For journalists, reviewers, newsletter writers, and anyone writing about Novel Engine.
Novel Engine is to book-writing what a build system is to software. It's a desktop application that organizes the entire editorial lifecycle — from initial concept through polished manuscript — into a structured, phase-gated pipeline powered by seven specialized AI agents.
It is not a chatbot. It is not an AI ghostwriter. It's an editorial production pipeline where the human author retains full creative authority. Each agent has a distinct role — ideation, ghostwriting, cold reading, developmental editing, copy editing, task planning, and publishing — and operates at precisely the right moment in the production process. The author drives every decision; the agents are the editorial staff.
15-phase pipeline with explicit completion gates. Each phase has defined inputs, outputs, and a designated agent. The author confirms advancement — no automatic progression. This is structured production, not freeform prompting.
Not one general AI doing everything. Spark ideates. Verity drafts. Ghostlight cold-reads. Lumen runs structural analysis. Sable copy-edits. Forge plans revisions. Quill manages publication, metadata, and query letters. Each agent has distinct instructions, thinking budgets, and context rules.
Before drafting, an interview process captures the author's writing voice — cadence, vocabulary, sentence structures, stylistic preferences. Verity then drafts prose in that captured voice, not generic AI-speak.
Everything runs on your machine. No cloud backend. No API key storage. The Claude Code CLI handles its own authentication. Your manuscripts, notes, and conversations never leave your computer.
AGPL-3.0 licensed. 197 TypeScript source files. ~50,000 lines of code. Full architecture documentation. Every design decision is inspectable, from the 5-layer clean architecture to the streaming CLI integration.
Ten published novels on Amazon, all built with this system. Independently evaluated by two AI reviewers (Claude and GPT) with scores ranging 7.0–9.4 out of 10. Not AI slop — actual books people read.
Pluggable provider architecture supports Claude Code CLI, OpenAI's Codex CLI, Ollama (fully local, with a built-in agentic tool loop), and llama-server — plus any OpenAI-compatible API endpoint. Capability-aware model routing and dual primary/secondary model pickers.
Quill researches literary agents and publishers via web search, auto-populates submission targets, generates personalized query letters, and tracks submissions through the full pipeline. The 15th pipeline phase, a dedicated IPC namespace, and a 5-component UI.
Cross-provider web search parity: DuckDuckGo-based WebSearcher for Ollama and llama-server, native web_search tool for Codex CLI, and WebSearch tool for Claude. All agents can research the web for query targets regardless of backend.
The AGPL license covers the application code only. Manuscripts, pitches, outlines, and exported books are the author's exclusive property — publish on Amazon, sell the film rights, keep it in a drawer.
10 manuscripts evaluated independently by Claude and GPT. Read the full evaluation →
“You bring the story. The agents build it into a book.”
“The pipeline is a build process: source material goes in, a production-ready manuscript comes out.”
“Seven agents. Fifteen phases. One manuscript.”
“Build is both metaphor and literal — the final phase compiles chapters via Pandoc.”
“Not a chatbot. Not an AI ghostwriter. An editorial production pipeline where the human author retains creative authority.”