AI时代，语音输入更高效！

• 语音输入支持账号渠道和自定义模型，最多配置三个识别渠道，提供文本优化与多语言自动翻译。
• 优化语音识别后的文字回填，减少长文本漏字、重复输入和换行误发送；支持复制与回听上一次结果。
• 灵动岛支持拖动、应用图标跟随、主题与音波配色；语音快捷键可选长按或点按，并提供动画使用指引。
• 划词后可翻译或提问，提问内容带入 AI 助手草稿，编辑后手动发送。
• 应用桌面优先显示上次的应用列表，后台刷新时保持可操作，重启后更快进入常用工具。
• 支持折叠导航、页面前进/后退和中右区域宽度拖拽，统一模型管理与账号管理的卡片布局。
• 更新日志随窗口大小自适应，放大顶部文字，修复折叠模式开关闪烁，优化设置与多语言文案。
• 完善更多工具的账号与模型切换体验，更新应用目录和推荐列表。

语音、翻译和模型切换能力取决于所选渠道的支持。

More efficient voice input for the AI era!

• Use saved accounts or custom models for voice input, with up to three recognition channels, text refinement, and automatic multilingual translation.
• Improve insertion of recognized text, reducing missing text, duplicates, and accidental sends from line breaks; copy or replay the latest result.
• Customize the draggable voice island with active-app icons, themes, and waveform colors; choose hold or toggle shortcuts and follow the animated first-use guide.
• Translate selected text or bring a question into an AI Assistant draft for editing and manual sending.
• Show the last app list immediately on startup and keep it usable during background refresh, so you can reach familiar tools sooner.
• Use collapsible navigation, page back/forward, and adjustable center/right widths, with consistent model and account card layouts.
• Resize the changelog with the window, improve title-bar readability, fix flickering collapsed switches, and refine settings and translations.
• Improve account and model switching for more tools and refresh the app catalog and recommendations.

Voice, translation, and model switching depend on the selected channel's capabilities.

6.0.7 新增语音输入与划词翻译/提问，完善智能接入、AI 助手和桌面布局。

• 新增语音输入，支持账号接入和自定义语音模型，按顺序使用最多三个识别渠道。
• 支持流式识别、麦克风自动增益、文本优化及多语言自动翻译。
• 完善原输入框回填与受支持控件的后台写入，减少长文本漏字、重复输入和换行误发送；可复制或回听上一次结果。
• 提供可拖动灵动岛、应用图标跟随、独立主题与音波配色，并新增流彩背景。
• 语音快捷键默认关闭，可自行设置长按或点按；新增首次使用动画指引。
• 新增划词翻译/提问浮条，提问可带入 AI 助手草稿，编辑后手动发送。
• 智能接入支持更多工具在原生客户端内切换模型，完善账号登录、取消与重试处理。
• 修复 Trae / Trae CN、MiniMax、ZCode 等模型与语音接入兼容问题。
• 新增可选的凭据保护，隐藏请求内容中的密钥和密码，完善并发与响应还原，并更新加密和压缩安全依赖。
• AI 助手支持检索本机已验证的任务经验，改善多语言划词提问的上下文理解。
• 模型管理优先获取上游模型 ID，清理过时预置，保留必要预设和手动输入。
• 新增 Claude Duet 应用入口，更新应用目录、桌面小技巧和多语言文案。
• 新增折叠导航、页面前进/后退和中右区域拖拽，统一模型与账号卡片的宽度规则。
• 优化协议地址排列、更新提示、意见反馈入口、图标及动画细节。

语音、模型切换及翻译能力取决于所选渠道和上游模型的支持。

6.0.7 adds voice input and selected-text translation/questions, and improves Smart Access, AI Assistant, and desktop layouts.

