This project provides an autonomous pipeline for generating complete novels using AI agents, transforming a seed concept into print-ready PDFs, ePubs, audiobooks, and landing pages. Aimed at researchers and power users exploring AI-driven creative content generation, it offers a fully automated workflow from initial idea to final publication, inspired by the autonomous research loop methodology.

How It Works

The pipeline employs a modify-evaluate-keep/discard loop, mirroring karpathy/autoresearch, applied to fiction writing. It operates across five co-evolving layers: voice, world, characters, outline, and chapters, with a cross-cutting canon layer for factual consistency. Changes propagate bidirectionally between layers, managed via a state tracking mechanism. Quality is maintained by two "immune systems": a mechanical scorer detecting clichés and writing anti-patterns, and an LLM judge assessing prose quality and adherence to established elements. A sophisticated multi-stage review process, culminating in a dual-persona critique by Claude Opus, iteratively refines the manuscript until major issues are resolved. The final output is then exported into various formats, including typeset LaTeX, generated art, and multi-voice audiobooks.

Quick Start & Requirements

• Installation: Clone the repository, navigate into the directory, copy .env.example to .env and populate it with API keys, then install dependencies using uv sync.
• Prerequisites: Requires Python environment managed by uv. Essential API keys include Anthropic (for writing, evaluation, review), with optional keys for fal.ai (art generation) and ElevenLabs (audiobook generation).
• Execution: Generate a seed concept with uv run python seed.py. Run the full pipeline from scratch using uv run python run_pipeline.py --from-scratch.
• Documentation: Refer to PIPELINE.md for detailed technical specifications.

Highlighted Details

• Successfully produced the first novel, "The Second Son of the House of Bells," comprising 19 chapters and 79,456 words.
• The pipeline orchestrates 27 distinct Python scripts covering all stages from concept generation to final export.
• Features a multi-layered AI quality assurance system, including mechanical checks, LLM scoring, reader panel simulation, and a final dual-persona review by Claude Opus.
• Generates multiple output formats: print-ready PDF, ePub, multi-voice audiobook, and a responsive landing page.
• Integrates AI art generation via fal.ai (Nano Banana 2) for covers and ornaments, and ElevenLabs for audiobook narration.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), sponsorships, or a public roadmap were provided in the README.

Licensing & Compatibility

The repository's license is not explicitly stated in the provided text. This lack of clarity may pose a barrier for commercial use or integration into closed-source projects, requiring further investigation.

Limitations & Caveats

Core functionality relies on external API services, necessitating API key provisioning and incurring associated costs. While art and audiobook generation are optional, their use also requires specific API keys. The pipeline's complexity and reliance on multiple AI models suggest potential setup challenges and variable output quality depending on model performance and prompt engineering.