Every bound, published.
Two things here cost real money per use — writing cards with AI and grading your recall — so both are bounded. The bounds are sized so a real study routine never touches them; they exist so one account can't run up unbounded model spend. This page renders the same constants the server enforces, and your Account shows your live usage against them.
Free forever, or Pro at $9.99/mo.
Free forever: the open-source MCP (your agent writes cards), curated starter decks, cloud sync, Self-review sessions, and FSRS scheduling. Every new account also gets the two things that cost real money — 200 free AI credits and 120 free built cards — with no card on file and no expiry date. Pro ($9.99/mo, or $64/yr) is the same two dials, monthly and much larger. Reviewing itself is never paywalled; downgrading never deletes anything.
What every account starts with.
One credit is one AI judgement of your own words: a graded Recall answer, a deep-dive follow-up, or a recurring miss distilled into a card of its own. One pool for all three, spent only when you ask for it — self-review never touches it.
One honest course from the Deck Builder. Everything built is yours whatever happens next; the open-source MCP keeps writing cards free after that.
Deck building.
The Builder sizes a deck by what the material needs; this is the structural ceiling.
One bound, not a stack of them: there is no daily building cap. Planning calls draw the same budget. Resets on the 1st; reviewing is unaffected. Need a bigger month? The open-source MCP builds through your own agent, free.
AI credits & deep-dive.
The same three spends — graded answers typed or spoken, deep-dive turns, and misses drafted into cards — from one pool. One bound, not a stack of them: there is no daily cap. About 3× the heaviest real study month; resets on the 1st, and self-review never counts against anything.
Each turn draws the same pool as a grade. The next session is the real test.
Speaking an answer is part of having it checked, not a separate meter: if you have credits, you can dictate. (An anti-abuse ceiling on uploaded audio exists in the background; no real study month comes near it.)
Anti-abuse bounds (not a business model).
Deleting a deck frees its slot.
Agent-written cards through the MCP count here too. A real learner won't reach this.
Adding a deck from the Library.
The Builder's own ceiling, applied to shared decks as well — publishing holds the same number, so nothing can be shared that nobody is allowed to add. Importing is free on every plan.
Publishing to the Library.
Listing a deck in the public Library is what this bounds — how many listings one account can make in a rolling seven-day window; re-submitting an already-listed deck counts too. Sharing a private link is never counted. Nothing you already listed is affected: wait a day or two, or withdraw a deck you no longer want listed. The bound exists so the shelf cannot be flooded in an afternoon.
Why publish the caps at all?
Because a limit you discover by hitting it is a trap, and a “unlimited” that secretly throttles is a lie. Every bound that shapes your studying is on this page and in your Account, with the same numbers, from the same source of truth in the code.