• Adds voice input through saved accounts or custom speech models, with up to three recognition channels tried in order.
• Supports streaming recognition, automatic microphone gain, text refinement, and multilingual translation.
• Improves insertion into the original field and supported background controls, reducing missing text, duplicates, and accidental sends from line breaks; copy or replay the latest result.
• Adds a draggable voice island, active-app icons, independent themes and waveform colors, and a flowing color background.
• Voice shortcuts start disabled and support configurable hold or toggle modes; adds an animated first-use guide.
• Adds a selected-text translation/question toolbar. Questions prefill an AI Assistant draft for editing and manual sending.
• Extends native in-client model switching in Smart Access and improves account login, cancellation, and retries.
• Fixes model and speech integration compatibility for Trae / Trae CN, MiniMax, ZCode, and other tools.
• Adds optional credential protection for keys and passwords in request content, improves concurrency and response restoration, and updates security-sensitive crypto and compression dependencies.
• Lets AI Assistant retrieve verified task experience from the current machine and better understand multilingual selected-text questions.
• Fetches model IDs from upstream catalogs instead of stale built-in lists while retaining required presets and manual entry.
• Adds a Claude Duet app entry and updates the app catalog, desktop tips, and translations.
• Adds collapsible navigation, page history, and a resizable center/right layout, with shared sizing rules for model and account cards.
• Refines protocol addresses, update indicators, the Feedback entry, icons, and animations.

Voice, model switching, and translation capabilities depend on the selected channel and upstream model.

6.0.6 汇总 6.0.5 的功能升级，并补充本次紧急修复。

• 修复豆包工作 / Dola 切换账号及闪窗，修正豆包每日签到请求。
• 批量领取与刷新失败统一汇总提示，避免重复弹窗。
• 千问办公移除手动领取入口，保留奖励余额与到期时间。
• Claude Code 的 Fable 档位同步所选第三方模型与 1M 设置。
• 统一额度进度条配色，余额为 0 时显示红色。
• 智能接入支持独立多路由、并发请求与顺序备用模型，减少无效探测。
• 完善额度不足和自动切换，避免单模型耗尽影响同账号其他模型。
• 总开关关闭时锁定路由，其他已启用路由橙色加粗标记且布局不抖动。
• 接入信息支持复制和空密钥提示，添加条目不再打断测速。
• 新增 DuMate、豆包工作、Dola、QwenWork、Devin 集成，移除 WorkBuddy 系列接入风险标记。
• Claude、Manus、Cue 等改用独立登录窗，登录时限延长至 10 分钟。
• 自动领取奖励默认关闭，启用后启动时及每小时领取；完善奖励到期、额度恢复与订阅显示。
• ZCode 新增 web_search 与 image 开关。
• WorkBuddy、MiMo、MiniMax、Qoder、DeepSeek Harness 补充相应 image、1M 与原生推理配置，保留手动设置。
• 完善 Codex 官方模式恢复、Hermes 配置、WorkBuddy 新任务模型选择与模型目录。
• AI 助手新增轻量逆向分析与诊断，辅助安装、配置和修复。
• 更新 TLS 安全依赖和检查工具，清理弃用依赖。
• 持续打磨用户体验、交互细节与精美 UI 动画。

图片、联网、1M 和推理能力以实际购买的上游支持为准。

6.0.6 combines the improvements from 6.0.5 with this hotfix release.

• Fixes account switching and flashing windows in Doubao Work / Dola, and corrects Doubao daily check-in requests.
• Summarizes batch claim and refresh failures once instead of showing repeated dialogs.
• Removes manual reward claims from QwenWork CN while retaining balances and expiry dates.
• Maps the Claude Code Fable tier to the selected third-party model and 1M setting.
• Unifies quota colors and shows zero balances in red.
• Adds independent Smart Access routes, concurrent requests, and sequential standby selection to reduce unnecessary probes.
• Improves quota handling and failover so one exhausted model does not block other models on the same account.
• Locks route switches when the master switch is off and marks other enabled routes in bold orange without layout shifts.
• Adds copy actions and empty-key guidance; adding entries no longer interrupts speed tests.
• Adds DuMate, Doubao Work, Dola, QwenWork, and Devin integrations; removes WorkBuddy account-access risk badges.
• Uses isolated login windows for Claude, Manus, Cue, and other supported tools, with a 10-minute deadline.
• Adds default-off reward claims at startup and hourly, with better reward expiry, quota recovery, and subscription displays.
• Adds web_search and image switches for ZCode.
• Extends supported image, 1M, and native reasoning settings across WorkBuddy, MiMo, MiniMax, Qoder, and DeepSeek Harness while preserving manual settings.
• Improves Codex official-mode restoration, Hermes configuration, WorkBuddy model selection for new tasks, and the model catalog.
• Adds lightweight reverse analysis and diagnostics to assist with installation, configuration, and repairs.
• Updates TLS security dependencies and development checks, and removes deprecated dependencies.
• Continues to refine the user experience, interaction details, and polished UI animations.

