uu201/character-arc

CharacterArc(弧光) AI 小说创作应用,集项目设定、角色关系、剧情大纲、章节写作与多模型 AI 协作于一体

What it solves

CharacterArc is a desktop application designed for novelists who need to manage complex, long-term projects. It solves the fragmentation of the writing process by integrating project settings, character relationships, plot structures, and the actual writing of chapters into a single, local-first workspace, reducing the need to jump between different tools for world-building and drafting.

How it works

Built with Electron, Vue 3, and SQLite, the app functions as a centralized workbench. It uses a local database to store all project data, ensuring privacy and control. The AI integration is handled via the Vercel AI SDK, supporting various providers (OpenAI, Anthropic, DeepSeek, etc.) and a local Codex CLI. A unique "Skill" system allows the AI to execute specific writing tasks—such as polishing, pacing adjustment, or world-building—by matching tasks to built-in or project-specific skill sets using an Agent Loop for iterative reasoning.

Who it’s for

It is primarily for web-novel authors and creative writers who maintain extensive world-building assets and require AI assistance for drafting, refining, and structuring their narratives without sacrificing data privacy.

Highlights

  • Local-First Architecture: All project data is stored in a local SQLite database, ensuring full user control.
  • Comprehensive World-Building: Includes tools for managing characters, organizations, and world settings, with a visual relationship graph (Cytoscape).
  • Chapter-Centric Workflow: A three-column layout (directory, editor, AI sidebar) that binds chapters to specific plot outline nodes.
  • Skill-Driven AI: An extensible system of "Skills" (e.g., humanizing text, analyzing pacing, designing character arcs) that can be automatically matched to tasks.
  • Cover Art Workbench: A dedicated tool for generating platform-specific cover prompts and previewing images via AI image models.
  • Reasoning Isolation: Specifically handles reasoning fields from models like DeepSeek and MiniMax to keep "thought" markers out of the final manuscript.

Related

  • Project
  • Project
  • Project
  • Project
  • Project