This page exists because a limit you discover at 2 a.m. before a midterm is not a limit, it is a surprise. Everything below is either a rule this site already states somewhere, or a fact that can be checked against the live site. Where a number comes from a page, the page is named.
One of those numbers was wrong until this page was written. That is covered near the end, because a page about limits that only lists the flattering ones is a brochure.
The rules, in one table
These are the same rules printed on the home page and the pricing page. They are repeated here in one place so you do not have to assemble them from three scrolls.
| What | Value | Detail |
|---|---|---|
| Max file size | 10 MB | One file at a time. |
| Max length | 500 pages or sections | Per document. A PDF counts pages; Word, TXT and Markdown count 2,000-character sections. |
| Credit per PDF page | 1 | Even if that page yields three cards. |
| Credit per text section | 2,000 characters | Includes a partial final section. |
| Page that yields nothing | 0 credits | Credited back automatically. No email needed. |
| System retry | 0 credits | A retry the system performs is not charged again. |
| Viewing your cards | 0 credits | Always free, including spaced review. |
| Public book decks | 0 credits | Free to read, no signup. |
| Demo | 5 cards | No account, no email. |
| Verified email | 10 credits | Once, when the address is verified. |
| Starter pack | $4.99 / 100 credits | About 100 PDF pages. |
| Standard pack | $9.99 / 250 credits | About 250 PDF pages. |
| Pro pack | $19.99 / 600 credits | About 600 PDF pages. |
| Expiry | Never | Credits do not expire and nothing renews. |
The credit rules are on the pricing page as a receipt, and the file rules are on the FAQ. Both are quoted rather than paraphrased above.
Seven things the tool will not do
Each of these is a deliberate boundary rather than a missing feature queued for later, so it is worth knowing before you plan around it.
- Read a scanned PDF or a photo of a page. There is no OCR anywhere in the pipeline. A PDF needs selectable text; a page that is an image is rejected rather than guessed at.
- Accept pasted text. You upload a file. If you have text in a chat window, save it as .txt first. This is not a missing feature to be added later; the input path is a file.
- Export to Anki. There is no automatic Anki export and no import from Anki either. Review happens inside the app. If deck portability matters more to you than speed of getting a first deck, this trade is not in WakeCards' favour.
- Schedule your reviews. There is no spaced-repetition algorithm deciding when you should see a card again. You rate each card Blank, Fuzzy, Solid or Easy, and you choose when to run the deck again. Anki and Mochi do the scheduling; this does not.
- Connect to Google Docs, Slides or Notion. There is no integration with any of them. The guides on this site describe exporting a file and then uploading it, because that is the only path that exists.
- Guarantee that a generated card is correct. The model writes the questions and the answers. AI can make mistakes. Every card links back to the page it came from so you can check it against your source, and you should.
- Work instantly. Generation is not instant and larger documents take longer. The server keeps processing after you close the tab, so closing the window does not cancel the job.
Why the no-OCR limit matters more than it sounds
Most study material people already own is a scan. Photocopied handouts, photographed textbook pages, a PDF exported from a scanner rather than a word processor — none of those have selectable text, so none of them will work here. If your material is a scan, run it through an OCR tool first and upload the result. That is a real extra step, and it is the step most people hit first.
Why there is no export
No Anki export, and no import from Anki either. The review loop lives inside the app: you open a deck, you get a card, you rate your recall with the notes closed. If your existing workflow depends on deck files moving between tools, this is the wrong tool and the comparison on the Anki alternatives page says so in its own column.
What the app returns when it refuses
The study app is not an open API and does not pretend to be one. These responses were measured against the live site, not copied from a design document.
| Situation | Response | Body |
|---|---|---|
| Missing the site's own request header | 403 | 请从本站页面提交 |
| Not signed in, calling a POST endpoint | 403 | Same header check fails first |
| Study app pages | 200 + noindex | x-robots-tag: noindex on every /study/ response |
| Crawlers on /study/api/ | Disallowed | Blocked in robots.txt for all 15 named AI bots |
The request header check is the first gate on every POST endpoint, which is why an unsigned caller and a signed-out caller get the same 403: the header test runs before any identity is looked at. The refusal message is in Chinese, so a reader who does not read Chinese may not recognise it as a deliberate guard rather than a malfunction.
There is also a hard boundary on indexing. Every response from /study/ carries a noindex header, and /study/api/ is disallowed in robots.txt for all fifteen named AI crawlers. The marketing pages are the surface meant to be read; the app is not, and a crawler that respects robots.txt will never see it.
What is free, and will stay free
Four things never cost a credit, and none of them has a paywall in front of it: opening the app and trying the five demo cards, viewing cards you already have, spaced review, and reading any of the public book decks. Paying is optional, and browsing is not metered to make paying feel necessary.
Five things this site will never offer
Each of these is a thing a reasonable reader might assume exists. Naming them is more useful than leaving them absent.
- A price per card. You pay per page, not per card. A page that produces three cards costs the same as a page that produces one, and a page that produces none costs nothing. There is no per-card metering to optimise against.
