A macOS menu bar translator: copy text or an image → press the hotkey (⌥⌘T) → a floating panel streams the translation.
English · 简体中文
A gloss is the line of small type written between the lines of a foreign text — an interlinear gloss, from Greek γλῶσσα, "tongue, language." The other word spelled the same way means shine, polish: translation and polish in one word. The icon draws exactly that — a glowing golden gloss line between two faint lines of ink. The five house rules are in CLAUDE.md (Chinese).
swift run # develop
Scripts/bundle.sh # package dist/Gloss.app (ad-hoc signed)
open dist/Gloss.app # everyday useRequires macOS 14+ and a Swift 6 toolchain (Xcode 16+). You can also open Package.swift in Xcode directly.
Menu bar icon → 「设置…」 (Settings — the app's own UI is Chinese for now):
- Base URL / model — any OpenAI-compatible service (DeepSeek, OpenAI, Ollama, a relay…), saved as you type. Keep several providers side by side and switch between them from the menu bar.
- API key — press 「保存」 (Save) and it goes into the Keychain, never into a plaintext file.
- Language — you set only what to translate into; Simplified Chinese by default. The source language is never declared: the model recognizes it on its own, and naming it would only constrain it.
- Hotkey — re-recordable, ⌥⌘T by default.
- 「图片翻译显示原图」 (show the source image) — on: the picture is laid out at natural size with the translation overlaid line by line, and the panel grows to fit; off: translation only.
Zero system permissions — no Accessibility, no Screen Recording, no Full Disk Access.
Copy text or an image, press the hotkey. The panel opens near the pointer and fills in as it streams, translating into the language set in Settings (Simplified Chinese by default); the source language is left to the model, and code blocks are left alone.
When the clipboard holds an image — a screenshot, a chat log, a picture from a web page — the panel shows the picture first while Apple's Vision framework reads the text and the position of every line, locally: no permissions, no cost, and Vision decides the language itself. Those lines then go out as one numbered batch and come back onto the picture in place, each translation over its own line, lighting up row by row as the stream arrives. That is the interlinear gloss, literally. The image is shown at natural size — exactly as large as it was on screen, neither scaled up nor down — and the panel grows with it; at 80% of the screen's visible area it stops growing and the image area scrolls instead. A translated line only grows rightward, so it never covers the line below; anything too long for the picture is repeated in full in the content area underneath, readable and selectable. Turn off 「图片翻译显示原图」 if you only want the text.
Changing the language: pick another target in Settings and the text on the panel is re-translated at once, no confirmation needed. It re-translates what is on screen, not whatever the clipboard holds by now; for an image, recognition doesn't run again — the lines are still there, they just go out once more in another language.
Closing the panel: press the hotkey again, or click × in the top-right corner. The panel stays put — clicking outside doesn't dismiss it, so you can keep the translation open while you work. It is a non-activating panel and never takes focus from the app in front, which is why Esc doesn't close it: the key never reaches it. A choice, not a defect.
- Translating a selection in place (needs the Accessibility permission — v2 at the earliest)
- History, rewriting, summarizing
- A generated
.icnsapp icon (the master isDesign/AppIcon.svg)
MIT © ZhangFan