How Novus Learn turns a source into a study space.

The core idea: extract, never generate

Most learning tools generate new text about a source. Novus Learn does the opposite. It reads a fixed source and builds everything — claims, summaries, notes, flashcards, quizzes, a concept map, and visuals — from that source's own sentences. There is no language model writing answers, so there is nothing to hallucinate and nothing to invent.

That single decision shapes the whole product. Because every claim is a real sentence from the source, you can trust the breakdown exactly as much as you trust the source — and you can check any claim in one click.

It starts with a fixed source

Whether you search a Wikipedia topic, paste an article URL, upload a file, or open a journal article, Novus Learn pins a specific version of that source — a Wikipedia revision, a retrieved page, your uploaded text, or a paper's full text. Everything downstream is built from that fixed snapshot, so a breakdown is reproducible and always points back to exactly what it read.

Claims are quoted, and every one links to its evidence

Novus Learn pulls out the load-bearing statements in a source as claims. Each claim quotes the source and stays attached to the block it came from, so opening a claim shows you the exact evidence text behind it. Nothing is paraphrased into something the source did not say.

Direct, derived, and Wikidata links are labelled — never blurred

Connections between concepts come in kinds, and Novus Learn keeps them honest. A direct relationship is stated in the source; a derived one is inferred from structure and marked as derived; a Wikidata link comes from structured open data. You always see which is which, so a connection is never dressed up as more certain than it is.

When the source can't answer, it says so

Real sources have gaps. A sparse article is marked limited-source instead of being padded with guesses; a title with several meanings is marked as needing a choice rather than having one picked for you. The Ask feature answers only from the source's own sentences, and tells you when the source cannot answer instead of making something up.

From claims to a whole study space

Once the claims exist, the study tools follow — all built only from supported claims. The summary and study notes reorganise the material; flashcards and the quiz turn claims into recall practice; the guided lesson walks them in order; the concept map shows how ideas connect; and the Visual Studio draws diagrams from the same evidence. Each tool links back to the source it came from.

The sources — all open, keyless, and attributed

• Wikipedia for article text, pinned to a specific revision.
• Wikidata for structured identity when an article links an entity.
• Wikimedia Commons for reusable media and licence metadata.
• Wiktionary for the free, source-cited dictionary.
• Any public web page you point it at, extracted with a readability pass.
• Open-access journal articles (Europe PMC), abstract or full text.
• Your own text and transcript files, parsed locally in the browser.

Local-first and accountless by design

Public pages open without a login. Saved projects, preferences, and history live on your device in local storage and IndexedDB — there is no account and no cloud profile. Content pages load no third-party trackers; optional analytics and ads run only if you allow them in the cookie banner.

How to check our work

Every claim opens its evidence. Every project can download a fully-referenced source list. And every source keeps a link back to the original — the exact Wikipedia revision, the page you read, or the paper — so you are never more than one click from verifying what you studied.