This AI doesn't want to chat—it wants to direct your novel

What it does This TypeScript monorepo orchestrates long-form novel generation as an agent-driven pipeline. Instead of the familiar chat-and-complete loop, it breaks the work into phases: auto-director brainstorming, world-building, character preparation, volume strategy, chapter breakdown, and execution. The system tracks state across the entire chain, supports checkpoint recovery, and can route planning, drafting, and review tasks to different LLM providers.

The interesting bit The project explicitly targets complete beginners over experienced authors—its core bet is that finishing a mediocre book is harder than polishing a short story, so the AI acts as a project manager rather than just a ghostwriter. The writing-style engine is the unusual standout: it extracts reusable prose features from sample text, compiles them into toggleable rules, and feeds them back into generation and quality-fixing loops so the voice stays consistent across hundreds of pages.

Key highlights

• Auto-director mode generates whole-book directions, character rosters, and volume outlines from a single vague idea, with checkpoint resume and model-swapping mid-stream.
• Creative Hub bundles dialogue, planning, tool calls, and execution status into one agent runtime with explicit Planner, Tool Registry, and approval nodes.
• Worldview, character, and knowledge-base assets are structured and vectorized via Qdrant for RAG, acting as long-term memory rather than static reference files.
• Style rules are treated as durable assets: extract features from existing prose, enable or disable them in a feature pool, and recompile the prompt rules automatically.
• Runs locally with SQLite by default; only needs Qdrant if you want the full knowledge-base experience, and supports OpenAI, DeepSeek, SiliconFlow, and xAI routing.

Caveats

• The README admits this is “not a one-click book machine” yet; the main chain moves forward but still requires human oversight at stages.
• It is explicitly aimed at novices who need to finish a book, so experienced authors looking for a lightweight co-writing tool may find the workflow heavy.
• The UI and feature set are evolving rapidly—recent release notes show heavy churn in auto-director state management and recovery logic.

Verdict Worth a look if you’re researching AI-native agent workflows or need to see how LangGraph orchestration maps to real creative processes. Skip it if you want a simple Markdown editor with an AI sidekick.