Glossary

The words we use

Manuscripts.ai uses a small vocabulary consistently. Where a term is doing specific work — different from how a general writing app or a general AI tool might use it — this page explains what we mean. Terms are alphabetical; each entry has a short definition, a concrete example where useful, and a link to the surface where you'll encounter it.

If a word in the UI confuses you and isn't on this page, email support and we'll add it. The glossary is meant to be a fast reference, not an exhaustive dictionary.

Accept flow

The rule that governs every AI-written character in your manuscript. Every AI suggestion appears in a side sheet with an Accept button. Only Accept enters the text into your manuscript, and only Accept logs the change to Provenance.

There is no back door. No auto-apply mode, no "silently improve while you type" background job, no chat-to-editor bridge. If AI text is in your book, you clicked Accept on it. This is the single most load-bearing product rule — if it ever breaks, the whole provenance guarantee unravels. See Write and Provenance.

Adjacent activity

AI work that happens in your project but doesn't touch your manuscript file. Chat threads, Pad sessions, Template runs, Desk drafts, Flags scans, Reports, Deep Analysis runs — none of them write into the manuscript.

Adjacent activity is counted separately in Provenance for transparency. If you spent a Sunday afternoon chatting with the AI about your protagonist, that's twenty messages of adjacent activity — visible in the "adjacent counts" section of Provenance — but zero AI-written characters in your manuscript. The distinction matters when an agent or publisher asks "how much of your book is AI?" — adjacent doesn't count. See Provenance.

Alias

An alternate name for a Lore Book entry. A character's nickname ("Ada" for Adelaide), a location's local name ("the Old Bridge" for the Miller's Crossing), a shorthand for a faction ("the Watch" for the City Constabulary).

Aliases resolve to the same entry when you type @ in the editor. If Ada is registered as an alias on the Adelaide entry, @Ada in your prose links to Adelaide's Lore record — which means when you run an AI command on that paragraph, the AI gets Adelaide's full context (her wound, her occupation, her distinguishing tic) even though the paragraph only used her nickname. See Overview, Lore Book, and Outline.

Atomize / Echoes

The pipeline that takes one source (a chapter from your manuscript, a Vault file, or pasted text) and produces one first draft per selected promotional channel — newsletter, LinkedIn post, Twitter thread, podcast cold open, Instagram caption, back-cover blurb, YouTube description, and ad hooks. Every channel runs in parallel through OpenRouter and comes back as a card in the output pane.

The surface is called Atomize internally (the route is /atomize, the DB table is atomization_runs) but Echoes in the sidebar. Same thing; the label just leaned into "your book, speaking again in other rooms" as a mental model. See Atomize / Echoes.

The floating menu that appears above a text selection in the editor. Formatting buttons on the left (bold, italic, headings, lists); the AI dropdown on the right. Twelve inline AI commands live in the dropdown, plus three "Transform" submenus (Humanize, Translate to, Paraphrase as).

If you select nothing, the bubble menu doesn't appear. If you select and it doesn't appear, you're likely inside a codeblock or a horizontal rule — click into normal prose and try again. See Write.

Chapter break

A visual separator between scenes inside a chapter — renders as a tracked-out * * * ornament in the editor (and in exports). Inserted via the slash menu on an empty line or by typing *** and pressing Enter.

Chapter breaks are visually and semantically distinct from H1 chapter headings. An H1 starts a new chapter; a chapter break marks a scene shift within a chapter (a time jump, a POV switch, a break in the flow). See Write.

Codex auto-injection

When you run an AI command on a paragraph that contains @-mentions, the referenced Lore Book entries' full fields (traits, wounds, aliases, distinguishing details, and any custom fields you added) are automatically included as context in the model's system prompt. That's Codex auto-injection.

The suggestion sheet header shows "+ N from lore" so you know what the AI is reading — if you @-mention three characters in a paragraph and run Rewrite, the sheet says "+ 3 from lore" and each of those three entries' fields go into the prompt. This is how the AI stays in-character rather than reaching for stock detail. See Overview, Lore Book, and Outline.

