更新日志（中文）
[1.3.75] - 2026-10-27
新增
- 在 API 服务模型中加入 GPT-6.1 Terra，并支持标准、快速、Flex 和长上下文价格的费用估算。
- 新增应用背景设置：提供四款渐变预设、本地图片选择与预览，支持调整强度、模糊和图片位置。尚无图片时“自定义”卡片显示加号，点击即可选图；修改立即生效并在本机自动保存，仅保留“恢复默认”操作，不改变账号卡片和控件样式。
- 根据实际 API 地址为 Codex 账号卡片应用对应主题，包括 APIKEY.FUN、ai.ahzm.top 和 yunmoor.com，保留卡片内容与操作。
- 账号卡片与仪表盘在原有位置使用统一的额度时间样式显示 Codex 最近重置到期时间，临近到期时突出提醒；到期明细持久缓存，按账号分散在约 6 小时刷新，并限制请求并发，打开详情时立即更新共享缓存。感谢 @iamvicliu（#2724）。
- 补充 Antigravity 的 Gemini 4 Argon、Gemini 3.8 Flash 和 Claude 5.5 模型及分组，保留旧模型、档位区别和自定义分组设置；默认自动切号仅检查当前账号实际返回的模型分组。感谢 @Chain813（#2915）。
修复
- 修复 Windows 等平台本地 API Key 账号使用客户端默认模型列表时，Sol 6.1 与 Astra 缺少超高速选项的问题；保留模型列表和自定义目录，选择与请求继续使用原生速度流程。
- 修正 New API 账号卡片额度条与剩余额度文字方向相反的问题，让进度条与可用／总额度使用相同口径。感谢 @specetator（#2923）。
- 取消单个 API 请求时，不再因取消消息被误判为超时而重启共享 Sidecar、打断其他请求。感谢 @limaomei-pixel（#2909）。
- 将展开后的协作工具调用恢复为声明中的命名空间与名称，避免子代理协作出现不支持的工具调用错误。感谢 @jamie950315（#2901）。
- 防止共享客户端模型目录中的第三方模型被分配给不支持它们的 ChatGPT 账号，同时保留官方 Reserve 模型调用能力。感谢 @jamie950315（#2902）。
- Gemini 流式响应中的参数错误仅使当前请求失败，不再错误冷却账号；调用方仍能看到原始上游错误。感谢 @jamie950315（#2904）。
- Kiro 企业账号设备授权与 Token 刷新使用 AWS IAM Identity Center 要求的 JSON 请求格式。感谢 @peisencn（#2799）。
- Claude Desktop Gateway 切换账号时保留用户的隐私、权限、网络限制及模型设置，写入失败时保留原有可用配置；切号在后台执行，不再等待旧网关的进行中请求结束。感谢 @fflow2023（#2907）。
- Antigravity 额度请求在缺少旧版安装元数据时使用后台探测的版本缓存，保留协议版本下限。感谢 @moyuan10086（#2900）。
- 数据管理设置首次进入后保留已挂载内容，减少反复切换页签引起的备份扫描，其他页签不提前加载数据管理内容。感谢 @xhrxgr（#2740）。
- 注入的速度菜单使用官方客户端内置的超高速说明翻译，避免被英文文案覆盖；保留速度档位选择和请求参数。
- 取消本地开发任务时同步清理开发版应用与前端服务，避免窗口残留和终端继续输出日志，同时保留其他应用实例。
- 减少 Windows 进程查询暂时不可用时的 Codex 启动误报：优先确认启动器返回的进程，会话检查无法安全开始时保留历史并跳过本次维护；启动结果不明确时显示待确认并提供只读状态检查，检测到目标实例运行后转为成功，避免引导重复启动。
- 统一 Codex 启动路径缺失时的引导，覆盖账号切换、API 服务、实例、CLI 和原生菜单入口。路径弹框置于其他弹框上方，支持选择程序、检测运行中应用及设置切号后自动启动；保存后继续原操作，取消或失败时不重复启动。
变更
- “修复会话索引”支持在 Codex 客户端和 CLI 运行时直接执行：备份并重置异常索引，保留聊天记录，提交前检查并发变化，并提示完成后重启客户端。
- 清空已发布公告，客户端成功刷新后不再显示旧公告或自动弹出公告。
Changelog (English)
[1.3.75] - 2026-10-27
Added
- Add GPT-6.1 Terra to API Service models and cost estimates, including standard, Fast, Flex and long-context pricing.
