Novel Engine: AI Editorial Team, From Pitch to Publish

Asking a chatbot to "write my novel" gets you generic prose in a voice that isn't yours, with no way to tell what's broken. Novel Engine gives your agent the editorial process a book actually goes through: seven specialists, a fixed 15-phase pipeline, and your approval between every phase. Ten published novels were built with this pipeline.

What it does

• Spark (pitch): pressure-tests a hunch into a premise with a protagonist, a conflict and stakes, then scaffolds the book folder.
• Verity (ghostwriter): builds a Voice Profile from your samples or a short interview, then the scene outline, story bible and every chapter. Each chapter runs draft → cold audit → surgical fix as separate passes.
• Ghostlight (first reader): reads the manuscript cold, without the outline or bible, and reports where she was hooked, confused, moved or bored.
• Lumen (developmental editor): eight structural lenses, including premise, arc, cast, pacing, scene necessity, theme, openings and endings, and a phrase-repetition audit.
• Forge (task master): turns the reports into a phased task list and self-contained revision sessions.
• Sable (copy editor): style sheet, continuity, grammar, crutch words and formatting.
• Quill (publisher): audits the build and writes the title, blurb, seven KDP keywords, BISAC categories, comps, query letter and synopsis.

How it works

1. One role at a time, in a fixed order: Pitch → Scaffold → First Draft → First Read → Structural Assessment → Revision Plan → Revision → Second Read → Second Assessment → Copy Edit → Fix Planning → Mechanical Fixes → Build → Publish & Audit → Query Agents.
2. Read rules keep feedback honest. The first reader never sees your outline. The planner never reads prose. Each role loads only the files its job allows.
3. Every phase writes a file at a fixed path, and completion is detected from disk. After each phase the agent summarizes what it produced, names the next phase, and waits for you.
4. Resume-safe. pipeline_status.py shows exactly where the book stands, so you can stop for a week and pick up where you left off.
5. Full-length novels fit. Readers and editors work in about 20,000-word batches with running notes, then synthesize from the notes.
6. A motif ledger tracks recurring phrases. Retired phrases are banned and limited ones are capped, and every third chapter the ledger is rebuilt from the real text.

Frameworks & tools

Plain Markdown files in a books/{slug}/ folder. Two optional Python scripts: pipeline status, and a build that joins chapters and uses Pandoc for DOCX and EPUB.

Why this beats prompting it yourself

One long chat forgets your voice by chapter 5 and grades its own homework. Here the voice profile is a file every chapter is checked against, the audit runs separately from the drafting, and the reader, the editor and the planner can't see each other's assumptions. You get feedback you can act on, not reassurance.

Use cases

• Write a novel from a one-line idea with you approving every step.
• Run a cold read, a developmental edit or a copy edit on a manuscript you already have.
• Turn editorial feedback into a concrete, session-by-session revision plan.
• Produce DOCX and EPUB files plus KDP metadata, a blurb and a query letter.

Known limitations

It writes and edits, but it's still your book: plan to read and steer every phase. A full novel is many sessions, not one. DOCX and EPUB need Pandoc installed; without it you get the assembled Markdown. Covers aren't generated.