Repository navigation
Lexora 0.19.1
0.19.1
中文
这是 0.19.0 的补丁版本。上个版本带来了伙伴向用户提问的能力，但提问卡片的样式与交互还没达到我们期望的效果，因此很快推出了这次更新：重新打磨卡片布局与视觉层级，支持逐题切换作答，让问题更聚焦、填写更从容。
此外，我们也处理了一些性能问题，改善长对话阅读与分屏拖拽的流畅度，并修复补充消息延迟显示、编辑和重新生成后的加载与阅读位置问题。
这些体验仍有进一步打磨的空间，我们会在后续版本中持续改进。
⚡ 改进
-
🙋 更清晰的提问面板与逐题作答 by @shanyuhai123：
调整提问卡片的排版、间距与交互方式，让你更专注于当前问题：- 逐题填写：一次展示一个问题，可前后切换并查看进度，保留各题已填内容，完成后统一提交；暂时离开任务再返回，也能继续填写；
- 问题更聚焦：优化伙伴的提问方式，背景、分析和计划先在对话中说明，面板聚焦简短的问题、选项与关键取舍，方便回看依据、快速作答；
- 独立滚动：问题面板位于输入区上方，较长内容在面板内滚动，减少对对话阅读位置的打扰；
- 层级更清楚：改善问题正文的 Markdown 展示，以及选项说明、推荐标记、自定义回答和按钮的视觉状态。
-
📜 长对话阅读更轻快 by @shanyuhai123：
优化长对话的渲染方式，减少屏幕外内容带来的界面负担。滚动浏览时保留相关内容的展开状态，并兼顾文字选择与正在编辑的内容，减少反复展开和操作中断。 -
📐 分屏拖拽更顺畅 by @shanyuhai123：
优化调整分屏尺寸时的界面更新，减少拖动过程中的重复布局计算，并改善宽度变化后的阅读位置保持，让长对话与多面板并排使用更稳定。
🛠️ 修复
-
💬 补充消息及时显示 by @shanyuhai123：
修复工作中发送的补充消息已经送达，却未及时出现在对话中的问题。补充消息送达或排队消息开始处理时，会及时刷新对话内容。 -
🔄 编辑与重新生成后的加载和阅读位置 by @shanyuhai123：
修复编辑消息、重新生成回复后，对话加载状态不正确的问题。重新生成时保留前文及相关过程记录，并改善阅读位置和底部跟随，减少历史内容短暂不显示与页面跳动。
English
This is a patch release for 0.19.0. The previous release introduced the ability for companions to ask you questions, but the question cards’ styling and interactions weren’t where we wanted them to be. We’re following up with a refined layout, clearer visual hierarchy, and one-at-a-time question navigation to make answering more focused and comfortable.
We also addressed performance issues to make long conversations and split-pane resizing smoother, alongside fixes for delayed supplementary messages and loading and reading-position issues after editing or regenerating replies.
There’s still room for improvement, and we’ll keep refining these experiences in future releases.
⚡ Improved
-
🙋 Clearer Question Panels & One-at-a-Time Answering by @shanyuhai123:
Refined question-card layout, spacing, and interaction to help you focus on the current question:- Answer one at a time: Move between questions, check your progress, and retain each answer before submitting them together. Temporarily leave a task and return to continue filling in the form;
- More focused questions: Refined how companions ask questions, keeping background, analysis, and plans in the conversation while the panel focuses on concise questions, choices, and essential tradeoffs;
- Independent scrolling: The panel stays above the composer, with longer content scrolling inside it to reduce disruption to your reading position;
- Clearer hierarchy: Improved Markdown prompts, option descriptions, recommendation labels, custom answers, and button states.
-
📜 Lighter Long-Conversation Reading by @shanyuhai123:
Optimized long-conversation rendering to reduce the interface cost of offscreen content. Expanded sections retain their state while scrolling, with handling for text selection and active editing to reduce repeated expansion and interrupted interactions. -
📐 Smoother Split-Pane Resizing by @shanyuhai123:
Optimized interface updates during resizing to reduce repeated layout calculations, with better reading-position preservation after width changes for more stable side-by-side work.