- Add app background customization with four gradient presets, local image selection, a preview, intensity, blur and image positioning. The Custom tile shows a plus sign for choosing an image when none is saved. Changes take effect immediately and save automatically on this device, with a single Restore default action; account cards and controls keep their appearance.
- Apply Codex account-card skins based on the actual API endpoint, including APIKEY.FUN, ai.ahzm.top and yunmoor.com, while preserving account content and actions.
- Show the next Codex reset-credit expiry in its original position on account cards and the dashboard using the quota time style, highlighting approaching expiry; persist details, stagger refreshes around six hours with bounded concurrency, and update the shared cache immediately when details are opened. Thanks @iamvicliu (#2724).
- Add Antigravity Gemini 4 Argon, Gemini 3.8 Flash, and Claude 5.5 model groups while preserving older models, effort distinctions, and custom group settings; default automatic rotation checks only model groups actually reported by the current account. Thanks @Chain813 (#2915).
Fixed
- Restore Ultrafast for supported Sol 6.1 and Astra models when local API Key profiles use the client's default model list, including Windows. Preserve the model list and custom catalogs, with selection and requests using the native speed flow.
- Align New API account-card quota bars with their remaining-quota labels, so the bar and available/total values use the same basis. Thanks @specetator (#2923).
- Prevent canceling one API request from restarting the shared Sidecar and interrupting other requests when cancellation is mistaken for a timeout. Thanks @limaomei-pixel (#2909).
- Restore flattened collaboration-tool calls to their declared namespace and name, preventing unsupported-call errors during agent coordination. Thanks @jamie950315 (#2901).
- Prevent third-party models in the shared client catalog from being routed to ChatGPT accounts that do not support them, while retaining official Reserve model access. Thanks @jamie950315 (#2902).
- Keep Gemini invalid-argument stream errors scoped to the affected request instead of incorrectly cooling down accounts; retain the upstream error for the caller. Thanks @jamie950315 (#2904).
- Send Kiro enterprise device-authorization and token-refresh requests in the JSON format expected by AWS IAM Identity Center. Thanks @peisencn (#2799).
- Preserve Claude Desktop Gateway privacy, permission, network, and model preferences when switching accounts, retaining the working configuration if a write fails; switch in the background without waiting for the old gateway’s in-flight request to finish. Thanks @fflow2023 (#2907).
- Use background-detected version metadata for Antigravity quota requests when legacy installation metadata is unavailable, retaining the minimum protocol version. Thanks @moyuan10086 (#2900).
- Keep data-management settings mounted after their first visit to reduce repeated backup scans, without loading them when other settings tabs are opened. Thanks @xhrxgr (#2740).
- Use the official client's localized Ultrafast description in injected speed menus instead of overriding it with English text; preserve speed-tier selection and request parameters.
- Stop the development app and frontend server together when cancelling a local development run, preventing leftover windows and terminal logs while preserving unrelated app instances.
- Avoid false Codex launch failures when Windows process queries are temporarily unavailable. Confirm the process returned by the launcher, preserve history when a session check cannot begin safely, and show an unconfirmed launch with read-only status checks instead of prompting duplicate launches; a detected running instance updates the dialog to success.
- Unify missing Codex launch-path recovery across account switching, API Service, instances, CLI and native menu actions. Show the path dialog above other dialogs, support app selection, running-app detection and launch-on-switch settings, and resume the original operation after saving without duplicate launches on cancellation or failure.
Changed
- Allow Repair chat index while Codex clients and CLI sessions are running. Back up and reset invalid indexes without changing chat records, detect concurrent changes before committing, and prompt a client restart afterward.
- Clear all published announcements so old notices and automatic announcement popups disappear after a successful client refresh.