THE 4ORT LAB  //  PROPERTY 14 / 21

4ort-books — The Novel Engine

An AI novel-generation pipeline that has produced two books (including an 8-chapter novel with a full audiobook), grounded in dual knowledge graphs.

/

What It Is

4ort-books is an AI novel-generation pipeline with actual output to show: two books produced, including an 8-chapter novel with a full piper-tts audiobook. Not a demo that writes a paragraph, but a system that takes a book from plan to outline to draft, then through style-lint and continuity passes, end to end.

The grounding is what keeps it coherent at book length. The pipeline runs on dual knowledge graphs: an append-only narrative KG that tracks the story's own internal truth, plus the 4ort.xyz factual graph for real-world grounding. The story can't quietly contradict itself, and it can't quietly contradict the world.

Quality isn't vibes either; it's calibrated gates: Vale prose rules, stemmed phrase-repetition detection, POV checks, and a 47-marker style-clone drift analyzer that catches when the prose drifts off-voice. Drafting runs on LLMs with prefix caching to keep long-form generation efficient.

/

Highlights

Python Postgres FastMCP LLM drafting w/ prefix caching
/

Status

  Built; book 3+ queued.