像现在实时更新，确保在所有集成平台中，您的在线状态和头像始终与最新身份保持一致。
  • 🔧 Milvus 配置改进：现在可直接在设置中配置 Milvus 的索引和度量类型；全面掌控您的向量数据库，获得更精确、健壮的 AI 搜索体验。
  • 🛡️ S3 标签切换以增强兼容性：通过环境变量可选开启 S3 标签功能，实现与所有存储后端（包括 Cloudflare R2 等不支持标签的后端）的完全兼容，确保附件与文档管理零错误。
  • 👨‍🦯 图标按钮无障碍改进：关键交互式图标按钮现在包含 aria-label 和 ARIA 描述，使屏幕阅读器能够精确指导每个按钮的操作，提升无障碍访问体验。
  • ♿ 增强无障碍性：模态框焦点陷阱：模态对话框和弹出窗口现在具备焦点陷阱和改进的 ARIA 角色，确保无缝导航和屏幕阅读器支持——使界面对所有用户（包括键盘和辅助技术用户）更友好。
  • 🏃 管理员用户列表加载指示器改进：管理员面板中的用户列表加载体验现在更清晰、响应更迅速。
  • 🧑‍🤝‍🧑 管理员用户列表页面容量增大：管理员现在可在管理界面中每页管理最多 30 名用户，大幅减少分页操作，使大型用户团队的管理更轻松快捷。
  • 🌠 澄清默认代码解释器提示：内置代码解释器提示现在更明确，防止 AI 在非必要时将代码包裹在 Markdown 代码块中——确保格式正确的代码每次都能按预期运行。
  • 🧾 改进默认标题生成提示模板：标题生成现在使用健壮的模板以获得可靠的 JSON 输出，改善聊天组织性和可搜索性。
  • 🔗 支持非根路径部署的 Jupyter Notebook：基于 Notebook 的代码执行现在支持非根路径部署的 Jupyter 服务器，为混合或多用户部署提供完全灵活性。
  • 📰 UI 溢出工具始终显示滚动条：当可用工具超出显示区域时，滚动条现在始终可见，并提供便捷的“显示全部”切换，使大型工具集的导航更快捷直观。
  • 🛠️ 通用后端重构以提升稳定性：对后端组件进行了多项底层改进，确保更流畅的性能、更少的错误，以及所有用户更可靠的体验。
  • 🚀 优化 Web 搜索以加快结果返回：Web 搜索速度和性能显著提升，以创纪录的时间提供答案和来源，加速您研究密集型工作流。
  • 💡 支持更多语言：扩展的语言支持确保更广泛的用户群体能够以其母语享受直观自然的界面。

修复问题

  • 🏃‍♂️ 修复因 Websocket 问题导致 Nginx 反向代理耗尽工作进程：Websocket 会话现在在 Nginx 后面完全兼容，消除了工作进程耗尽问题，即使在复杂部署中也能恢复实时聊天的 24/7 可靠性。
  • 🎤 修复 OpenAI 音频转录问题：基于 OpenAI 的音频转录现在可以无误处理 WebM 和较新格式，确保每次都能实现无缝的语音转文字工作流。
  • 👉 修复消息输入框从右到左显示问题：聊天消息输入框现在正确支持从右到左语言，为阿拉伯语、希伯来语等提供完美的输入和阅读体验。
  • 🀄 Katex：修复中文字符与数学公式相邻时的渲染：数学公式现在即使在紧邻中文（CJK）字符时也能完美渲染，提升多语言团队和跨语言文档的视觉清晰度。
  • 🔂 消除重复的 Web 搜索 URL：搜索结果现在可靠地过滤掉重复 URL，确保您的知识和搜索引用始终清晰、精简、易于审阅。
  • 🗂️ 修复知识库中的 Markdown 渲染：Markdown 现在在知识库中正确显示，实现更丰富的文件格式和信息清晰度。
  • 🌐 修复 LDAP 导入/加载问题：LDAP 用户导入现在正确处理，确保顺畅的入职和访问流程，无中断。
  • 🌎 Pinecone 批量操作与异步安全性：所有 Pinecone 操作（批量插入、更新、删除）现在在异步环境中高效安全运行，提升性能并防止大规模 RAG 作业中的速度下降。

