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 is CharacterArc?

CharacterArc (弧光) is a Windows/macOS desktop application that helps novel writers organise their projects and collaborate with large‑language‑model (LLM) services. It isn’t just a chat window – it provides a full‑featured writing workspace where you can store world‑building data, outline plots, edit chapters, and run AI‑powered assistants (called Skills) to generate, polish, or analyse text. All data lives locally in a SQLite database, so you keep full control over your manuscript.


✨ Core Features

Area What you get
Project Management Create separate novel projects, each with its own settings, knowledge base, and AI run‑logs.
World‑building & Relations Structured tables for characters, organisations, and settings; visual relationship graph (Cytoscape).
Plot Outline Double‑column timeline view, drag‑and‑drop ordering, AI‑assisted expansion of outline nodes.
Chapter Workspace Three‑pane layout – file tree, rich‑text editor (TipTap), and AI side‑panel. Supports auto‑save, version snapshots, word‑count targets, focus/reading modes, and export to .txt or .docx.
AI Assistance • Draft generation, continuation, rewriting, polishing, pacing adjustments.
• Summaries, foreshadow detection, next‑scene suggestions.
• Multi‑step “draft → audit → repair” pipelines.
• Agent Loop: the model can automatically select and run the appropriate Skill based on the task.
Skill System Built‑in collections of 28 Skills (e.g., long‑form writing, style‑fingerprinting, anti‑AI‑detection).
You can also import custom Skill packages per project.
Cover Designer Generate platform‑specific cover prompts, call an image model, and compare versioned previews.
Local‑first & Multi‑vendor Works with any OpenAI‑compatible API (OpenAI, DeepSeek, Anthropic, Kimi, etc.) and Anthropic‑specific endpoints.
All requests go through a configurable base URL and API key you set in the Settings panel.
Data Privacy Everything is stored locally (workspace.db in the Electron user data folder). No cloud sync unless you add it yourself.

🛠️ Technical Stack

  • Framework: Electron + Vue 3 (TypeScript) – provides a native‑feel desktop UI.
  • State Management: Pinia.
  • UI Library: Naive UI.
  • Rich‑text Editor: TipTap (with search/replace, formatting, selection actions).
  • Persistence: SQLite (handled in the Electron main process).
  • Graph Visualisation: Cytoscape.
  • AI SDK: Vercel AI SDK (@ai-sdk/openai, @ai-sdk/anthropic).
  • Build: electron‑vite (Vite 7).
  • Document Parsing: mammoth for .docx, marked for Markdown.

🚀 Getting Started (Windows/macOS)

# 1. Install prerequisites
#    Node 18+ and pnpm 10+

# 2. Install dependencies
pnpm install

# 3. Run in development mode (Electron + Vite hot‑reload)
pnpm run dev

# 4. Build for production
pnpm run build

# 5. Package installers
pnpm run dist:win   # Windows NSIS installer
pnpm run dist:mac   # macOS DMG/ZIP (requires running on macOS)

After the first launch, open Settings → AI and add one or more API configurations (base URL, API key, model name). You can switch between them from the title bar.


📦 Skill Packages (built‑in)

The repository ships three groups of Skills (28 total) under resources/skills/:

Source Example Skills Purpose
oh-story-claudecode story-long-write, story-chapter-exec Core novel‑writing workflow (long/short stories, chapter execution, cover generation, anti‑AI‑detection, etc.).
community-skills humanizer-zh, style-fingerprint Chinese‑language polishing, style extraction and fusion.
Distilled-Novel-Toolbox novel-pacing, novel-worldbuilding, novel-commercialization Full‑stack novel toolbox – pacing, world‑building, platform‑specific commercial advice, compliance, emotion curves, etc.

Each Skill is a folder containing a SKILL.md description and the prompt/template files; the app scans them at start‑up. You can drop additional Skill folders into a project’s project-skills/ directory to extend functionality.


📄 License

MIT License – free to use, modify, and distribute.


Bottom line: CharacterArc is a locally‑run, Electron‑based writing environment that tightly integrates LLMs through a modular Skill system, giving Chinese‑language novel authors a powerful, privacy‑preserving AI co‑author.

Related

  • Project
  • Project
  • Project
  • Project
  • Project