Mochocyang / QMAI

What is QMAI?

QMAI is a desktop application for Chinese-language web novel writers tackling the nightmare of maintaining consistency across million-character stories. Built with TypeScript and React on a Tauri (Rust) desktop shell, it solves the classic AI writing failure mode where models forget earlier plot points, characters act out of character, and timelines collapse by chapter 50. The system automatically extracts "memories" from each saved chapter, building a searchable context engine that feeds relevant backstory to the LLM before generation. It also tracks character "souls" (personality templates), monitors foreshadowing debt, and includes a six-dimension review system checking everything from setting consistency to hook effectiveness.

Why is it gaining traction?

The killer feature is the draft isolation mechanism. AI-generated content sits in a separate draft layer until you manually confirm it, preventing bad generations from poisoning the memory system. Combined with "de-AI" rewriting that strips robotic template language, the output reads less like chatGPT output. The knowledge graph visualization showing character relationships and plot threads gives writers an actual map of their story rather than hoping the AI remembers. For the niche of Chinese web novel authors publishing 300-chapter serials, this addresses real pain that generic AI writing tools ignore entirely.

Who should use this?

Chinese-language web novel authors writing serial fiction at scale (200k+ characters) who are tired of AI forgetting character personalities by chapter 30. Writers using LLM APIs who need structured context injection rather than hoping the model pays attention to earlier chapters. The tool assumes you already have an LLM API configured and understand novel structure concepts like foreshadowing tracking and timeline management. English-only developers or casual writers working on short fiction will find the Chinese-first interface and specialized novel-writing features unnecessary overhead.

Verdict

At 78 stars with a 1.0% credibility score, this is an early-stage, highly specialized tool from a single developer solving a specific problem for a specific audience. The architecture is thoughtful and the feature set addresses real novel-writing pain points, but test coverage, documentation depth, and community size suggest proceeding with caution for production use. If you write Chinese web serials and your AI keeps breaking character consistency, this is worth evaluating; for everyone else, wait for more maturity.