发布日期: 2026-02-14
版本号: v0.8.1

0.8.1版本于2026年2月14日发布，包含重要的数据库架构变更。强烈建议在生产环境升级前备份所有数据库及相关数据。如果部署在多任务、多服务器或负载均衡环境中，所有实例必须同步更新，不支持滚动更新，否则将导致应用故障。

此次更新新增了多项功能：优化了频道用户活跃状态的数据库查询速度；新增带模型路由的Responses API端点，可将请求根据模型字段路由至正确后端，支持vLLM等后端；通过预获取用户组ID优化了模型和提示列表的加载性能；用批量查询替代了多项单独的访问控制检查，显著降低了数据库负载；延迟加载用户头像使用户列表加载更快；内置搜索工具现在会遵循管理员配置的搜索结果数量设置；支持SCIM外部ID以改善与身份提供商的集成；并更新了葡萄牙语（巴西）翻译。

修复了多个问题：包括公共分享权限显示和API绕过的安全漏洞；修复了对没有数据库条目的直接Ollama/OpenAI模型的访问控制更改无法持久化及错误消息显示异常的问题；解决了模型推理轨迹在流式响应时渲染成碎片导致浏览器卡顿的问题；修正了0.8.0版本在ARM设备上因torch版本问题无法启动的情况；修复了PostgreSQL环境下创建/列出技能和按模型ID获取聊天ID的兼容性问题；使SQLite在删除聊天时能正确级联删除相关消息；修复了使用Ollama云模型时因错误附加“:latest”后缀导致模型未找到的问题；调整了知识选择器工具提示的层级；使模型选择器在打开时能正确滚动并居中选中项；修复了启用社区共享功能后同步模态框可能意外出现的问题；修正了知识集合菜单中长名称的布局和截断；解决了处理包含特定键的文档元数据时因迭代中字典变更导致的崩溃问题；并修正了密码验证正则表达式中转义字符的处理。

更新内容 （中文）

[0.8.1] - 2026-02-14

[!CAUTION] ⚠️ 数据库迁移：此版本包含数据库架构变更；我们强烈建议在生产环境升级前备份数据库及相关数据。若使用多 worker、多服务器或负载均衡部署，所有实例必须同步更新；滚动更新不支持，将因架构不兼容导致应用故障。

新增

  • 🚀 频道用户活跃状态。通过优化数据库查询，现在检查频道中的用户活跃状态更快。提交
  • 🔗 支持模型路由的响应 API 端点。OpenAI API 代理现支持 /responses 端点，可根据请求中的模型字段将请求路由至正确的后端，而非始终使用第一个配置端点。这实现了对提供 /skills 和 /v1/responses 端点的后端（如 vLLM）的支持。提交，提交
  • ⚡ 模型与提示列表优化。通过预取用户组 ID 代替多次数据库查询，提升了加载模型和提示时的性能。提交
  • 🗄️ 批量访问控制查询。通过将多个单独的访问检查替换为单次批量查询，提升了加载模型、提示和知识库时的性能，显著降低了大规模部署的数据库负载。提交
  • 💨 用户列表加载更快。通过延迟加载用户头像，用户列表现在加载显著更快；头像由浏览器单独并行获取，改善了缓存并降低了数据库负载。提交
  • 🔍 网络搜索结果数量。内置的 search_web 工具现在会遵循管理员配置的“搜索结果数量”设置，而不是在使用原生函数调用模式时始终返回 5 个结果。#21373，#21371
  • 🔐 SCIM externalId 支持。启用了 SCIM 的部署现在可以存储和管理用于用户配置的 externalId，从而更好地与 Microsoft Entra ID 和 Okta 等身份提供商集成。#21099，#21280，提交
  • 🌐 翻译更新。已更新葡萄牙语（巴西）翻译。