Image, search, 1M, and reasoning capabilities depend on the upstream model you purchase.

6.0.5 升级智能接入、账号登录与第三方模型配置，带来更顺畅的自动切换体验。

• 智能接入支持多组独立路由与编辑；强化并发请求和共享上游处理，按顺序只准备一个可用备用模型，减少无效探测与反复切换。
• 完善典型错误、额度不足和自动切换处理，避免一个模型额度耗尽影响同账号下其他可用模型。
• 总开关关闭时锁定并淡化路由开关；已开启的其他路由以橙色加粗标记，切换时保持布局稳定。
• 优化路由编辑：协议地址、模型 ID 与 API Key 支持复制，密钥为空时明确提示；添加新条目不再打断当前测速，新条目留待下次测试。
• 新增 DuMate、豆包工作、Dola、QwenWork 与 Devin 的账号集成，补充相应客户端的模型接入与智能路由能力；移除 WorkBuddy / WorkBuddy AI 在账号接入列表中的高风险标记。
• Claude、Manus、Cue 等适用工具改用独立内置登录窗，登录时间延长至 10 分钟；改善登录完成识别、取消清理和客户端进程处理。
• 新增默认关闭的“自动领取奖励”，启用后启动时及运行期间每小时复用现有领取功能；完善 WorkBuddy 奖励到期时间、额度恢复信息与订阅显示。
• ZCode 新增 web_search 与 image 开关；WorkBuddy / WorkBuddy AI、MiMo 桌面端 / Code CLI、MiniMax 桌面端 / Code CLI、Qoder / Qoder CN、DeepSeek Harness 补充 image、1M 上下文及原生推理档位支持，并保留已有手动配置。
• 修复 Codex 恢复账号模式时的设置保留与官方路由；改善 Hermes 原生提供商配置、WorkBuddy 新任务模型选择及模型身份保留，更新模型目录。
• AI 助手新增轻量逆向分析与诊断能力，帮助解决更棘手的安装、配置和修复问题。
• 更新 TLS 证书校验与根证书依赖，修复近期公布的证书信任漏洞；更新开发检查工具并清理弃用依赖。
• 持续深度优化用户体验，继续增加大量细节与精美 UI 动画效果。

模型的图片、联网、1M 上下文与推理能力取决于所购上游的实际支持；新增开关用于配置客户端能力。

6.0.5 upgrades Smart Access, account login, and third-party model configuration for smoother automatic switching.

• Adds independently editable route groups, strengthens concurrent requests and shared upstream handling, and prepares one usable standby in sequence to reduce unnecessary probes and repeated switching.
• Improves common error, quota exhaustion, and failover handling so one exhausted model does not block other usable models on the same account.
• Locks and dims route switches when the master switch is off; highlights other enabled routes in bold orange without shifting the layout.
• Adds copy actions for protocol URLs, model IDs, and API keys, with clear empty-key validation. Adding entries no longer interrupts an ongoing speed test; new entries wait for the next test.
• Adds account integrations for DuMate, Doubao Work, Dola, QwenWork, and Devin, with corresponding client model and Smart Access support. Removes the high-risk badge for WorkBuddy / WorkBuddy AI in the account access list.
• Uses isolated embedded login windows for supported tools including Claude, Manus, and Cue, with a 10-minute deadline and improved completion detection, cancellation, and client process handling.
• Adds an optional, default-off automatic reward claim setting that reuses existing claim actions at startup and hourly. Improves WorkBuddy reward expiry, quota recovery information, and subscription displays.
• Adds web_search and image switches for ZCode; extends image, 1M context, and native reasoning settings across WorkBuddy / WorkBuddy AI, MiMo desktop / Code CLI, MiniMax desktop / Code CLI, Qoder / Qoder CN, and DeepSeek Harness while preserving existing manual settings.
• Preserves user settings and official routes when restoring Codex account mode; improves Hermes native provider configuration, WorkBuddy model selection for new tasks, model identity preservation, and the model catalog.
• Adds lightweight reverse analysis and diagnostics to the AI assistant to help with tougher installation, configuration, and repair problems.
• Updates TLS certificate verification and root certificate dependencies to fix recently disclosed trust vulnerabilities; refreshes development checks and removes deprecated dependencies.
• Continues to refine the user experience with many thoughtful details and polished UI animations.