🛠️ Fixed
-
💬 Prompt Display of Supplementary Messages by @shanyuhai123:
Fixed supplementary messages being delivered during ongoing work without promptly appearing in the conversation. The conversation now refreshes when a supplement is delivered or a queued message begins processing. -
🔄 Loading & Reading Position After Editing or Regeneration by @shanyuhai123:
Fixed incorrect conversation loading states after editing messages or regenerating replies. Regeneration now retains earlier conversation content and related activity records, with improved reading-position preservation and bottom following to reduce temporarily missing history and sudden jumps.
📥 下载地址 / Downloads
Windows
macOS 15+
Linux
文件校验 / Checksums
Intel／AMD 电脑请选择 x64，ARM 设备请选择 ARM64。macOS 当前仅支持 Apple Silicon。
Choose x64 for Intel/AMD computers or ARM64 for ARM devices. macOS currently supports Apple Silicon only.
Lexora 0.19.0
中文 · 0.19.0
这个版本，我们围绕日常使用体验做了一轮集中打磨：减少不必要的等待与打断，让操作衔接更顺畅、过程状态更清楚、结果反馈更可靠。从启动应用、发送消息，到朗读回复、生成图片和切换视图，让你少一些反复确认，更专注于任务本身。
同时，新增了 Agent 向用户提问的能力：伙伴需要补充信息或确认选择时，可以直接在任务中发起问题，通过选项或文字收集你的回答。浏览器操作也有了独立的执行页面，默认在后台打开，减少对当前页面的打扰。
你还可以选择稳定版或尝鲜版更新通道，并根据模型能力调整工具调用方式，在熟悉的工作流程中获得更灵活的选择。
✨ 新增
-
🙋 新增 Agent 向用户提问能力 by @shanyuhai123：
在需要澄清需求、补充信息或确认选择时，伙伴可以在任务中向你发起问题：- 灵活回答：支持单选、多选和文字回答，也能填写自定义内容或补充说明；
- 任务内处理：问题通过卡片展示，多个提问可以切换查看，也可以选择不回答；
- 保留问答记录：处理后的问题与回答进入任务历史。提交或送达失败时有明确提示，已保存的回答支持重试送达，无需重新填写。
-
🌐 浏览器后台打开与执行页面隔离 by @shanyuhai123：
伙伴与子任务使用各自的浏览器执行页面，减少并行操作互相干扰。页面默认在后台打开，不替换你正在查看的内容；需要查看或接管时，再展示对应页面。 -
🧪 稳定版与尝鲜版更新通道 by @shanyuhai123：
在应用更新设置中选择适合自己的通道：- 稳定版：仅接收正式版本，适合日常使用；
- 尝鲜版（Beta）：接收正式版及 Beta 更新，提前体验新功能；Beta 可能不稳定，也可能暂不提供。
加入尝鲜版前会提示备份重要数据。更新仍由你决定何时安装，切回稳定版不会自动降级。
⚡ 改进
-
⚙️ 按模型能力选择工具调用方式 by @shanyuhai123：
在运行设置中选择三种模式：- 关闭：停用 CodeMode，由模型直接调用可用工具；
- 混合调用：简单操作直接调用工具，批量查询通过代码编排并过滤结果；
- 代码编排：普通工具必须通过代码调用，适合能稳定生成正确脚本的模型。
模式变化从下一次运行生效，所有调用仍遵循任务权限。
-
🚀 更顺畅的启动与加载反馈 by @shanyuhai123：
复用已准备的启动状态，减少重复初始化。启动时会显示读取本地数据、准备历史记录、准备模型与能力、恢复工作台等阶段，让等待过程更清楚。 -
🔎 添加服务时快速搜索模型 by @Clannad47：
添加模型服务时，可以按模型名称或模型 ID 搜索列表，不必在大量模型中逐项寻找。 -
🔊 更简洁的朗读操作与状态反馈 by @lexora-agent：
朗读状态直接呈现在回复按钮上，区分准备、生成语音、加载音频与播放阶段。等待时可以点击取消，播放时可以点击停止，不再额外占用输入区展示播放状态。
🛠️ 修复
-
🖼️ 图片生成结果与重复提交 by @shanyuhai123：
