AI Disclosure

Reading your files: sorting each file (syllabus, slides, study guide, past quiz) and pulling out the topics, dates and how the grade is split.

Exam Prediction: ranking every topic by how likely it is to be on your exam, with a reason and how sure we are.

Test Plan: summing up public web pages about how your course's exams usually work, like course pages, syllabi and the professor's own faculty page. Review sites aren't used. Each line shows its source and the source's year.

Practice: writing questions and flashcards from your files, checking written answers, explaining mistakes, answering questions about a topic, and the one-page review sheet.

Where AI isn't used: your study plan, sprint timing, Hours Left and Grade Map math are ordinary calculations. Our own prediction formula is tuned with plain math from pooled course facts (see Course insights on the Privacy page), not by training an AI model.

Anthropic runs the AI, Claude (today Claude Haiku for quick steps and Claude Opus for rankings, practice exams and the Test Plan).

What we send Anthropic: the text of your files; images and scanned PDFs as the images themselves; class details (course, professor, exam names and dates, topic names); and what you type into the app, like questions about a topic and notes you ask about. Never your email.

Tavily runs the web search for the Test Plan. We send it searches built from your school, course code, course name and professor name, and the addresses of public pages to read. Never your files, notes, name or email.

Both are listed with everything they receive on the Privacy page.

We don't train AI on your files. Anthropic's business terms say it won't train its AI on what we send unless we allow it, and we haven't.

That covers your notes and answers too. We don't sell or give any of it to anyone who would train AI on it.

Tavily: its terms are different. They let Tavily and the AI companies it works with keep search queries and use them to improve and train AI models, and it may pass queries to other search engines. That's why we only ever send it those four public course details. We've asked nothing of Tavily yet; if it offers a way to opt out, we'll take it and update this page.

Anthropic says it deletes what we send, and what it writes back, within 30 days. Content its safety systems flag can be kept up to 2 years, and some data longer where the law requires.

Tavily keeps queries as long as it needs them to run and improve its service, with no fixed period.

On our side, we save AI results so we don't pay for the same work twice. They're deleted 90 days after they're made, and with your account. What the AI wrote for you (rankings, questions, cards) stays in your class until you delete it.

A ranking is a prediction from your files, your course's history and what classmates reported. Early in a course, with little data, it can miss. It isn't a statement of what's on your exam.

Practice questions and answers can contain mistakes. Check anything important against your course materials.

The Test Plan depends on what's public online and can be out of date. Each line shows where it came from, and anyone can report a problem with it.

It's built to decline homework, essays and answers to graded questions, and to explain the idea instead. That can still slip on an unusual question.

Files you upload are handled as material to read, not as instructions: the app is built to ignore instructions hidden in a file, though no safeguard is perfect.

Follow your course's rules about AI tools. If your syllabus says no AI, don't use it for that course. The Academic Integrity page has the full rules.

Don't hand in what the AI writes as your own work unless your instructor allows it.

Tell us when it gets something wrong: use "Report a problem" on a Test Plan line, flag a question, or use the Contact page.