How this works
Google indexes web pages. Wikipedia organises knowledge by topic. Curio organises it by the questions people keep asking — because the question is usually the part you actually have, and the page is the part you have to go and find.
Cards are not written by hand and they are not copied from anywhere. A pipeline watches public discussion for questions that keep resurfacing, merges the different phrasings of the same curiosity into one, gathers evidence from several independent sources, and writes an explanation that starts from plain intuition and goes down five levels to expert detail. A second pass checks the draft against the same evidence and scores how well supported it is.
That score is shown on every card, and it is not decorative. Below a threshold, a card is never published at all. Where sources genuinely disagree, the card says so rather than picking a side quietly.
Most of what gets collected never becomes anything. You can see the discards for yourself on the incoming queue — every question the platform has found, where it came from, and whether it was rejected and why.
No card may use a term it has not already explained. Every explanation is checked for it, and a violation costs the card confidence. It is the difference between an answer you can read and an answer you have to go and research first.
Browsing, search, graph, saves, offline
Always available
Search by meaning as well as words
Embeddings active
Fresh explanations when one doesn't land
Generated on demand
Discovering and writing new cards
Model: gemini-3.5-flash
There is no account and no password. Your browser generates a random key and keeps it locally; the server uses it to remember which cards you have read so it can suggest what follows. Nothing is linked to an email, and you can see and copy the key on your page.
A fixed share of every recommendation is drawn from subjects you have never touched, and it cannot be turned off. A system that only ever shows you more of what you already clicked stops being worth opening.