Frequently asked questions

Upload a text — a novel, a history, a scripture, a course reader — and Constellate charts its cast: every figure as a node, every relationship as a line, each grounded in a short quotation from the text. The result is an interactive constellation you can search, filter, and share.

PDFs with real, selectable text. The free plan accepts up to 30 pages per document; Pro up to 400; Scholar up to 600. Scanned or photographed pages (images of text) can't be read yet — if you can't select the words in your PDF viewer, Constellate can't either.

A parse is one full read of an uploaded document. Free includes 3 parses a month, Pro 15, Scholar 30. A failed parse never counts against your allowance.

The free shelf holds one graph at a time — uploading a new text replaces it (only after the new graph builds successfully, so a failed upload never costs you the one you had). Pro keeps a library of 20 graphs, Scholar 50.

Each node is a figure. Size tracks how often they're mentioned — protagonists grow large, walk-ons stay small. Color marks the figure's group (a family, a faction, a court…), and the Legend button lists every group and line style in your graph. Only prominent figures show a name label by default, to keep dense casts readable.

Solid lines are kinship (parent, child, sibling…), dashed lines are unions and allegiances (spouse, lover, ally, mentor…), and dotted lines are social bonds (friend, rival, enemy, colleague…). Arrowheads mark directed ties like parent → child. The Legend shows the exact set found in your text.

The info panel opens with their epithet, a summary, their aliases, their relationships (each one clickable — the panel navigates to that person), and a short evidence quote from your text. The ✕ closes the panel.

Type any part of a name in Find a figure… and the map glides to the best match while its story opens in the panel. Accents don't matter — “ursula” finds Úrsula — and nicknames work too, because search also looks through each figure's aliases.

That's the spotlight: the found figure and everyone directly connected to them stay at full strength (with all their names shown) while the rest of the cast dims for context. Faded figures are still clickable. Clear the search box — or click any empty space — to bring the whole constellation back.

The ⌂ pill in the corner names the text's central figure; clicking it (or any white space in the graph) resets the view — search cleared, spotlight off, centered back on the protagonist.

Yes — the slider in the lower-left adjusts node spacing live; the layout gently re-settles as you drag. Large casts open with extra room automatically. You can also zoom (scroll), pan (drag the background), and pull individual nodes around.

Filters narrow the constellation: pick groups, pick connection types, or hide secondary figures, and everything else fades. Match: Any shows a figure if it has any selected connection; Match: All keeps only figures having every selected type — handy for questions like “who is both a parent and a spouse here?” Active filters appear as removable chips above the graph.

Your document's text is split into large overlapping sections, and an AI reader works through them in order, recording every named figure it meets: names, nicknames and titles, a one-line epithet, a running mention count, and every relationship the text states or clearly implies. The sections are then merged into one cast.

Each figure's aliases are collected as the reader goes (“Sarai” and “Sarah”, “the colonel” and “Colonel Aureliano Buendía”), and the merge step folds records that refer to the same person into one node. The panel's “Also called” row shows what was folded together. If two truly different people share a name, their distinct relationships usually keep them apart.

Groups aren't a fixed menu — they're whatever organizes your text. A family saga yields families; a war history yields armies and parties; a scripture yields eras or houses. The reader assigns each figure the grouping the text itself suggests, and colors are handed out by group size, biggest first. That's why every graph's legend is different.

Prominence in the text. Figures who recur and drive events are marked primary and get always-on name labels; passing mentions are kept as secondary — present in the constellation and the counts, but unlabeled until you select them, spotlight their neighborhood, or (on Pro/Scholar) un-hide them in Filters.

Each connection is typed from a fixed vocabulary — parent, child, spouse, sibling, lover, relative, friend, ally, mentor, rival, enemy, servant, colleague, and other — and stored with a short description and, where available, an evidence quote. Only relationships the text states or strongly implies are drawn; Constellate does not invent connections to fill gaps.

By design. Evidence quotes are capped at 50 words, at most two per figure — enough to ground a claim, never enough to reconstruct the work. Your reading happens in the book; the map just points you to the right page.

The reader makes judgment calls — where a group boundary sits, whether a passing name deserves a node — and those calls can vary a little between runs, like two careful readers disagreeing at the margins. The core cast and relationships stay stable.

A second, slower pass that lingers on the text — surfacing more secondary figures, more relationship evidence, and steadier grouping on very large casts. Scholar accounts choose it at upload.

Only the derived map: figures, relationships, summaries, and the short evidence quotes. Your uploaded document is never reachable through a share link — not by page, not by download. Free-plan share pages carry a small Constellate watermark; paid plans share clean.

From Account — upgrades go through Stripe checkout, and Manage billing opens the portal where you can switch plans, update cards, or cancel anytime. Your graphs stay yours when a plan lapses; you just fall back to free-tier limits going forward.

Deleting a graph removes it and its derived data. Deleting your account (also under Account) removes your documents, graphs, and profile — everything.

The usual cause is a PDF without extractable text — scans and photographed pages. Try a version where the words are selectable. Failed parses never count against your monthly allowance, and the failed card can be dismissed with its ✕.

Judgment calls at the margins do happen, especially with very common names. A re-parse often resolves it; Scholar's deep parse is more thorough. Graph editing tools for hand-corrections are on the roadmap.

Write to support@constellate-atlas.app — include your account email and the graph title, and we'll dig in.