The tools for thinking with a whole library

People, works, places and concepts

Scholaris reads the whole library looking for entities: people, works, places, organisations, concepts, events and dates (and it tells fictional characters apart). A model proposes them in batches of about 48,000 characters, and then code finds each mention in the text, so every occurrence has its exact anchor and can be opened at its page. Each entity is linked, where possible, to its Wikidata record.

Two entities are joined when they appear close together: the edge weight falls with distance inside a passage and is lower between neighbouring passages. For the strongest relations in each document, a model names the relation. From the graph you can see an entity's neighbours, its timeline and the shortest path between two. Measured in the benchmark, extracting the entities of 300 pages costs about $0.03.

The concept map

A view of the library as a chart of the sky: the vectors of passages and figures are clustered, projected to two dimensions with UMAP (on a sample of up to 8,000 points) and each cluster gets a name. It shows what your library talks about and what lies far from what.

Separately there is Concepts: you type a concept and Scholaris gathers where it is defined, applied, criticised and merely mentioned, each with its passage and page. It exports to CSV, JSON, JSONL, HTML, BibTeX, TEI and XLSX.

The citation graph

How the documents in your library cite each other. The bibliography entries of each document are resolved against your library by DOI, by title (with a compatible year) or by first author and year when there is a single candidate, and every link carries its confidence. What your books cite and you do not have shows up as orphan references: a list of what you still have to read. The graph only looks inside your library; it does not query external databases.

Watchers

A watcher is a saved search that repeats itself: by hand, on every upload, daily or weekly (in the cloud, every day at 06:17 UTC, and the weekly ones on Mondays). It tells you when a new document enters the top results or when the answer changes: if confidence rises or falls, or if it says something else.

Notebooks

Markdown notes with cards: a passage, a page, a figure, a stretch of audio, a note of yours or a synthesis. Each card keeps a verified copy of the passage, its anchor and a hash, and with verify again Scholaris checks that the library still says the same thing: living citations. A notebook's synthesis may only cite the notebook's own cards; any other mark is removed.

Perspectives

Documents you have not opened in a while, gaps (topics you have little on) and recommendations from within what you already have.