Image, search, 1M context, and reasoning availability depend on the upstream model you purchase; these switches configure the client's declared capabilities.

6.0.4 修复账号套餐显示与界面细节，更新第三方模型目录和配置支持。

• 修复 ZCode 已确认无订阅时显示横杠的问题，区分“未开通”、体验套餐与未知状态。
• 修复 Open Design 免费账号的 Free 等级识别，并在查询返回异常时保留已有数据。
• 修复本地模型页面首次打开时，硬件检测完成后内容向上跳动的问题。
• 修复智能接入卡片模式中，暂停卡片内的穿透虚线流动方向。
• 更新第三方 Codex 模型能力目录，新增 Step Plan 配置支持，补充 MiniMax、Qwen 等平台的模型信息。

更新后，已有 ZCode / Open Design 账号请手动刷新一次，以更新缓存中的套餐标签。

6.0.4 fixes account plan labels and UI details, and updates third-party model catalogs and configuration support.

• Fixes ZCode showing a dash for a confirmed absence of a subscription, distinguishing No plan, trial, and unknown states.
• Recognizes Free accounts in Open Design and preserves cached data when the subscription response is invalid.
• Prevents the local model welcome content from jumping upward when hardware detection finishes on first entry.
• Corrects the direction of dashed lines passing through paused cards in Smart Access.
• Updates third-party Codex capability catalogs, adds Step Plan configuration support, and refreshes model information for MiniMax, Qwen, and other providers.

After updating, refresh saved ZCode / Open Design accounts once to update their cached plan labels.

6.0.3 优化智能接入的测速与超时处理，更新免费模型目录和平台图标。

• 智能接入列表新增 [测试速度]，等待图标靠右显示，测速时使用光丝波束动画。
• 模型管理与智能接入统一采用五次连接采样，去除最高、最低值后取平均；同轮相同地址共享结果。
• 统一延迟颜色与“超时”提示，完善中途取消和迟到结果的清理。
• 取消智能接入的 20 个模型上限，允许 API 密钥留空。
• 增加首段内容与流式输出的静默超时保护，保留整体重试预算。
• 新增 Agnes AI 免费入口，完善 ModelScope 的接入地址与首页，更新平台图标并适配明暗主题。
• 优化账号额度标记，避免空日志触发本地模型页面滚动。

6.0.3 improves connection speed tests and timeout handling in Smart Access, and updates the free model directory and provider icons.

• Adds [Test Latency] to the Smart Access list with a loading icon aligned to the right and a light thread animation while testing.
• Uses five connection samples in both Model Nexus and Smart Access, removes the fastest and slowest, and averages the remaining three. Matching addresses share a result within each test round.
• Aligns latency colors and timeout labels, and improves cancellation and cleanup of late results.
• Removes the 20 model limit in Smart Access and allows API keys to be left blank.
• Adds timeout protection before the first content and when streaming output stalls, while preserving the routing retry budget.
• Adds Agnes AI to the free directory, completes ModelScope endpoints and website links, and refreshes provider icons for light and dark themes.
• Refines account quota markers and prevents empty logs from scrolling the local model page.

6.0.2 优化了 AI 助手及相关页面的界面细节。

• 移除 AI 助手快捷短语、输入框、推理强度控件、模型列表及命令执行卡片的外边框，统一扁平化风格。
• 统一问号、模型选择、推理强度与发送按钮的高度和间距，复用应用桌面的问号样式。
• 优化模型列表的悬停高亮，增加可显示的选项数量，并根据模型名称和 ID 动态调整宽度。
• 移除 AI 生涯统计卡片的可见边框。
• 统一本地大模型安装引擎栏的状态高度，避免扫描完成后的界面抖动。

6.0.2 refines the AI Assistant and related pages.