修复

  • 🛡️ 公共共享安全修复。修复了一个安全问题：具有写入权限的用户无论其实际公共共享权限如何都能看到“公共共享”选项，并且直接 API 调用可以绕过前端共享限制。#21358，#21356
  • 🔒 直接模型访问控制修复。模型访问控制变更现在能正确应用于没有数据库条目的直接 Ollama 和 OpenAI 模型，并且错误信息能正确显示，而不是显示“[object Object]”。提交，#21377
  • 💭 推理轨迹渲染性能。来自模型的推理轨迹现在可以正确渲染，不会被分割成多个片段，从而防止流式响应期间的浏览器卡顿。#21348，提交
  • 🖥️ ARM 设备兼容性修复。修复了升级到 0.8.0 后，由于 torch 2.10.0 导致 SIGILL 错误，ARM 设备（如 Raspberry Pi 4）无法启动的问题；现已固定为 torch<=2.9.1。#21385，#21349
  • 🗄️ Skills PostgreSQL 兼容性修复。修复了 PostgreSQL 兼容性问题，该问题导致创建或列出技能时出现 TypeError 错误，而 SQLite 工作正常。#21372，提交，#21365
  • 🗄️ PostgreSQL 分析查询修复。修复了由于不兼容的 DISTINCT 排序，在 PostgreSQL 上按模型 ID 检索聊天 ID 会失败的问题，而 SQLite 工作正常。#21347，提交
  • 🗃️ SQLite 级联删除修复。在 SQLite 中删除聊天现在会正确删除所有关联消息，与 PostgreSQL 行为一致，并防止孤立数据。#21362
  • ☁️ Ollama Cloud 模型命名修复。修复了使用 Ollama Cloud 模型时，由于错误地在模型名称后附加了“:latest”而导致“Model not found”错误的问题。#21386
  • 🛠️ 知识选择器工具提示 z-index。修复了“选择知识”下拉菜单中的工具提示被菜单遮挡，导致难以阅读知识项名称和描述的问题。#21375
  • 🎯 模型选择器滚动位置。模型选择器下拉菜单现在在打开时会正确滚动并居中显示当前选中的模型，并在重新打开时重置滚动位置。提交
  • 🐛 同步模态框意外出现。修复了如果用户之前通过同步参数访问过应用，在启用“社区共享”功能后同步模态框会意外出现的问题。#21376
  • 🎨 知识集合布局修复。修复了知识集成菜单中的布局问题，长集合名称现在可以正确截断并显示省略号，而不会导致缩进问题。#21374
  • 📝 元数据处理崩溃修复。修复了一个潜在错误，处理包含特定键（content, pages, tables, paragraphs, sections, figures）的文档元数据时，会因迭代过程中字典变更而导致 RuntimeError。#21105
  • 🔑 密码验证正则表达式修复。通过添加原始字符串前缀修复了密码验证正则表达式，确保 \d 和 \w 等转义序列被正确解释。#21400，#21399

变更

  • ⚠️ 数据库迁移：此版本包含数据库架构变更；我们强烈建议在生产环境升级前备份数据库及相关数据。若使用多 worker、多服务器或负载均衡部署，所有实例必须同步更新；滚动更新不支持，将因架构不兼容导致应用故障。

更新内容 （原始）

[0.8.1] - 2026-02-14

[!CAUTION] ⚠️ Database Migrations: This release includes database schema changes; we strongly recommend backing up your database and all associated data before upgrading in production environments. If you are running a multi-worker, multi-server, or load-balanced deployment, all instances must be updated simultaneously, rolling updates are not supported and will cause application failures due to schema incompatibility.

