An AI Flashcard Generator That Already Has Your Material

When you generate, the request carries the course's own material rather than whatever you happened to paste. Three sources go in.

The most recent syllabus parse for that course: up to 60 extracted items, each as its title and type, trimmed to 8,000 characters. That is what gives a deck the shape of the course — its topics, its units, what the instructor said it would cover.

Up to the 10 most recently uploaded note files for that course, sharing a combined 24,000-character budget of extracted text, each labelled with its filename. Notes are read newest first, on the reasonable assumption that recent material is what you are studying.

And a list built from the course's other tracked tasks — up to 60 of them, ordered by date — so the generator knows what the term contains around the thing you are studying.

• Syllabus: up to 60 items, 8,000 characters
• Notes: 10 newest files, sharing 24,000 characters of extracted text
• Tracked tasks: up to 60, ordered by date

The generate panel asks two things and then confirms. The first is what to focus on: the whole course, or one specific item you are already tracking as a deadline.

That second option is the one that earns its place. A deck generated for the whole course is diluted by definition — a midterm review that includes material from finals is a worse deck for the midterm, no matter how good the individual cards are. Pointing it at a tracked exam narrows the source material to what that exam is about.

It works because the exams are already in the app as real dated items with types, extracted from the syllabus. A generator with no knowledge of your term cannot offer this, because it has no idea what your midterm covers or when it is.

The second question in the panel is optional study material: attach a PDF or a photo and it becomes part of what the deck is generated from, alongside the syllabus and notes.

This is the highest-signal input available, and it is worth using when it exists. A review sheet is the instructor telling you what the exam covers, in their own words, with their own emphasis.

The file goes to a private storage bucket, filed under your own user ID. Nothing is parsed on your phone: the first time a generation or a tutor request needs that file, the server reads it, extracts the readable text while preserving structure, and caches the result.

The shape of the output is fixed rather than left to the model. The front is a short question or a term. The back is a concise answer or definition, one to three sentences.

The target is between 10 and 20 cards, with an explicit instruction that fewer good cards beat padding — if the material genuinely does not support twenty, it should not invent them.

Then everything is validated before it is saved. Both faces are trimmed and each is capped at 300 characters. Any entry missing a front or a back is dropped rather than failing the whole batch, so if fourteen of sixteen came back clean you get the fourteen. No more than 30 cards are inserted from a single run.

The 300-character cap is a defensive clamp on model output specifically. Cards you type yourself are not truncated.

Every card carries four values: an ease factor, an interval in days, a due date, and a count of consecutive successes. A brand-new card starts at ease 2.5, interval 0, due now — so anything you generate is due in your very next session, with no waiting.

Grading a revealed card runs a compact SM-2 variant. Again drops ease by 0.20, resets the success count, and brings the card back in about ten minutes rather than tomorrow. Hard drops ease by 0.15; a new card graduates to one day, an established one gets its interval multiplied by 1.2. Good is the standard ladder: one day, then six, then multiplied by the card's own ease. Easy raises ease by 0.15 and multiplies by that ease plus a further 1.3 bonus, so a new card graded Easy jumps straight to six days.

Ease is floored at 1.3 — the SM-2 floor — both in the scheduler and by a database constraint, so a bad week of Again and Hard cannot drive a card into a permanent loop.

In practice a card you keep getting right goes one day, then six, then fifteen, then about thirty-eight, then roughly three months. A card you keep missing stays in front of you.

• New card: ease 2.5, due immediately
• Again ≈ 10 minutes; Good ladder 1 day → 6 days → × ease
• Easy multiplies by ease and a further 1.3
• Ease never falls below 1.3

Starting a session snapshots the due queue at that moment. Grading a card pushes its due date forward, and without the snapshot the list would reshuffle underneath you mid-session. A counter shows your position, like 4 / 12, with an exit beside it.

Generation is entirely optional. New Deck creates an empty deck with a title of up to 80 characters, scoped to the course you came from or left uncategorized if you opened Flashcards on its own. Add Card gives you two multi-line fields, front and back, both required — a card with a blank side is not saved.

Hand-written and generated cards live in the same deck and are scheduled by the same algorithm. There is no second-class citizen here.

Flashcards are part of Pro, at $4.99 a month or $29.99 a year — about $2.50 a month annually — bought with a card on the web or inside the iOS app, and applied account-wide across iPhone, iPad and the browser.

The honest limitation is the source material. This generator is strong when a course has a scanned syllabus and uploaded notes, and weak when it has neither, because there is nothing to build from. If you want to turn an arbitrary PDF or a YouTube lecture into a deck with no course attached, a general-purpose generator is a better fit and you should use one.

It is also not a shared deck library. There is no browsing other students' decks; everything here is generated from your own course's material or written by you.