notnotype/neuro-book

NeuroBook – AI‑augmented writing environment

What it is – NeuroBook is a desktop‑first application that lets you write long‑form fiction (or any structured narrative) while keeping the story world consistent with the help of AI agents. It treats the novel as a software project: the world state, plot outlines, and the manuscript are stored as plain‑text Markdown and a local SQLite database, and a World Engine can recompute the exact state of any character, location, or resource at any point in the timeline.

Why it matters – Large language models are good at generating short passages but they drift over long works, forget planted foreshadowing, and leave a mechanical “AI‑flavour” in the prose. NeuroBook solves these problems by:

• Persisting a world state that evolves over time instead of relying on conversational memory.
• Providing a commit‑style “promise ledger” that tracks every foreshadowed element from planting to payoff.
• Running a linting tool (llmlint) with 360 rules to detect and automatically fix typical AI‑generated writing artifacts.
• Giving the author fine‑grained control over AI agents (leader, writer, researcher) and a three‑mode workflow (discussion → plan → apply) that requires explicit approval before the AI makes changes.

Core components

Component	Role
World Engine	Stores subjects (characters, kingdoms, etc.) with time‑stamped snapshots; any moment’s state can be recomputed from the event‑sourcing log.
Plot Workbench	Visual “承载树” (story‑structure tree) and “因果树” (causality tree) for arranging chapters, scenes, and the cause‑effect chain.
Multi‑Agent Studio	A harness that runs separate AI agents (leader, writer, researcher) with configurable permissions and prompts.
llmlint	Linter that flags filler words, mechanical transitions, formulaic phrasing, etc.; can auto‑repair many issues.
Profile & Workflow editor	Visual TSX‑based editor for defining an agent’s context, tools, and step‑by‑step jobs without writing code.

How you get it

• Windows – Download the portable zip from the Releases page and run Start Neuro Book.cmd.
• Linux / macOS – Run the provided install script (curl … | sh).
• Docker / multi‑instance – Use the NeuroBook Manager (bunx @notnotype/neuro-book-manager@canary) which can spin up isolated instances, handle updates, and expose the app via a web UI.
• All installations perform an environment check and let you choose the LLM provider and API keys you prefer.

Typical workflow

1. Initialize a new “book” – the tool creates a lorebook/, manuscript/, and world-engine/ folder of Markdown/TS files.
2. Define the world – add characters, factions, calendars, etc. in the World Engine UI; each change is logged with a timestamp.
3. Plan the plot – use the Plot Workbench to map chapters → scenes and attach foreshadowing promises.
4. Invite AI agents – configure a leader (plot planner), writer (draft generator), and researcher (fact‑checking). The UI shows inline suggestions; you approve before they are written.
5. Write – edit the Markdown manuscript directly; the integrated AI can rewrite selected text, suggest continuations, or perform a lint pass.
6. Audit – at any time you can query the world state at a past date, view the promise ledger, or run llmlint to clean the prose.
7. Export – the finished book lives in plain Markdown and can be compiled to PDF, ePub, or any format you like.

Who it’s for

• Novelists and world‑builders who want AI assistance without losing narrative control.
• Game designers or tabletop‑RPG creators needing a consistent setting and traceable plot promises.
• Anyone who likes the idea of treating a story like a software project (versioned state, linting, automated checks).

Community & contribution

• Official Discord, QQ group, and a public documentation site (both Chinese and English).
• The repo is a monorepo with ~15 workspaces (core app, manager, various libraries, the llmlint package, etc.).
• Contributions are welcome: prompt‑engineering, UI redesign, documentation, testing, or even non‑technical community work.

License – AGPL‑3.0‑only. Your own writings remain your property; only modifications to NeuroBook itself are subject to the copyleft terms.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.