修复请求断连、响应异常或图片保存出错时，结果状态判断不准确的问题。现在会区分生成失败、结果未知和保存未完成，并显示出错阶段。同一轮中，内容或输出位置相同的请求仍在执行、结果未确认或保存未完成时，会阻止重复提交，并提示先检查已有产物。 -
📊 上下文用量与压缩状态 by @shanyuhai123：
修复同一上下文连续对话时，用量短暂清空、压缩后仍显示旧值的问题。轮次切换时保留最近一次有效用量；压缩后清除旧读数，等待新的有效数据，并及时刷新。模型返回空白摘要时按失败处理，不再误判为压缩成功。 -
💬 发送后的输入框与消息顺序 by @shanyuhai123, @lexora-agent：
修复发送时的输入焦点、发送后的撤销历史及草稿切换问题，避免撤销操作重新带回已发送内容。已接受的消息会及时显示，后续刷新保持历史与附件衔接；失败、取消等没有回复正文的结束结果，也会排在触发它的用户消息之后。 -
🧵 命令并发与产物写入 by @shanyuhai123：
修复长命令不必要地阻塞同一工作区内其他命令和文件编辑的问题，同一文件的受控写入仍会协调冲突。图片等产物保存前会重新检查授权、目标路径与文件版本，检测到权限变化、链接替换或并发修改时阻止冲突写入，降低误覆盖、写错位置的风险。 -
📐 侧栏动画中的视图跟随 by @QAyong, @shanyuhai123：
修复侧栏展开、收起过程中，插件视图与内置浏览器未及时跟随布局变化的问题。视图会在动画过程中持续更新位置与尺寸，动画结束或中途取消后也会重新对齐，减少错位和突然跳动。 -
📄 Markdown 预览中的相对路径图片 by @QAyong, @shanyuhai123：
Markdown 文档预览现在可以按当前文档位置加载相对路径图片，支持中文、空格及父级目录引用，不必改写成绝对路径。图片加载失败时会显示明确提示，并支持原地重试，无需重新打开文档。 -
🎙️ 关闭语音后的按钮闪烁 by @QAyong：
修复关闭语音后，新建对话或从设置返回时，录音、朗读按钮仍短暂出现的问题。按钮会等待语音偏好加载完成后再显示，并分别遵循语音输入与输出的开关状态，减少界面闪烁。
English · 0.19.0
This release focuses on making everyday use smoother, clearer, and less disruptive. From starting the app and sending messages to reading replies aloud, generating images, and switching views, these refinements reduce unnecessary waiting and repeated checks so you can stay focused on your task.
Agents can now ask you questions within a task when they need information or a decision, collecting your answers through choices or text. Browser operations also have separate execution pages that open in the background by default, reducing interruptions to what you are viewing.
You can also choose between Stable and Beta update channels and adjust tool invocation to suit your model, adding flexibility to your familiar workflow.
✨ Added
-
🙋 Agents Can Ask You Questions by @shanyuhai123:
Companions can ask questions within a task to clarify requirements, gather information, or confirm a choice:- Flexible answers: Single-choice, multiple-choice, and text answers, with custom responses and additional notes;
- In-task interaction: Questions appear as cards. Switch between requests or choose not to answer;
- Question history: Processed questions and answers remain in task history. Submission and delivery failures provide clear feedback, and saved answers can be delivered again without re-entering them.
