青幕AI写作 qingmuai

QMAI is a memory-based AI writing desktop system for full-length novels, designed for serialization on the scale of two to three million characters. Its workflow extracts context before writing, consolidates chapter memory after writing, tracks relationship changes in a knowledge graph, reviews drafts to prevent character and plot drift, and leaves final confirmation to the author. Structured outlines run from a master outline to volume outlines and chapter outlines, each producing both Markdown text and visual HTML. Chapter memory is ingested and retrieved through mixed BM25, vector, and graph search, a character system keeps personalities consistent, six-dimensional review flags drift, and a plot deduction room runs multi-agent reasoning over the story. You bring your own model: configure at least one large language model API, such as an OpenAI-compatible endpoint or Ollama. Projects stay on your machine.