Repository navigation
0.18.0
中文
这个版本，你可以 用语音记录想法，也可以听伙伴朗读回复；识图、生图与 OCR 有了集中配置入口，不必为了处理图片反复切换对话模型。
你还可以在插件市场中发现更多社区插件，为工作台添加喜欢的功能。本次带来了 DSH 风格的任务状态栏，并继续打磨下载、并行执行、输入和文件操作体验。
✨ 新增
-
🎙️ 语音输入与回复朗读 by @shanyuhai123：
在输入框中录音，将语音转成文字，编辑确认后再发送；也可以直接朗读伙伴的回复：- 语音输入：选择转写服务，并按服务能力设置识别语言，支持麦克风测试与试说；
- 回复朗读：选择系统声音或语音合成模型，按来源能力切换声音、调整语速与试听；
- 随时停止：录音和朗读都可以中途停止，服务中断时保留已识别的文字。首次使用前，请在「设置 → 语音」中完成配置，并允许麦克风访问。
-
🖼️ 集中设置识图、生图与 OCR by @shanyuhai123：
在「设置 → 模型」中选择处理图片的方式：- 识图：使用当前模型，或指定图像模型与插件服务，理解和比较多张图片、提取文字、读取文档图片；
- 生图：选择生成服务与补充模型，按服务能力设置尺寸、质量、背景，并使用参考图编辑；
- 独立 OCR：为文字提取单独配置 OCR 服务；关闭时沿用识图设置。
-
🧩 浏览和安装社区插件 by @shanyuhai123：
现在可以在插件市场中浏览和安装社区贡献的插件，扩展你的工作台。安装前可以查看所需权限，安装包与收录记录不一致时会提示风险。
如果你也有插件想分享，欢迎向 useLexora/plugins 提交 PR。源码可以保留在自己的仓库中，发布安装包后提交目录条目即可。
⚡ 改进
-
⚙️ 调整 CodeMode 并发与排队方式 by @shanyuhai123：
默认最多同时运行 8 个脚本。你可以在运行设置中调整所有任务共享的并发上限，并选择容量满时排队等待或立即返回繁忙。修改立即生效，降低上限不会打断正在运行的脚本；也可以在确认资源占用风险后取消数量限制。 -
🌐 管理应用更新与插件下载代理 by @QAyong：
可以在网络代理设置中添加、编辑、排序和启停 GitHub 下载代理。软件更新包与插件包会按顺序尝试下载，全部失败后使用原地址；暂停使用代理不会清空列表，也不影响其他网络请求。 -
💬 选择工作中消息的默认发送方式 by @shanyuhai123：
在运行设置中选择「排队」或「补充当前工作」：前者等当前工作结束后处理，后者在执行边界加入当前工作。设置只影响之后发送的消息，不影响回答待处理问题。 -
📎 窄窗口输入更从容 by @QAyong, @shanyuhai123：
输入工具栏会根据窗口宽度逐步精简空间、模型与权限控件的显示，减少分屏时的拥挤；同时修正权限按钮的背景、对齐与图文间距。附件较多时，可以使用鼠标滚轮横向浏览。 -
📂 文件命名出错后可以直接修改重试 by @shanyuhai123：
新建文件提供默认名称；创建或重命名遇到同名冲突时，提示会显示在原对话框中，并保留已输入的名称，不必重新开始。移动文件、复制路径等操作失败时也有更清楚的提示。 -
📈 用量趋势查看更顺畅 by @shanyuhai123：
优化日期查看、时间范围切换与加载提示，让你更连贯地浏览不同时段的用量记录。 -
📊 在插件中查看任务统计与上下文用量 by @shanyuhai123：
支持相关能力的插件现在可以展示当前任务的轮次、步骤、Tokens、平均输出速度与上下文占用，并随任务进展更新。安装时会单独展示所需的数据读取权限。
🛠️ 修复
-
🧵 命令并行执行与状态显示 by @shanyuhai123：
修复命令并行调度和执行状态显示问题。现在可以更清楚地区分排队、等待资源和实际执行，并查看资源被哪个操作占用；超时提示也会区分「等待资源超时，命令未执行」与「命令执行超时」。 -
🔌 模型列表、能力识别与错误提示 by @shanyuhai123：
修复兼容服务模型列表同步异常、可用模型被误标为不可用的问题，改善 Magpie 等服务的模型能力识别，并隐藏选择列表中的禁用模型。模型请求失败时，也能更准确地区分权限、额度与连接问题。 -
💾 备份期间的服务状态 by @shanyuhai123：
修复备份暂停本地服务时的时序问题。备份期间会明确提示暂时不可用，避免将正常备份过程误认为服务异常。
🧩 插件
- 📟 DSH 状态栏 by @shanyuhai123：
在任务输入区下方显示 DSH 风格状态栏，一眼查看 对话轮数、模型步数、平均输出速度、Token 用量、缓存命中率与上下文占用估算。主题与语言跟随 Lexora，窄窗口自动精简显示，可在插件市场中安装。
English
You can now capture ideas with voice input and listen to companion replies. Image understanding, generation, and OCR also have centralized settings, so you no longer need to keep switching conversation models to work with images.