-
🌐 Background Browsing & Separate Execution Pages by @shanyuhai123:
Companions and subtasks use separate browser execution pages to reduce interference between concurrent operations. Pages open in the background by default without replacing what you are viewing, and can be shown when you need to inspect or take over. -
🧪 Stable & Beta Update Channels by @shanyuhai123:
Choose your preferred channel in application update settings:- Stable: Receives official releases only, recommended for everyday use;
- Beta: Receives official and Beta releases for earlier access to features. Beta versions may be unstable or temporarily unavailable.
Joining Beta prompts you to back up important data. You still decide when to install updates, and switching back to Stable does not automatically downgrade your application.
⚡ Improved
-
⚙️ Choose Tool Invocation to Suit Your Model by @shanyuhai123:
Runtime settings now offer three modes:- Off: Disable CodeMode and let the model invoke available tools directly;
- Mixed: Invoke tools directly for simple operations, or orchestrate batch queries and filter results through code;
- Code orchestration: Require ordinary tools to be invoked through code, for models that reliably generate correct scripts.
Changes apply from the next run. All calls continue to follow task permissions.
-
🚀 Smoother Startup & Clearer Loading Feedback by @shanyuhai123:
Reuse prepared startup state to reduce repeated initialization. Loading feedback now identifies stages such as reading local data, preparing history, preparing models and capabilities, and restoring the workbench. -
🔎 Search Models When Adding a Provider by @Clannad47:
Filter the model list by display name or model ID during provider setup instead of searching through long lists manually. -
🔊 Simpler Read-Aloud Controls & Status Feedback by @lexora-agent:
Reply buttons now show preparation, speech generation, audio loading, and playback states directly. Click to cancel while waiting or stop during playback, without an additional playback indicator occupying the composer.
🛠️ Fixed
-
🖼️ Image Generation Outcomes & Duplicate Submissions by @shanyuhai123:
Fixed inaccurate outcome reporting after disconnections, abnormal responses, or image-saving errors. Results now distinguish failed generation, unknown outcomes, and incomplete saving, with clearer failure stages. Within the same turn, requests with matching content or output locations are blocked from being submitted again while running, unconfirmed, or incompletely saved, with guidance to check existing outputs first. -
📊 Context Usage & Compaction Status by @shanyuhai123:
Fixed usage briefly disappearing between turns in the same context or showing stale values after compaction. The latest valid reading is retained between turns; compaction clears outdated readings while awaiting fresh data and triggers timely refreshes. Blank summaries are now treated as failures rather than successful compaction. -
💬 Composer State & Message Ordering After Sending by @shanyuhai123, @lexora-agent:
Fixed input focus, undo history, and draft-switching issues around sending, preventing Undo from restoring already-sent content. Accepted messages appear promptly, with history and attachments preserved during refreshes. Terminal results without reply text, including failures and cancellations, now appear after the user messages that triggered them. -
🧵 Concurrent Commands & Artifact Writes by @shanyuhai123:
Fixed long-running commands unnecessarily blocking other commands and file edits in the same workspace, while controlled writes to the same file still coordinate conflicts. Saving images and other artifacts rechecks authorization, destination paths, and file versions, rejecting conflicting writes when permission changes, link replacements, or concurrent modifications are detected. -
📐 Views Following Sidebar Animations by @QAyong, @shanyuhai123:
Fixed plugin views and the built-in browser lagging behind layout changes while the sidebar expands or collapses. Position and size update throughout the animation, with final alignment after completion or interruption to reduce misalignment and sudden jumps. -
📄 Relative Image Paths in Markdown Previews by @QAyong, @shanyuhai123:
Markdown previews now resolve image paths relative to the document, including Chinese characters, spaces, and parent-directory references, without requiring absolute paths. Failed images show clear feedback and can be retried in place without reopening the document. -
🎙️ Speech Button Flashing When Disabled by @QAyong:
Fixed recording and read-aloud buttons briefly appearing in new conversations or after returning from settings when disabled. Buttons wait for speech preferences to load and respect the separate input and output switches, reducing interface flicker.