notnotype/neuro-book

What it solves

NeuroBook is designed to solve the common failures of AI-assisted long-form writing, such as "hallucinations" where the AI forgets established settings (setting drift), forgotten plot holes (unresolved foreshadowing), and the generic, formulaic style often associated with AI-generated text ("AI flavor"). It also integrates fragmented tools—like separate notes for settings and editors for text—into a single engineering-driven workspace.

How it works

The project applies software engineering principles to creative writing. It uses a World Engine to track world states via a timeline of "slices," ensuring that character attributes or historical events remain consistent over time. A Plot Workbench manages the narrative structure using a "promise system" to track foreshadowing and a decision log to record the reasoning behind plot twists.

Writing is handled by a Multi-Agent Studio, where specialized agents (Leader, Writer, Researcher) collaborate under human supervision. To ensure quality, llmlint acts as a code-linter for prose, scanning for 360 rules to identify and remove mechanical AI writing patterns.

Who it’s for

Novelists and long-form writers who want to use AI as a collaborative partner without losing creative control or narrative consistency.

Highlights

• World Engine: Prevents setting drift by calculating world states based on event sourcing rather than model memory.
• Promise System: Tracks foreshadowing as "technical debt" to ensure all plot points are eventually resolved.
• llmlint: A dedicated tool to detect and automatically fix formulaic AI writing styles.
• Local-First Data: Stores all settings and manuscripts as local Markdown and SQLite files for full data ownership.
• Agentic Workflow: Features three distinct modes (Discussion, Plan, and Execution) to ensure the human author remains the primary decision-maker.