The default engine is not AI
The private deterministic engine is selected first and works with no model provider involved. AI is a choice you make per source, not the path of least resistance.
Study material you can trace back to the sentence it came from.
An AI flashcard generator that has to prove each card
RecallQuill will use a model if you ask it to — and then makes the model's output earn its place. Each returned card names a paragraph and quotes it, RecallQuill re-checks the quote against the stored paragraph, and anything that fails is dropped before you ever see it.
- Verbatim source anchors
- Private engine by default
- Export and delete anytime
Try ai flashcard generator before you sign up
Paste any notes below and the page itself drafts a few anchored cards with the same private engine the product saves with — no account, no upload, nothing leaves this browser tab.
This preview runs entirely in this browser tab, powered by the same private deterministic engine the product uses — no model provider is involved and nothing you paste here is uploaded or stored.
What RecallQuill actually does here
What separates this from a generic AI card maker.
Unverifiable cards never reach the deck
A card whose quote is absent from the paragraph it cites is discarded. If fewer than two cards survive the check, the generation is treated as failed and the deterministic engine completes the deck instead.
Bounded, logged, and attributable
Sources are capped at 30,000 characters, requests carry a 25-second timeout, and a pending record with the source hash is written before the model call begins. Model, tokens, cost, and gateway generation ID are stored with the deck.
Edits are yours, anchors are not
You can rewrite anything a model produced. What you cannot do — deliberately — is detach a card from the paragraph and quote it was built on.
How the workflow works
Using the AI path safely.
Bound the source
Paste or upload the section you want cards for. The character counter shows how much of the per-generation allowance it uses.
Opt in explicitly
Select Luna and tick the consent box naming Vercel AI Gateway and OpenAI GPT-5.6 Luna. Five generations per UTC day are available per account.
Audit the result
Read each cited paragraph next to its card. Edit the wording, or delete the deck, before it becomes part of your review schedule.
Current limits, stated up front
Constraints on the AI path:
- Nothing reaches a model provider unless you pick Luna and tick the consent box for that specific generation.
- Luna generation is limited to five generations per account per UTC day; the private engine has no such limit.
- A source anchor shows where wording came from; it does not prove a card is a good question. Read the anchor and edit the card.
- Sources are UTF-8 text and Markdown you paste or upload — not PDF, OCR, image, video, audio, or office documents.
Questions about AI flashcard generation
Do I have to use AI to make flashcards here?
No. The private deterministic engine produces complete, anchored decks without contacting a model provider, and it is the default choice in the builder.
What does RecallQuill send to the model?
The bounded source you supplied for that generation, split into numbered paragraphs, together with the grounding instructions. The request goes through Vercel AI Gateway with zero data retention requested, and is tagged to your account for rate limiting.
How do I know which engine built a deck?
The deck shows its mode and, for Luna decks, the model name. A deck that fell back to the deterministic engine says so explicitly, and decks generated privately by choice are never labelled as fallbacks.
What if the model returns something wrong but quotable?
The citation check only proves the quote exists in your source. Judging whether the question is worth asking is your job, which is why prompts, answers, and key terms all stay editable.