• Removes outer borders from quick phrases, the message composer, reasoning controls, the model menu, and tool execution cards for a consistent flat appearance.
• Aligns the help, model, reasoning, and send controls with uniform heights and spacing, reusing the help icon from App Desktop.
• Restores model menu hover highlighting, shows more options at once, and adjusts menu width to model names and IDs.
• Removes visible borders from AI Career statistics cards.
• Keeps the local model engine installation bar at a consistent height across states, preventing layout shifts after scanning.

6.0.1 改善了 Claude 桌面端 Code 页的本地对话记录保留。

• 切换 Claude 账号或第三方 API 模型时，自动汇总 Claude 桌面端 Code 与本机 Claude Code 的对话记录，按时间显示并可继续原会话。
• 保留原始聊天正文及归档、删除状态，更新已有会话索引前自动备份。
• 应用桌面为 Claude 桌面端、ChatGPT / Codex CLI、WorkBuddy 国内版及海外版增加本地对话记录保留提示。
• ChatGPT / Codex CLI 同时保留第三方模型需支持 Responses 的接入提示。

6.0.1 improves local conversation history retention in the Code page of Claude Desktop.

• Automatically brings together local Claude Desktop Code and Claude Code conversations when switching Claude accounts or third-party API models, keeping them ordered by activity and available to continue.
• Preserves conversation content and archived or deleted states, with automatic backups before updating existing session indexes.
• Adds local history retention guidance in App Desktop for Claude Desktop, ChatGPT / Codex CLI, and both WorkBuddy editions.
• Keeps the Responses compatibility reminder for third-party models used with ChatGPT / Codex CLI.

6.0.0 改善了 DeepSeek Harness 配置兼容性、Windows 应用识别和自定义入口。

• 修复部分合法 DeepSeek Harness 配置导致模型或账号切换失败的问题，保留其他插件配置、表达式和凭据。
• 修复模型配置失败时直接显示 accountError.format 等内部错误标识的问题。
• 改善 Windows 下 TRAE CN 用户级安装的识别，支持通过注册信息定位自定义安装目录。
• 应用桌面的齿轮按钮增加「自定义」文字，便于配置自定义路径和可执行文件名。
• 关闭智能路由后，母体聊天不再显示或保留该模型选项；调整 Pi 图标留白。

6.0.0 improves DeepSeek Harness configuration compatibility, Windows app detection, and the Customize action.

• Fixed model and account switching failures with supported DeepSeek Harness configuration structures while preserving other plugin settings, expressions, and credentials.
• Configuration failures now show translated messages instead of internal error identifiers such as accountError.format.
• Improved detection of per-user TRAE CN installations on Windows, including custom installation directories resolved from the registry.
• Added a visible Customize label to the gear button in App Desktop for custom executable paths and filenames.
• Disabled Smart Router models are now removed from Mother chat choices and selection. Adjusted Pi icon padding.

5.9.9 优化了账号接入、Claude 模型切换和界面一致性。

• 新增 CodeArts 桌面端接入，支持账号登录与切换、API 模型配置、积分及订阅到期显示、每日积分领取和智能接入。
• 删除已保存账号时同步清理智能接入中的关联项，并完善授权失败后的重新授权流程。
• Claude 桌面端和 Claude Code 改为直接使用模型地址，移除 API Router 开关；桌面端自动处理模型 ID，保留独立的 Web 搜索和 1M 开关。修复快速切换开关可能相互覆盖配置的问题。
• 统一页面操作按钮、重新授权按钮和积分显示，更新 Pi、Hermes Agent、OpenCode 和 Aider 图标。
• 更新退出提示，并在应用桌面增加长按拖动图标排序的提示。

5.9.9 improves account integrations, Claude model switching, and interface consistency.

• Added CodeArts desktop integration with account sign-in and switching, API model configuration, credit and subscription expiry display, daily credit claims, and Smart Access.
• Deleting a saved account now removes its linked Smart Access entries. Improved reauthorization after authentication failures.
• Claude Desktop and Claude Code now connect directly to the configured model endpoint without an API Router toggle. Desktop model IDs are handled automatically, with independent Web Search and 1M controls. Fixed overlapping configuration writes when toggling controls quickly.
• Standardized page actions, reauthorization buttons, and credit labels. Updated the Pi, Hermes Agent, OpenCode, and Aider icons.
• Updated the exit warning and added guidance for holding and dragging app icons to reorder them.