Credit

A single-use pass to run one Deep Analysis simulation. Priced at $55, purchased separately from your subscription plan, and completely separate from your weekly token budget. One credit runs one simulation of your book across 100 fictional readers over five simulated days.

Credits don't expire and roll over between subscription cycles. If you buy three, they sit in your account until you use them or delete the account. See Deep Analysis credits.

The premium simulation surface. 100 fictional readers read your manuscript across five simulated days on Goodreads, Amazon, Reddit, X, and TikTok. Costs one $55 credit per run, burns roughly 100k tokens across the LLM cascade, and takes 5–15 minutes wall-clock to complete.

The output is a stack of simulated reviews, engagement metrics, and reaction summaries — designed to give you a rough read on how your draft would land with a real reader population before you send it to real readers. See Deep Analysis.

Deduction multiplier

How many tokens come off your weekly budget per token a model actually generates. Different models have different multipliers based on their real upstream cost — a cheap model like Claude 3 Haiku deducts 3, a premium model like GPT-5.4 deducts 75.

The multiplier is why swapping models on the Settings page can dramatically change how fast your weekly budget drains. See Models and AI providers for the full multiplier table.

Flags

The diagnostic pass that reads your book for continuity contradictions, anachronisms, missed setups, and unintentional repetition. Reports each finding with a paragraph reference, a rationale, and a suggested fix.

Flags never edits your manuscript — it's observational only. The typical use is: click Flags on a completed draft, read every finding, click into each to jump to the paragraph, and decide fix / defer / dismiss. Expect 30–90 seconds on a novel. See Flags.

Focus paragraph mode

The iA-Writer-style feature that fades every paragraph except the one your cursor is in. Toggled by ⌘. (Ctrl+.) or the "focus" button in the editor footer.

Use it when a wall of text is distracting you from the specific paragraph you're revising — the fade forces your eye to the one block that's fully lit. Move your cursor and the focus follows. See Write and Keyboard shortcuts.

Lore Book

The typed graph of every named thing in your book — Characters, Locations, Objects, Factions, Events, Concepts, Notes (seven kinds). Each entry has kind-specific recommended fields (a character has traits and wounds; a location has a region and a distinctive detail) plus aliases for entity resolution.

The Lore Book is the source of truth for Codex auto-injection: the fields on each entry are what the AI sees when you @-mention that entry in the editor. Richer entries produce richer AI drafts. See Overview, Lore Book, and Outline.

Manuscript

Externally, what we call a book you're writing in the app. Internally, we sometimes call the same thing a project (particularly in database schemas and API endpoints — projects.id, /api/projects/[id]). They mean the same thing.

A manuscript contains one or more documents, a Lore Book, an Outline, a Vault, chat threads, Pad sessions, a Provenance ledger, and (optionally) a voice fingerprint. Each user account can have multiple manuscripts side by side. See Dashboard.

Narrative Flow

The pacing map. Each chapter gets a tension score derived from an LLM read of what happens in it, and the chart shows the shape of your book from opening to close.

The typical use is: after Flags is clean, run Narrative Flow to see whether the middle sags or the third act rushes. Common fixes surfaced by the chart: split a sagging chapter into two, tighten a slow opener, add a beat to a chapter that reads flat. See Narrative Flow.

Outline

The three-level structure view: Acts → Chapters → Scenes. Each node has a title, a summary, and a status (planned / drafting / revising / done) with a colored chip. Scenes can optionally link to a document in Write, which makes the outline your project map — "Open" on any scene jumps you into the editor on that document.

The Outline is where you'll spend the most time after the Wizard finishes seeding it. It's a living document — expect to reorganise as you draft. See Overview, Lore Book, and Outline.

Portrait

A rendered image for a Lore Book entry (Character, Location, or Object). Generated on demand from the entry's fields. For your reference only; portraits never appear in the manuscript or in any exported file — they're mood board, not book cover.

Provenance

The trust ledger. Every AI-accepted change in your manuscript is recorded here with before-text, after-text, task kind (Rewrite, Expand, Draft, and so on), paragraph IDs, and timestamp.

Provenance is the answer to "how much of this book is AI?" — a single "N% human-typed" number for the whole manuscript, plus per-document breakdowns, plus a chronological event log with side-by-side diffs. Exportable as JSON (machine-verifiable) or Markdown (human-readable) for an agent, publisher, or your own records. See Provenance.