Added

  • 🚀 Channel user active status. Checking user active status in channels is now faster thanks to optimized database queries. Commit
  • 🔗 Responses API endpoint with model routing. The OpenAI API proxy now supports a /responses endpoint that routes requests to the correct backend based on the model field in the request, instead of always using the first configured endpoint. This enables support for backends like vLLM that provide /skills and /v1/responses endpoints. Commit, Commit
  • ⚡ Model and prompt list optimization. Improved performance when loading models and prompts by pre-fetching user group IDs once instead of making multiple database queries. Commit
  • 🗄️ Batch access control queries. Improved performance when loading models, prompts, and knowledge bases by replacing multiple individual access checks with single batch queries, significantly reducing database load for large deployments. Commit
  • 💨 Faster user list loading. User lists now load significantly faster by deferring profile image loading; images are fetched separately in parallel by the browser, improving caching and reducing database load. Commit
  • 🔍 Web search result count. The built-in search_web tool now respects the admin-configured “Search Result Count” setting instead of always returning 5 results when using Native Function Calling mode. #21373, #21371
  • 🔐 SCIM externalId support. SCIM-enabled deployments can now store and manage externalId for user provisioning, enabling better integration with identity providers like Microsoft Entra ID and Okta. #21099, #21280, Commit
  • 🌐 Translation updates. Portuguese (Brazil) translations were updated.

Fixed

  • 🛡️ Public sharing security fix. Fixed a security issue where users with write access could see the Public sharing option regardless of their actual public sharing permission, and direct API calls could bypass frontend sharing restrictions. #21358, #21356
  • 🔒 Direct model access control fix. Model access control changes now persist correctly for direct Ollama and OpenAI models that don’t have database entries, and error messages display properly instead of showing “[object Object]”. Commit, #21377
  • 💭 Reasoning trace rendering performance. Reasoning traces from models now render properly without being split into many fragments, preventing browser slowdowns during streaming responses. #21348, Commit
  • 🖥️ ARM device compatibility fix. Fixed an issue where upgrading to 0.8.0 would fail to start on ARM devices (like Raspberry Pi 4) due to torch 2.10.0 causing SIGILL errors; now pinned to torch<=2.9.1. #21385, #21349
  • 🗄️ Skills PostgreSQL compatibility fix. Fixed a PostgreSQL compatibility issue where creating or listing skills would fail with a TypeError, while SQLite worked correctly. #21372, Commit, #21365
  • 🗄️ PostgreSQL analytics query fix. Fixed an issue where retrieving chat IDs by model ID would fail on PostgreSQL due to incompatible DISTINCT ordering, while SQLite worked correctly. #21347, Commit
  • 🗃️ SQLite cascade delete fix. Deleting chats now properly removes all associated messages in SQLite, matching PostgreSQL behavior and preventing orphaned data. #21362
  • ☁️ Ollama Cloud model naming fix. Fixed an issue where using Ollama Cloud models would fail with “Model not found” errors because “:latest” was incorrectly appended to model names. #21386
  • 🛠️ Knowledge selector tooltip z-index. Fixed an issue where tooltips in the “Select Knowledge” dropdown were hidden behind the menu, making it difficult to read knowledge item names and descriptions. #21375
  • 🎯 Model selector scroll position. The model selector dropdown now correctly scrolls to and centers the currently selected model when opened, and resets scroll position when reopened. Commit
  • 🐛 Sync modal unexpected appearance. Fixed an issue where the Sync Modal would appear unexpectedly after enabling the “Community Sharing” feature if the user had previously visited the app with the sync parameter. #21376
  • 🎨 Knowledge collection layout fix. Fixed a layout issue in the Knowledge integration menu where long collection names caused indentation artifacts and now properly truncate with ellipsis. #21374
  • 📝 Metadata processing crash fix. Fixed a latent bug where processing document metadata containing certain keys (content, pages, tables, paragraphs, sections, figures) would cause a RuntimeError due to dictionary mutation during iteration. #21105
  • 🔑 Password validation regex fix. Fixed the password validation regex by adding the raw string prefix, ensuring escape sequences like d and w are interpreted correctly. #21400, #21399

Changed

  • ⚠️ Database Migrations: This release includes database schema changes; we strongly recommend backing up your database and all associated data before upgrading in production environments. If you are running a multi-worker, multi-server, or load-balanced deployment, all instances must be updated simultaneously, rolling updates are not supported and will cause application failures due to schema incompatibility.

下载链接