Repository navigation
GoRss 1.0.2
GoRss 1.0.2
简体中文
- 更新三栏阅读界面、文章操作菜单和统一设置页;阅读排版与 AI 配置独立保存。
- 新增按需摘要、结构化全文翻译、原文/译文/双语切换及右键划词翻译。
- 支持自定义 OpenAI 兼容服务与 Magpie,包含真实协议检测和模型测试;API Key 保存在 macOS Keychain。
- 添加有界持久 AI 缓存、已完成批次复用、取消及过期响应防护。
- 优化翻译连接复用、输出预算和结构化响应兼容;修复 Markdown 正文排版。
- 更新双语文档、私密文件忽略规则及依赖许可说明。保留本地 MyGo 选区访问适配及完整 runtime 源码。
安装:Apple Silicon,macOS 12+。下载 ZIP,解压后把 GoRss.app 移到“应用程序”,更新前退出旧版。附 SHA256SUMS.txt。此次提供 ZIP;没有制作新版 DMG。安装包未使用 Developer ID 签名或公证。
验证:Linux 上全项目 go test ./... -count=1、go test -race ./... -count=1、go vet ./... 通过,Darwin ARM64 App 交叉构建成功;ZIP 内容、权限、版本与 SHA-256 已校验。实际 macOS GUI、CoreText、原生对话框、Keychain、真实 AI 服务及系统安装体验未验证。
限制:AI 当前使用 stream: false,完成一批后更新界面;双语跨独立结构块选区受限。远程图片需联网,订阅只提供摘要时不抓取网页全文。AI 可能有误,主动请求才发送正文或选区;本地明文缓存可能包含私人文章派生内容。
English
- Updated the three-pane reader, article action menus, and unified Settings page, with separate reading and AI saves.
- Added on-demand summaries, structured article translation, original/translated/bilingual views, and selection translation from the context menu.
- Added custom OpenAI-compatible and Magpie providers, protocol detection, and real model tests; keys are stored in macOS Keychain.
- Added bounded persistent AI results, completed-batch reuse, cancellation, and stale-response protection.
- Improved translation connection reuse, output budgeting, and structured-response compatibility; fixed Markdown typography.
- Updated bilingual documentation, private-file ignores, and dependency notices. Included the complete local MyGo runtime and selection accessor adaptation.
Install: Apple Silicon, macOS 12+. Extract the ZIP and move GoRss.app to Applications, quitting the old version first. SHA256SUMS.txt is included. This release preparation provides a ZIP; no new DMG was produced. The package is not Developer ID signed or notarized.
Validation: full tests, race detection, and vet passed on Linux; the Darwin ARM64 App cross-build succeeded. ZIP contents, permissions, version, and SHA-256 were checked. Actual macOS GUI, CoreText, dialogs, Keychain, real AI services, and installation behavior were not exercised.
Limits: AI uses stream: false and updates after each complete batch. Selection across independent bilingual blocks is limited. Remote images need connectivity; summary-only feeds do not trigger full-page fetching. AI can be inaccurate. Only explicit requests send article or selection content; local plaintext results may contain private article information.