Muse — AI Companion

The Pen Stays Yours

Muse never writes your prose — no drafting, rewriting, or paste-ready lines. It inspires, grounds its suggestions in real craft and your own research, and asks the question that helps you find your words. When it does propose a change to your story world, nothing is written until you approve it in the Suggestions window.

What Muse Can Propose

• New entities — a character, location, faction, item, or glossary term, drawn from your chapters and research, as a full card you approve
• Enrichments — fill a thin card's profile from what your material actually says, adding only, never overwriting what you wrote
• Character connections — wire two existing characters together with the right relationship, both-way labels, and direction
• Everything shows up as a preview card in Suggestions, with its source passages attached — approve, dismiss, or Add all

Deep Scan — Find What's Missing

Point Muse at your manuscript and research and let it mine them: pick a focus — characters, locations, factions, items, connections, terms & lore — or scan for everythingat once. It reads versioned, passage-sized evidence in resumable batches and proposes what it finds, so a book you've been writing for months can have its cast, places, and links caught up in one pass.

• Find missing — surface people, places, and things named in the prose but not yet in your world
• Enrich existing — deepen thin cards from the material, section by section
• Connections — propose relationships between characters who share the page
• Everything runs to Suggestions for your review; nothing lands without approval
• A Deep Scan needs a book scope and snapshots that scope when it starts — only that book's chapters, its attached research, and research shared at its universe level — so switching the scope picker afterward cannot redirect a scan in progress
• Its second pass matches each discovery against your existing story world on your own device, skipping ineligible names without spending AI Tokens; Find missing suggests only new entities, Enrich existing only additions to cards you already have
• Every candidate must match an exact quote in the passage it cites; if a source changes, the scan rewinds that source and a stale suggestion cannot be applied
• Partial extraction and unsupported files are listed as coverage warnings, never silently counted as fully read

Grounded, Cited, and Remembered

• When you name someone or something, Muse goes to the chapters that actually mention them, follows a glossary term through to its full backing — the entity it stands for and any research attached — and treats nicknames and spelling variants as the same subject
• Every fact it takes from your work carries an inline link on the meaningful words themselves — chapter titles open the editor, quoted or named manuscript evidence opens its exact paragraph, and research terms open the source passage
• Citation links are checked against what Muse actually read in that turn; a mismatched or invented destination is quietly disabled without adding repeated warning text to the answer
• Before answering, pending local index changes are flushed; incomplete or unsupported sources anywhere in the question’s scope are disclosed even when they returned no passage
• For a thread you are tracking, Muse starts with your Marginalia and the Plot Grid — what you meant to track — then follows into the chapters, characters, and research to see it in context; if nothing is tracked, it points you to Manuscript Lens or Marginalia
• It reads your saved AI reports too — Chapter Analysis, Chapter Pulse, and Manuscript Lens — and connects their findings; name the chapter you mean, since if you say only “last” it asks which rather than guessing the newest report
• The reasoning control defaults to Auto: light for simple chat and deeper after substantive reading; only the current action and thought are shown while Muse works, then they disappear from the completed reply
• It remembers your preferences and working agreements per universe, across chats, selecting the relevant ones without deleting older intentions from your device
• For a private note or idea of your own, it can search your mind-map cards, chapter notes, and snippets when you mention them

Continuity — Flagged and Checked

Continuity checking lives inside Muse now. As it reads, it flags contradictions it notices in passing — prose that disagrees with a card, or a detail that changes between chapters — with both sides linked, and offers to help you resolve them. For a focused question (“is Tam's age consistent?”) it simply checks with its ordinary reads.

• Notices and flags clashes while it answers — a changed eye colour, a broken timeline, a rule quietly bent — never silently smoothing them over
• Runs a systematic continuity check on request — sweep the characters, or the whole book, for contradictions against your world and against itself, returned as conflicts, watch-outs, and gaps with paragraph links
• That systematic sweep is a metered run that uses AI Tokens, so Muse names the cost when it offers one and keeps the default scope light unless you ask for the whole book
• Diagnostic only: it surfaces the issue and helps you plan the fix — the prose stays yours