更新内容 （原始）

[0.6.8] - 2025-05-10

Added

  • 🏆 External Reranker Support for Knowledge Base Search: Supercharge your Retrieval-Augmented Generation (RAG) workflows with the new External Reranker integration; easily plug in advanced reranking services via the UI to deliver sharper and more relevant search results, accelerating research and insight discovery.
  • 📤 Unstylized PDF Export Option (Reduced File Size): When exporting chat transcripts or documents, you can now choose an unstylized PDF export for snappier downloads, minimal file size, and clean data archiving—perfect for large-scale storage or sharing.
  • 📝 Vazirmatn Font for Persian & Arabic: Arabic and Persian users will now see their text beautifully rendered with the specialized Vazirmatn font for an improved localized reading experience.
  • 🏷️ SharePoint Tenant ID Support for OneDrive: You can now specify a SharePoint tenant ID in OneDrive settings for seamless authentication and granular enterprise integration.
  • 👤 Refresh OAuth Profile Picture: Your OAuth profile picture now updates in real-time, ensuring your presence and avatar always match your latest identity across integrated platforms.
  • 🔧 Milvus Configuration Improvements: Configure index and metric types for Milvus directly within settings; take full control of your vector database for more accurate and robust AI search experiences.
  • 🛡️ S3 Tagging Toggle for Compatibility: Optional S3 tagging via an environment toggle grants full compatibility with all storage backends—including those that don’t support tagging like Cloudflare R2—ensuring error-free attachment and document management.
  • 👨‍🦯 Icon Button Accessibility Improvements: Key interactive icon-buttons now include aria-labels and ARIA descriptions, so screen readers provide precise guidance about what action each button performs for improved accessibility.
  • ♿ Enhanced Accessibility with Modal Focus Trap: Modal dialogs and pop-ups now feature a focus trap and improved ARIA roles, ensuring seamless navigation and screen reader support—making the interface friendlier for everyone, including keyboard and assistive tech users.
  • 🏃 Improved Admin User List Loading Indicator: The user list loading experience is now clearer and more responsive in the admin panel.
  • 🧑‍🤝‍🧑 Larger Admin User List Page Size: Admins can now manage up to 30 users per page in the admin interface, drastically reducing pagination and making large user teams easier and faster to manage.
  • 🌠 Default Code Interpreter Prompt Clarified: The built-in code interpreter prompt is now more explicit, preventing AI from wrapping code in Markdown blocks when not needed—ensuring properly formatted code runs as intended every time.
  • 🧾 Improved Default Title Generation Prompt Template: Title generation now uses a robust template for reliable JSON output, improving chat organization and searchability.
  • 🔗 Support Jupyter Notebooks with Non-Root Base URLs: Notebook-based code execution now supports non-root deployed Jupyter servers, granting full flexibility for hybrid or multi-user setups.
  • 📰 UI Scrollbar Always Visible for Overflow Tools: When available tools overflow the display, the scrollbar is now always visible and there’s a handy “show all” toggle, making navigation of large toolsets snappier and more intuitive.
  • 🛠️ General Backend Refactoring for Stability: Multiple under-the-hood improvements have been made across backend components, ensuring smoother performance, fewer errors, and a more reliable overall experience for all users.
  • 🚀 Optimized Web Search for Faster Results: Web search speed and performance have been significantly enhanced, delivering answers and sources in record time to accelerate your research-heavy workflows.
  • 💡 More Supported Languages: Expanded language support ensures an even wider range of users can enjoy an intuitive and natural interface in their native tongue.

Fixed

  • 🏃‍♂️ Exhausting Workers in Nginx Reverse Proxy Due to Websocket Fix: Websocket sessions are now fully compatible behind Nginx, eliminating worker exhaustion and restoring 24/7 reliability for real-time chats even in complex deployments.
  • 🎤 Audio Transcription Issue with OpenAI Resolved: OpenAI-based audio transcription now handles WebM and newer formats without error, ensuring seamless voice-to-text workflows every time.
  • 👉 Message Input RTL Issue Fixed: The chat message input now displays correctly for right-to-left languages, creating a flawless typing and reading experience for Arabic, Hebrew, and more.
  • 🀄 Katex: Proper Rendering of Chinese Characters Next to Math: Math formulas now render perfectly even when directly adjacent to Chinese (CJK) characters, improving visual clarity for multilingual teams and cross-language documents.
  • 🔂 Duplicate Web Search URLs Eliminated: Search results now reliably filter out URL duplicates, so your knowledge and search citations are always clean, trimmed, and easy to review.
  • 📄 Markdown Rendering Fixed in Knowledge Bases: Markdown is now displayed correctly within knowledge bases, enabling better formatting and clarity of information-rich files.
  • 🗂️ LDAP Import/Loading Issue Resolved: LDAP user imports process correctly, ensuring smooth onboarding and access without interruption.
  • 🌎 Pinecone Batch Operations and Async Safety: All Pinecone operations (batch insert, upsert, delete) now run efficiently and safely in an async environment, boosting performance and preventing slowdowns in large-scale RAG jobs.

下载链接