Novel Parallel Writer

Problem to Solve

When continuing a long novel, the hard part is not whether text can be generated, but whether parallel chapters stay coherent: character voice drifts, foreshadowing is left unresolved, worldbuilding conflicts, word counts swing wildly, and the prose often carries a visible AI tone. Novel Parallel Writer turns these constraints into a reviewable pipeline, using source material, character profiles, world_view.yaml, and repeated checks to keep output aligned.

Core Workflow

• Data extraction: extracts chapter structure, character candidates, and style fingerprints from uploaded text, spoken settings, or authorized retrieval, with low-confidence results flagged.
• Persona refinement: builds core character traits, speech patterns, behavior boundaries, and forbidden-expression lists, then writes world_view.yaml for all sub-agents to follow.
• Parallel drafting: the main session handles key chapters while sub-agents draft in batches, injected with character cards, foreshadowing state, style samples, and word-count targets.
• Review and repair: runs fast or full checks covering CJK length, AI tone, POV, dialogue quality, and foreshadowing recovery; supports single-chapter pinpoint fixes and whole-volume batch cleanup.
• Delivery and archiving: outputs .docx and .txt, and archives project memory for future continuation under the same IP.

Scope and Limits

It is aimed at Chinese long-form fiction, especially continuation from source text, worldbuilding from scratch, and multi-book IP expansion. If web retrieval is enabled, the user must explicitly authorize it and takes responsibility for copyright. Embedding checks require sentence-transformers; otherwise the system falls back to statistical metrics. It does not support real-time multi-user editing and does not auto-publish to any platform.