StudyNotch

How it works

It is one app. No server, no account.

Structure

On this Mac The AI you already use Notch screens cards, answers, time left Sessions and timer course, length, time left Distraction checks front app, window title, tab address Records sessions, questions, review cards Selection and screenshots puts a button on the right-click menu Course material finds the sentence’s page in your PDFs AI connection runs the tool or calls the API Settings and Keychain config.json, API key answer Claude app with Claude Code inside Codex ChatGPT subscription Claude API API key Any one of the three is enough question + page answer
Only one line leaves the Mac. Nothing goes out from the distraction checks.

The path of one question

Mac app Your AI ① Select text and right-click press “Ask” above the menu ② Pick a button, press “Send” ③ Find the sentence in your PDFs which page, and that page’s text ④ course notes + found page + selected text + question ⑤ the answer streams back ⑥ Show the answer and save it math is drawn, not left as code
Step ③ is what makes this app different. The matching page of that course’s material goes along with the question, so the answer follows your professor’s notation and where you are in the course. The course’s PDFs are also the AI’s project: it can open the rest of them when that one page isn’t enough.

Finding the page

The first time it sees a PDF, it extracts the text of every page and keeps it (next to a copied PDF in materials/_text/, and for a PDF in a linked folder in ~/StudyNotch/.cache/text/, so the linked folder is never written to). It removes all spaces and punctuation from both the selected sentence and each page’s text and lowercases them, then looks for the page that contains the beginning, middle and end of the sentence. Words broken by line breaks or hyphens still match.

Files with no text layer, such as scanned PDFs, can’t be searched. Ask with a capture instead and it reads the picture itself.

macOS doesn’t let an app add items to another app’s right-click menu; the only allowed route is inside Services at the bottom. So instead of putting the button in the menu, it attaches the button right above it.

  1. It notices that a right-click happened. It looks only at where you clicked, not at what you clicked.
  2. It carries on only if the front app has selected text. For apps that don’t say, it judges by whether you just dragged.
  3. It reads the position and size of the menu window that just opened and shows the button above it at the same width. It never reads the window’s contents.
  4. When the menu closes, the button disappears.

How selected text is read

  1. Nothing is read before you press the button.
  2. When you press it, it asks the front app for its current selection, using the Accessibility permission.
  3. In apps that don’t answer, it copies once and immediately restores your clipboard.

Judging drift

During a session it checks the frontmost app every 2 seconds, and for a browser, the address of the tab.

  1. Allowed if you pressed “It’s for study” for it in this session.
  2. Allowed if it is on that course’s allow list or the shared allow list.
  3. A distraction if it is on the distraction list.
  4. Otherwise it isn’t counted.

In the lists, an entry with a dot is a site: youtube.com matches m.youtube.com but not notyoutube.com. An entry without a dot is an app name.

Where your data lives

DataWhereLeaves your Mac
Apps and sites you visitedOnly in memory, then goneNever
Session history~/StudyNotch/sessions.jsonlNever
Course PDFs and extracted textCourses/<course>/materials/Only the one page that matches a question
Questions and answersCourses/<course>/log/To the AI, when you ask
Captures~/StudyNotch/Captures/, deleted after a weekTo the AI, when you ask
API keyMac KeychainOnly to Anthropic

To draw math in answers, the KaTeX and marked libraries are loaded from a public CDN, which leaves an access record there. The next version will bundle them in the app and remove that connection.

Permissions

PermissionReadsNever reads
AccessibilitySelected text, the frontmost window’s titleKeystrokes, screen contents
Screen RecordingThe area you capture with “Capture and ask”The rest of the screen
AutomationThe address of the browser’s front tabPage contents, other tabs