Explore community plugins in the Plugin Marketplace to add features you enjoy. This release introduces a DSH-inspired task status bar, alongside refinements to downloads, concurrent execution, the composer, and file operations.
✨ Added
-
🎙️ Voice Input & Reply Read-Aloud by @shanyuhai123:
Record speech in the composer, turn it into text, and edit it before sending. You can also listen to companion replies:- Voice input: Choose a transcription service and, when supported, a recognition language. Microphone and transcription tests are available;
- Read-aloud: Use system voices or a speech synthesis model, with voice selection, playback speed, and previews where supported;
- Stop anytime: Stop recording or playback whenever you like. Recognized text is retained after service interruptions. Before first use, configure Settings → Speech and allow microphone access.
-
🖼️ Centralized Image Understanding, Generation & OCR Settings by @shanyuhai123:
Choose how images are handled under Settings → Models:- Understanding: Use the current model or select image-capable models and plugin services to understand and compare images, extract text, and read document images;
- Generation: Choose generation services and supplementary models, with size, quality, background, and reference-image editing options where supported;
- Separate OCR: Configure a dedicated OCR service for text extraction, or follow image-understanding settings when disabled.
-
🧩 Browse & Install Community Plugins by @shanyuhai123:
Browse and install community-contributed plugins in the Plugin Marketplace to extend your workbench. Review requested permissions before installation, with a warning if package contents differ from the catalog record.
Have a plugin to share? Submit a pull request to useLexora/plugins. Keep the source in your own repository, publish an installation package, and contribute its catalog entry.
⚡ Improved
-
⚙️ Adjust CodeMode Concurrency & Queueing by @shanyuhai123:
Up to 8 scripts run simultaneously by default. In runtime settings, adjust the shared limit across tasks and choose whether new calls queue or immediately report busy when capacity is full. Changes apply immediately; lowering the limit does not interrupt running scripts. Unlimited concurrency is available after acknowledging resource risks. -
🌐 Manage Download Proxies for Updates & Plugins by @QAyong:
Add, edit, reorder, enable, or disable GitHub download proxies in network settings. Update and plugin downloads try enabled proxies in order before falling back to the original URL. Pausing proxies preserves your list and does not affect other network requests. -
💬 Choose How Messages Sent During Work Are Handled by @shanyuhai123:
Choose Queue or Supplement current work in runtime settings. Queued messages wait until current work finishes; supplements join at execution boundaries. The preference applies only to subsequently sent messages and does not affect answers to pending questions. -
📎 A More Comfortable Composer in Narrow Windows by @QAyong, @shanyuhai123:
Space, model, and permission controls progressively compact as the window narrows, reducing crowding in split views. Permission-button backgrounds, alignment, and icon spacing were also corrected. Use the mouse wheel to scroll horizontally through longer attachment lists. -
📂 Correct File Names Without Starting Over by @shanyuhai123:
New files receive a default name. Naming conflicts appear in the original dialog with your input preserved, so you can correct the name and retry immediately. Failures when moving files, copying paths, and performing other operations also provide clearer feedback. -
📈 Smoother Usage Trend Browsing by @shanyuhai123:
Improved date browsing, period switching, and loading feedback make it easier to explore usage records across different reporting periods. -
📊 View Task Statistics & Context Usage in Plugins by @shanyuhai123:
Compatible plugins can now display turns, steps, Tokens, average output speed, and context usage for the current task, updating as work progresses. Required data-access permissions are shown separately during installation.
🛠️ Fixed
-
🧵 Concurrent Commands & Execution Status by @shanyuhai123:
Fixed concurrent command scheduling and status display. Queueing, resource waits, and actual execution are now easier to distinguish, including which operation is holding a resource. Timeout messages also distinguish commands that never started from commands that timed out during execution. -
🔌 Model Lists, Capability Detection & Error Messages by @shanyuhai123:
Fixed compatible-service model synchronization and available models being incorrectly marked unavailable. Improved capability detection for services such as Magpie and hid disabled models from selection lists. Request failures also distinguish permission, quota, and connection issues more accurately. -
💾 Service Status During Backups by @shanyuhai123:
Fixed timing issues when pausing the local service for backups. Temporary backup-related unavailability is now clearly identified rather than appearing as an unexpected service failure.
🧩 Plugins
- 📟 DSH Status Bar by @shanyuhai123:
Adds a DSH-inspired status bar below the task composer, showing conversation turns, model steps, average output speed, Token usage, cache hit rate, and estimated context usage at a glance. It follows Lexora’s theme and language, with a compact layout for narrow windows. Available in the Plugin Marketplace.
📥 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.