Reception

The sidebar label for Deep Analysis. In customer-facing headings we say "Deep Analysis"; the sidebar label kept the shorter word "Reception" because it fits in the nav strip.

Scope chip

The pill row at the top of a Chat thread — This scene, Whole book, or Clean room. Controls how much of your book the AI can see when it responds.

This scene limits retrieval to the current scene's paragraphs (fast, cheap, useful for close-in questions). Whole book adds a lore overview, recent docs, and FTS5 excerpts across the manuscript (slower, more context, useful for "does this motif land?" style questions). Clean room sends nothing about your book (useful for generic writing-craft questions where your specifics would bias the response). See Chat.

The / typeahead in the editor. Opens on an empty line and offers structural inserts: heading levels (H1 for chapters, H2 for scenes, H3 for subheadings), chapter break, horizontal rule, lists, quotes.

Deliberately narrow — the slash menu is not a general command palette. If we ever build a global palette it will live on ⌘K, not on /. See Write.

Sync

The ~30-second loop that pushes changes from the local database to the cloud mirror and pulls down changes from other devices. Autosave (the 1.2s local flush) is what makes typing feel instant; sync is what makes the same document appear on your laptop when you left it on your phone.

Sync is designed to be boring — the footer chip in Write flips between "local only" and "synced" every half-minute and that's the whole story. See How your data works.

The Pad

The Brainstorm Pad. The room across the hall from your manuscript — warm amber paper (a deliberate visual contrast from the cool stone paper of Write), structural wall, copy-out disabled on the AI's output.

Nothing typed there can cross into your book. Three enforcements: a system prompt that refuses to draft manuscript prose, DOM-level copy-out suppression, and no "send to editor" button anywhere in the UI. Use it to think out loud without any risk of the language creeping in. See Brainstorm Pad.

The wall

The structural rule that AI-generated language in exploratory surfaces (the Pad, Chat) cannot cross into your manuscript. Three enforcements: a system prompt that refuses to draft prose for the book, DOM-level copy-out suppression on the Pad, and no "send to editor" button anywhere.

The wall is the app's most-tested product invariant. If you find a way around it, that's a bug and we want to know. See Brainstorm Pad.

Token budget

Your weekly rolling AI allowance. Ticks down as you run AI features (each call bills at the model's deduction multiplier times the raw tokens generated). Refills on a 7-day rolling window — older calls roll out of the window continuously as time passes, freeing budget in the background rather than resetting all at once at midnight.

The Tokens page shows your current usage broken down by task kind, so you can see where the week went and which model contributed to which run. See Tokens.

Vault

Where reference material lives — research notes, uploaded PDFs, prior drafts, world bibles, character interviews, anything you want the AI to be able to see but not treat as part of the manuscript. Files are chunked, indexed into SQLite FTS5, and searchable from Ask and available as context to whole-book Chat.

Vault files never enter your manuscript file unless you copy them in yourself. They're research material, not prose. See Vault.

Voice fingerprint

The style profile the AI generates from your prose. A heavy model reads up to ~8,000 words of your existing writing (manuscript-kind Vault sources and editor documents) and returns a structured JSON profile: summary, sentence rhythm, vocabulary, distinctive habits, avoid list, examples.

Once generated and toggled on, the fingerprint injects into four surfaces: Write's inline commands, Templates, Desk, and Atomize/Echoes. The Pad, Chat, and Flags deliberately don't inject — Pad and Chat are for thinking, Flags is for observation. See Voice fingerprint.

Wall

Short for "the wall." See above.

Wizard

The 10-step interview that seeds a new manuscript — idea dump, title, genre + style line, synopsis, characters, locations, objects, chapter outline, scene plan, chapter drafting. Every accepted paragraph in the drafting step is logged to Provenance with the exact before/after text.

Budget about an hour for a full run. The Wizard is the sanctioned drafting flow (in contrast to the wall-enforced exploration surfaces like the Pad), and every step's output populates a real downstream surface — Lore Book, Outline, Write — so what you build during the interview is immediately usable after. See The Wizard.