- A monthly plan. There are three one-time packs and no subscription. No card is stored on the site, nothing renews, and there is nothing to cancel.
- A discount for buying two packs. Two packs cost exactly the sum of the two. There is no bundle tier to unlock, which also means there is no reason to buy more than you need.
- A guarantee of a refund on request. A pack is not refunded on request. Two things are automatic: pages that produced no cards are credited back, and any refund the payment platform or local law grants takes back the credits that refund bought, so the balance matches the money actually paid.
- A named author for every deck. One book deck, The Eclipse of Russia, is not listed anywhere on this site. Its page shows no named author and points at a source file that does not match the text. It stays out until that is fixed, and it is named here rather than quietly dropped.
The one deck that is deliberately missing
The Eclipse of Russia does not appear in the book deck list or in llms.txt. Its page shows no named author and points at a source file that does not match the text beside it. Publishing a deck whose provenance does not hold up would be worse than publishing ninety-four, so it stays out until the source is corrected. It is named here so the omission is visible rather than silent.
A number we published was wrong, and how it was found
Until this page was written, the machine-readable summary at /llms.txt said there were 92 public-domain book decks holding 19,485 cards. That is what an AI assistant reading this site would have repeated.
Counting the pages themselves on 2026-10-29 gives 95 decks and 19,912 cards. The figure was stale, and it was stale in the worst possible place: the file written specifically to be quoted.
WHY THIS IS ON THE PAGE
A COUNT IN A DOCUMENT GOES STALE
- Stale claimtill 2026-10-29
- Decks then92
- Decks counted95
- Cards then19,485
- Cards counted19,912
- Where it was stale/llms.txt
A count in a document goes stale the moment a deck is added, and a stale count is a small but real inaccuracy in every citation that repeats it. Putting the date next to the count is the only fix that survives the next addition. The number above is the counted one, and the date it was counted is printed with it.
The count is now stated as 95 decks and 19,912 cards, counted page by page, with the counting date attached. If you are reading this long after 2026-10-29, treat the count as a floor and check the book cards index for the current one — the index is generated from the same manifest, so it is the number that actually changes.
How to check any number on this page
File and credit rules — compare with /pricing/, which prints them as a receipt, and /faq/, which states them in prose.
Deck and card counts — open /book-cards/ and count. Each deck page prints its own card count; the totals above are the sum of those printed numbers, not an estimate.
Crawler rules — read /robots.txt. The order inside each block matters: every Disallow line comes before its Allow line, because the Python parser on this server takes the first matching rule rather than the longest one.
Refunds — /terms/ is the binding text. Where this page and the terms disagree, the terms win, and that sentence is the only promise on this page that has a lawyer behind it.
Frequently asked questions
What is the largest file I can upload?
10 MB, and up to 500 pages or sections per document. Those two numbers are the whole size limit; there is no separate limit on card count or on output length.
Why does a page that produced nothing cost nothing?
Because the charge is for pages that turned into cards. A page that yielded no cards is credited back automatically. You do not have to write in, and no support ticket is involved: the refund is part of how the charge is calculated.
If I stop a job halfway, do I pay for the rest?
No. The pages the job had not reached are refunded too. You only ever pay for pages that actually turned into cards.
Does an analysis that says my document is unusable cost me?
A page that produces no cards costs none, whether the reason is a layout the parser could not read or a page with nothing to ask about. The charge follows cards produced, not attempts made.
Can I get a pack refunded if I change my mind?
A pack is a one-time purchase and is not refunded on request. Two exceptions are automatic and need no email: pages that produce no cards are credited back, and a refund granted by the payment platform or by law takes back the credits that refund bought so the balance still matches the money paid.
Do credits expire, and does anything auto-renew?
Credits never expire. Nothing renews on its own. There is no subscription, and no card is stored on this site.
What happens to my documents if I never pay?
Nothing happens to them. Documents you upload stay in your account and stay yours. Studying, viewing and spaced review are free; only turning new pages into cards spends credits.
Why can I not paste text in?
The input path is a file upload, and that is a design decision rather than a gap. If your text is somewhere you can copy from, save it as a .txt file and upload that.
Is the study app indexed by search engines?
No. Every response from /study/ carries a noindex header, and /study/api/ is disallowed in robots.txt for all fifteen named AI crawlers. The marketing pages are the indexable surface; the app is not.
Why is one book deck missing from the list?
The Eclipse of Russia is deliberately excluded. Its page shows no named author and points at a source file that does not match the text it sits beside. It stays out until that is corrected, and it is named here rather than silently omitted.
How many free book decks are there, and how many cards?
95 decks holding 19,912 cards, counted on the pages themselves on 2026-10-29. Note that llms.txt said 92 decks and 19,485 cards until this page was written; that figure had gone stale and has since been corrected there too. The count above is the one you get by counting the pages, which is the only count worth trusting.
Can I trust the numbers on this page to stay right?
Each one is tied to a page you can check. Where a number is a limit set in code, the page that states it is named. Where a number is a count, it was counted rather than recalled, and the date it was counted is printed beside it. The previous version of one of these counts was wrong for at least a day, which is exactly why the date is printed.