AI text-selection tool for macOS · Free forever

Select to summon AI —
translation lands straight in Obsidian

Select or double-click, and the floating bar appears on the spot: translate, explain and summarize in real time; copy, search, save to Obsidian or prefill a calendar — results land right where you are. Text in images, videos and scanned documents can be captured by region as well — all handled in place, never locked in a chat window.

Pops in ~0.1s · bring your own model service · OCR & extraction run on-device · No usage tracking

任意应用中的文字

Self-recognition has long been treated as a mark of animal consciousness. The magpie is one of the few animals that recognises itself in a mirror.

翻译解释总结搜索5
2
capture methods
Selection · OCR
12
built-in actions
3 AI · 9 local
10
custom actions
write your own prompts
6
built-in search engines
custom engines supported

Actions

12 built-in actions, in AI and local families

Translate, explain and summarize run through your configured model service; copy, search, open link, compose email, information extraction and Save to Obsidian run locally on your machine. The bar shows 4 actions by default with the rest folded into an expand panel; available actions adapt to the selection — a URL offers Open Link, an email offers Compose Email.

Translate

Picks the translation direction automatically and shows results in real time. Works with AI translation, Baidu Translate and DeepL; a default service can be set.

Explain

Explains terms, concepts, allusions and long sentences — no need to open a separate chat window.

Summarize

Turns a paragraph or a whole page into key points, kept in a standalone window.

Copy / Search

Everyday actions handled on your machine — no configuration, no model quota.

Extract info

Detects URLs, emails, verification codes and phone numbers in the selection. A single hit gets a direct action; multiple hits open a grouped panel.

Open link / Compose email

Select a URL to open it, select an email address to compose, copy phone numbers with one click.

Add to Calendar / Open in Maps

Select text containing a date, time or place — prefill a calendar event or open it in Maps with one click. Information lands where it belongs instead of staying in a window.

Save to Obsidian

Writes the selection straight into your vault from templates: excerpts append to the daily note, task lines go to your to-dos (due dates detected in the selection), and cards are one file each. Entirely local file writes — no plugin needed.

Screen Capture

Text you can't select? Draw a box — it reads

Selection only works on selectable text. Text inside images, video subtitles, scanned documents and app interfaces can be captured by region and recognized on-device, then processed with the same set of actions.

01

Capture

Pick OCR from the hotkey or the menu bar, then drag a region on screen.

02

On-device recognition

Recognition runs locally — the screenshot is never uploaded or stored.

03

Keep processing

Recognized text can be translated, explained or summarized right away — works for images, video subtitles, scanned documents and app interfaces.

04

Pin

Screenshots can be pinned to the screen for re-recognition, copy, search and more.

Four OCR hotkeys

Press a hotkey to enter capture; the matching action runs automatically when recognition finishes. Keys can be changed in settings.

  • OCR capture⌥ S
  • OCR translate⌥ T
  • OCR explain⌥ E
  • OCR summarize⌥ D

Other capture & display rules

  • Drag or double-click both trigger the bar. When an app draws its own UI and selection can't be read, use OCR instead.
  • Screenshots and recognition run in the background; selection pauses automatically during capture.
  • Screenshots can be pinned to the screen for re-recognition — copy, search and AI actions all work.
  • Specific apps can be added to a blocklist; re-selecting the same text never pops the bar twice.

Use cases

Who uses it: anyone who works with words

Writing, editing, translation, research and reading — looking things up, verifying, excerpting and keeping happen at the highest frequency; for people who already keep notes in Obsidian or Notion, collecting also needs a destination. Magpie provides capture and processing actions for exactly that flow; content creation is out of scope.

Writers / translators / language learners

Citations, fact-checking and term explanations are one selection away — no window juggling. Prompts can follow your personal translation style; frequent conversions become custom actions.

Researchers / deep readers / students

Text in scanned PDFs, slides, video subtitles and figures can't be selected — draw a box to recognize it, then translate, explain or summarize, and keep excerpts into your reading notes.

Creators / information keepers

Translate a quote, verify a claim, lift material from a video, image or long article — text turns into usable draft material instead of screenshots lost in your album.

And all the small moments

Verification codes, unfamiliar words, text trapped in screenshots, a line worth keeping — scattered text tasks, one set of actions.

Customization

Models, prompts, actions, search and appearance — everything is configurable

Customizable: model services, action prompts, custom actions, search engines, translation services, hotkeys, capsule action count and ordering, the app blocklist, and appearance.

AI related

Prompt settings · Custom actions · Model services

Per-action prompt editing

The prompts behind Translate, Explain and Summarize are all editable — what the editor shows is exactly what gets sent, with view-and-restore for system defaults.

Frequent commands become actions

Set a name, a prompt and a sample, confirm with a try-run, and save. Custom actions sort and display alongside the built-ins, up to 10, each with a pickable icon.

Multiple model services

Endpoint, model name and key are yours to set and change anytime. The default service backs all AI actions; keys stay on your machine.

新建 AI 动作自定义动作 3 / 10
名称(显示在浮动条上,建议不超过 6 字)语气改写
提示词(发给模型的指令,所选文字随后附带)把下面这段文字改写成更克制的书面语气: 保留事实与数字,删去夸张副词,短句优先。只输出改写结果。
样例(作为所选文字发给模型,不保存)这个方案真的超级无敌爆炸好用!!!
试跑结果这个方案很好用。

Search & display

Search engines · Translation services · Hotkeys · Floating bar · Apps & appearance
搜索引擎{q} 为选中文本占位符
  • Google默认
  • 百度
  • GitHub
  • 术语库

内置引擎支持停用与排序,自定义引擎可增删;调整即时生效。

Search engines

Six built-in engines with toggle and ordering. Add custom engines — the query is marked by the {q} placeholder.

