Changelog

Small, clear updates.

1.6.1

Icons and menu order

  • Refined some icons and the order of the menu bar items.
1.6.0

A scene for where you write

  • Settings has a new choice: Scene. Standard, Casual for posts on X, Formal for email and documents. In Casual, first person, contractions and short fragments stay as they are and are not marked as errors. Formal goes the other way. Choose once; every translation and grammar check follows.
  • Instructions inside your text are translated, not followed. A line like "keep it short and natural" is treated as part of the text. To steer a result, use the field under it.
1.5.2

The panel lands where you're reading, and Settings got a new window

  • Select text, press ⇧⌘G, and the panel opens right under the selection — or above it when there's no room below — instead of wherever the pointer happened to be. Works in native apps, Safari and Chrome.
  • Settings is a new window: a sidebar, one card per group, translucent like the panel. The AI Provider page is down to the essentials with a single Validate & Save, and shortcuts can be reset to their defaults.
1.5.1

A connection-test fix, and brew install

  • Testing a custom OpenAI connection no longer fails on gpt-5 or the o-series. Graker was sending max_tokens, which those models reject; it doesn't send it at all now. Grammar check and translate were never affected.
  • Graker is on Homebrew: brew install --cask eryouhao/tap/graker.
1.5.0

A wider panel, a friendlier toolbar

  • The panel can go wide now: one click switches between 440 and 840 points, so a long translation reads in a few lines instead of a tall column. Your choice sticks.
  • The toolbar got a redo. On macOS 26 the mode switch is real Liquid Glass, the model picker moved next to it, and a Check / Translate button with a Return hint sits on the right — you always know what Enter will do.
  • Paste into the input and it just runs — pasted text is usually a complete sentence. Typing by hand never triggers anything.
  • "Quick Translate" is now just "Translate", and menu items dropped their trailing ellipses. Fewer words, same tools.
1.4.0

Markdown, properly rendered

  • Quick-translate results now render as real Markdown: headings, lists and code blocks take shape while the text streams in, and your original line breaks survive.
  • The panel grows with long results, up to 80% of the screen, and streaming updates got smoother.
  • Custom providers no longer send a fixed temperature. Endpoints that only accept their default, like Kimi k3, just work now.
  • Graker now requires macOS 15. On macOS 13 or 14, 1.3.0 keeps working and won't nag you to update.
1.3.0

Speaks your language

  • The interface now speaks English, 简体中文 and 繁體中文. Follow the system or pick one in Settings — switching applies instantly, no relaunch.
  • The menu bar menu was redone: icons on every action, and your recorded shortcuts shown natively on the right, always in sync.
  • Settings moved to a native preferences layout with four tabs: General, AI Provider, Shortcuts, About.
1.2.0

Auto-updates and a faster panel

  • Graker now keeps itself up to date: quiet checks in the background, one click in the menu bar when a new version is ready.
  • The quick-translate panel was rebuilt: results stream in, requests can be cancelled any time, and providers run side by side.
  • New ⌘1 / ⌘2 / ⌘3 shortcuts: translate clipboard, open quick translation, open grammar check.
1.1.0

Format-preserving translation

  • New format-preserving translation: select text or a rich HTML email and translate it without breaking images, links, or layout.
  • New Open Recent: the last 10 grammar checks from this run can be reopened from the menu bar.
1.0.10

A smoother default shortcut

  • The default grammar-check shortcut is now Command + Shift + G.
  • Fewer Keychain prompts on first launch, and a refined follow-up input below the result.
1.0.9

Context-aware refinement

  • New contextual grammar refinement: add your real intent and Graker keeps revising based on the original text and the previous result.
  • Legacy or incomplete model responses now also get a readable explanation.
1.0.8

Bring your own provider

  • New Custom OpenAI-compatible provider with configurable Base URL and model.
  • New MiMo cluster selection and API endpoint display.