How it works, in the open.
LearnReally is the memory layer for everything you learn with AI: cards get written for you — by your own agent or by the Deck Builder — and the app makes you recall them right before you'd forget, graded honestly against the reference. These pages document the whole machine: the loop, the methodology behind it, and every limit the server enforces.
Five steps, closed.
1 — Learn. You study the way you already do: a conversation with an agent, a course, a textbook, a topic you type into the Deck Builder.
2 — Cards get authored for you. Your agent (through the open-source MCP) or the Builder writes proper spaced-repetition cards: one idea per card, phrased for active recall, with a reference answer the grader can judge against. You never hand-write a card.
3 — They sync. Cards land on your phone and the web dashboard automatically.
4 — You recall. When a card is due, you produce the answer from memory — typed or spoken — into an empty field. No multiple choice, no word bank; recognition is the weakest form of memory and we don't test it.
5 — You're graded, and the schedule adapts. The AI marks your answer against the stored reference — never against its own knowledge — and FSRS schedules the next review from how you actually did. Then it tells you to stop: when nothing is due, the honest amount of studying is none.
Read in order, or jump.
Your first deck three ways, and your first review.
How review worksTyped recall, honest grading, FSRS spacing — the methodology, in the open.
Deck BuilderFrom a one-line brief to a full course-sized deck.
Connect your agentThe open-source MCP: setup for every client.
Limits & pricingEvery bound the server enforces, published — the same numbers your Account shows.
FAQThe parts you’ll want to check.