Translation services

AI translation, Baidu Translate and DeepL support toggles and ordering, with a default service. Baidu and DeepL need your own keys.

Hotkeys

The four OCR features each take a hotkey — must include ⌘ / Ctrl / Alt / Shift, with one-click reset to defaults. The menu bar shows current keys.

Floating bar

Shows 4 actions by default, configurable to 2–8 with the rest in an expand panel. Actions support toggling and ordering, applied instantly.

Apps & appearance

Specific apps can join a blocklist so the bar never pops there. Light, dark and follow-system themes; the Dock icon can be hidden.

Positioning

Capture, process, hand off, keep — four stages in one continuous flow

Magpie covers capture, processing, hand-off and keeping: direct-to-vault Obsidian writes are live — excerpts, tasks and cards land in your vault from templates — while Notion, flomo and more integrations are on the way, so results keep flowing into the apps and note systems you already use.

AI actions

Live
  • Translate (three services)
  • Explain
  • Summarize
  • Up to 10 custom actions
  • Per-action prompt editing

Local actions

Live
  • Copy / Search
  • Open link / Compose email
  • URL, email, code & phone extraction
  • Standalone result window — pinnable and draggable

Obsidian integration

Live
  • Save to note: excerpts append to the daily note
  • Save to task: due date detected from the selection
  • Save to card: one file per card, titled from the selection
  • All three templates editable — no plugin required

App integrations

In development
  • Send to Notion / flomo / Apple Notes
  • Similar hand-offs: vocabulary, read-later

Flow: select → explain → copy → paste into notes.Download Magpie

Privacy & Price

Permissions, data flow and price

What Magpie reads, which system permissions it uses, where data is stored and sent, and where free ends and bring-your-own begins.

Bring your own model service

Works with mainstream providers — endpoint, model name and key are yours to set and change anytime. Keys stay in a local config file.

Processed on your machine

OCR, info extraction, search and copying run locally. AI actions send the selected text to the model service you configured.

No usage data collected

No analytics, no ads, nothing shared with third parties. Update checks only read public GitHub Releases info.

Free forever

Built-in actions, custom actions, OCR and settings are all free — no subscription, no account. Translate, explain and summarize need your own model service; costs are whatever that service charges.

Three system permissions in total: Accessibility and Input Monitoring for text capture, Screen Recording for OCR region selection. Each permission's purpose and status is listed in the in-app Permissions page.

Download

Download & install

Free forever, no account. Translate, explain and summarize need your own model service connected.

macOS

Magpie for macOS

Universal .dmg · runs on Apple Silicon and Intel

Download from GitHub ReleasesProject home & changelog

Version numbers and release notes on GitHub Releases.

A Windows build is in development.

Installation steps

  1. 01

    Download

    Get the universal .dmg from GitHub Releases, open it and drag Magpie into Applications.

  2. 02

    First launch (unsigned build)

    A double-click may be blocked by Gatekeeper — any one of these works: ① Right-click the app → choose Open ② System Settings → Privacy & Security → Open Anyway ③ Or run the command below in Terminal

    xattr -cr /Applications/Magpie.app
  3. 03

    Re-sign the app (required, or permissions won't stick)

    The build is ad-hoc signed — granting permissions directly shows as granted in System Settings but never takes effect in the app. Run in Terminal:

    codesign --force --deep --sign - /Applications/Magpie.app

The build is ad-hoc signed: after every update or reinstall, run the re-sign command and re-confirm permissions once — they then stay valid.

Grant permissions (final step, one by one)

Quit the app first, then for each item: System Settings → Privacy & Security → the matching pane → "+" to add /Applications/Magpie.app → toggle on. If you granted before re-signing, remove the old entries first. Each card's Open-pane link jumps straight to the matching settings pane (macOS only).

Accessibility

Read the selected text for translate, explain and summarize

Open pane

Input Monitoring

Watch for selection gestures so the floating bar pops up

Open pane

Screen Recording

Capture and recognize screen regions (OCR & pinned images)

Open pane

FAQ

FAQ

For anything else, open an issue on the GitHub repository.

01

Which systems are supported?

macOS, universal for Apple Silicon and Intel. A Windows build is in development.

02

Do I need an API key?

No. Copy, search, open link, compose email, info extraction, OCR and pinning work with zero configuration. Only translate, explain and summarize need a model service.

03

Is my text uploaded?

Recognition and extraction run on your machine; screen content is never uploaded. AI actions send the selected text to the model service you configured yourself.

04

The app won't open after download?

This build isn't signed and notarized by Apple yet: right-click the app and choose Open to confirm the first launch. After updates or a reinstall, permissions need re-confirming.

05

No floating bar on selection?

Check Settings → Permissions for Accessibility and Input Monitoring, and make sure the app isn't in the blocklist. Apps that draw their own UI (WeChat, Office…) may not yield text — use OCR capture instead.

06

How is this different from other selection tools?

PopClip is known for its extension ecosystem; Easydict and Pot focus on selection translation. Magpie differs in three ways: one, beyond selectable text, it reads text from images, videos and scanned documents. Two: results aren't locked in a chat — they stay in a standalone window you can copy and search, write straight to Obsidian, or turn into calendar events and map searches. Three: bring your own model service — no vendor lock-in.

View release notes →