Turn one page of notes into a 20-card deck in about 30 seconds.
Paste your notes → a deck you can study, in about 30 seconds.
Make My DeckAbout 30 seconds · No signup · Works offline in your browser
- Forge four card styles from one source — recall, cloze, concept or definition.
- Review with a real 3D flip and a Leitner-style queue that reorders as you go.
- Spot leeches — cards you've missed twice — and drill just those in one click.
- Export to CSV for Anki or Markdown for your notes.
Drop in your material
Mode: not generated yet
Where does my text go? If a model is configured on this deployment, your material is sent to it to generate cards. If not, everything happens locally in your browser and nothing is uploaded. The mode pill above always shows which one just happened.
Check what you get. Generated cards can be confidently wrong — skim each one against your source before you trust it in review.
Your deck
0 cardsNo deck yet
Paste material on the left, or load a sample, then forge a deck to see it here.
A little more material, please
Flashcards need at least 80 characters of source text to work with — right now there's 0. Paste a fuller paragraph or two, or load a sample.
Couldn't extract enough distinct cards
The material may be too short, too repetitive, or mostly numbers and code for this style. Try a longer passage, add a heading or two, or switch to a different card style.
Generated locally in your browser
No model is configured on this deployment, so cards were built here, on this page. Nothing was uploaded.
Generated by the configured model
Your text was sent to the model to produce these cards.
Deck state lives only in this browser tab's memory — reload the page and it's gone. Export it before you close the tab.
Session complete
Here's how it went.
0 leeches — missed more than once
These cards got an "Again" grade twice or more this session. Repetition alone isn't fixing them — reword them, split them, or drill them on their own.
How this is calculated
How the cards and the queue actually work
- Two card sources, never a silent failure. If this deployment has a model bound (Workers AI or an Anthropic API key), your text is sent to it with a structured prompt and the response is parsed into cards. If no model is configured, or the request fails or times out, a local generator runs the exact same job inside your browser — no server round trip, no upload.
- The local generator scores sentences, then extracts patterns. Each sentence is scored on length, whether it matches a definitional pattern ("X is/means/refers to Y"), whether it contains numerals or capitalised terms, and its position in the document. Strong sentences feed four extractors: definitions, cloze deletions (blanking the highest-information word), heading-to-question pairs, and cause-and-effect statements.
- Cards are deduplicated and spread across the source. The document is divided into as many position buckets as cards requested, and the best candidate from each bucket is chosen — so a long source doesn't end up as twelve cards from the first paragraph. Duplicate or near-duplicate fronts are dropped before the deck is capped to the requested count.
- The review queue is a Leitner-style box scheme, not spaced repetition. Grade a card "Again" and it goes back to the front of the queue. "Hard" reinserts it two positions ahead. "Good" reinserts it five ahead. "Easy" parks it — it won't come back this session. This reorders one sitting; it does not schedule anything across days, which is what real spaced repetition requires.
- Leeches are tracked, not hidden. Any card graded "Again" two or more times in a session is flagged as a leech at the end, with a one-click action to start a new session containing only those cards.
What the honesty claims in this tool rest on
- Ebbinghaus, H. (1885). Über das Gedächtnis — the original forgetting-curve research establishing that unreinforced memories decay predictably over time, the basic case for reviewing material more than once.
- Leitner, S. (1972). So lernt man lernen — the source of the box-based review scheme this tool's queue is named after and loosely modelled on. The original moves cards between boxes reviewed on different calendar days; this tool reorders a single sitting instead, which is a meaningfully smaller claim.
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). "Distributed practice in verbal recall tasks: A review and quantitative synthesis." Psychological Bulletin, 132(3). The meta-analysis behind the general claim that spacing study out over time beats massing it — the reason real spaced repetition needs days, not one sitting.
- Roediger, H. L., & Karpicke, J. D. (2006). "The power of testing memory: Basic research and implications for educational practice." Perspectives on Psychological Science, 1(3). The evidence for the "testing effect" — that retrieving an answer strengthens memory more than re-reading it — which is the argument for using flashcards at all rather than rereading notes.
None of the citations above were measured against this specific tool. They support the general mechanisms — spacing and retrieval practice help — not a claim that this implementation reproduces their effect sizes. The in-session queue is explicitly not spaced repetition; treat it as a smarter shuffle, not a study-scheduling algorithm.
Flashcards, answered directly
How do I turn my notes into flashcards?
Paste your notes into the box, pick how many cards you want and which style — recall, cloze, concept or definition — and generate. If a model is configured on this deployment your text is sent to it; otherwise a local generator in your browser extracts definitions, cloze deletions, headings and cause-and-effect sentences into cards. Either way you get an editable deck in seconds, not minutes.
What makes a good flashcard?
One fact per card — not three facts crammed onto one. A question you genuinely could not answer without knowing the material, not one you could guess from the phrasing. No lists longer than three items, because longer lists turn into recognition rather than recall. And write it in your own words: the textbook's exact sentence tests memorisation of the sentence, not understanding of the idea.
What is spaced repetition?
Spaced repetition is reviewing material at increasing intervals — a day later, then a week, then a month — because retrieving a memory after a delay strengthens it more than reviewing it again immediately. It requires tracking difficulty per card across multiple days or weeks, which is what algorithms like SM-2 do. This tool's review queue reorders cards within one sitting using a simple Leitner-style scheme; it is not spaced repetition, because nothing here spans days.
Are AI-generated flashcards any good?
They are a fast first draft, not a finished deck. A model can misread emphasis, invent a plausible-sounding but wrong detail, or ask a question the source doesn't actually answer — confidently and without any visual sign it happened. A wrong card studied repeatedly teaches the wrong answer more effectively than no card at all. Skim every generated card against your source once before you start reviewing it seriously.
How many flashcards should I make?
Enough to cover the material once, not enough to make review a chore — for most single-session study material that's somewhere between 12 and 25 cards. More than that and cards start overlapping or splitting into trivia; fewer and you're testing recognition, not recall. If a topic won't fit in 30 cards, it's really several topics — split the source and make separate decks.
A few more questions
Is my study material private?
Only as private as this deployment's configuration. When no model is bound, everything happens in your browser and nothing is uploaded — you can check this yourself in your browser's network tab. When a model is configured, your pasted text is sent to it over HTTPS to generate cards, the same as pasting it into any AI chat tool.
Can I get my deck into Anki?
Yes — export as CSV. Each row is front, back, hint, which Anki's "Basic" note type or a simple three-field import maps onto directly. The Markdown export is for reading or pasting into notes, not for importing into spaced-repetition software.
What happens if I close the tab?
The deck is gone. Nothing here is saved to a server or to local storage — it exists only in this tab's memory for the length of your session. Export to CSV or Markdown before you close the tab or navigate away if you want to keep the deck.
What is a "leech" card?
A card you graded "Again" two or more times in one session — the term comes from spaced-repetition software generally, where a card you keep failing is a sign the card itself is broken, not that you need more repetitions of it. This tool surfaces leeches at the end of a session with a one-click "study just these" option.
Can I fix a card the generator got wrong?
Yes. Press E during review, or use the delete button, to edit or remove the current card in place. Changes only exist in this session's memory, so export the deck again afterwards if you want to keep the fix.
Next
One tool in a set of ten.
The Forge exists to demonstrate a strategy: a small, genuinely useful tool answers a question people already type, earns attention no blog post earns, and survives the shift to AI answers because the value is the interaction, not the fact. More of the toolkit and the argument behind it are next door.