Novel Studio

AI-powered novel writing assistant — browser-based, multi-agent workflow + RAG consistency.

简体中文 | English

---

An author's workbench that assists with novel creation, not an auto-writer. The author controls pace and creativity throughout.

Features

• Outline Generation — provide a synopsis, get a structured chapter outline
• Staged Chapter Generation — split into planning (A+B) and writing (C+D+E+F) phases, with discussion between stages
• Agent Discussion — chat with planner/reviewer/writer agents at any point to refine plans and drafts
• Review & Polish — automated 10-dimension review with actionable feedback, or stylistic polish pass
• Chapter Editor — Monaco-based editor with plan/review side panel; plan-only chapters shown with "planning" badge
• Bible (Settings Collection) — character profiles, worldbuilding, plot notes with RAG-powered consistency checking
• Prompt Studio — edit agent system prompts and skill files directly in the browser
• Task History — persistent task tracking with retry support
• LLM Settings — configure LLM provider and model from the UI
• WebSocket Streaming — real-time pipeline progress via WebSocket events

Quick Start

Option A: Docker (recommended)

cp .env.example .env.local
# Edit .env.local — fill in your LLM API key

docker compose up --build
# Open http://localhost:8080

For China-based servers, use domestic mirrors:

docker build \
  --build-arg NPM_REGISTRY=https://registry.npmmirror.com \
  --build-arg PIP_INDEX=https://mirrors.aliyun.com/pypi/simple/ \
  -t novel-studio:1.0 .

Option B: Setup Script

./setup.sh
# Edit .env.local — fill in your LLM API key

.venv/bin/uvicorn backend.main:app --host 0.0.0.0 --port 8080
# Open http://localhost:8080

Option C: Manual

# Backend
python3 -m venv .venv
.venv/bin/pip install -e .

# Frontend
cd frontend && npm install && npm run build && cd ..

# Config
cp .env.example .env.local
# Edit .env.local

# Run (single port, production mode)
.venv/bin/uvicorn backend.main:app --host 0.0.0.0 --port 8080

Development Mode

# Terminal A — backend (auto-reload)
.venv/bin/uvicorn backend.main:app --reload --port 8080

# Terminal B — frontend (Vite dev server with HMR)
cd frontend && npm run dev
# Open http://localhost:5173

Workflow

Synopsis → Outline → [Discuss] → Plan (per chapter) → [Discuss] → Write → Review → Polish → Finalize
                        ↑                                  ↑                   ↑
                   ChatPanel                          ChatPanel           ChatPanel

Each step can be triggered independently. The author can pause, discuss with agents, edit files manually, and resume at any point.

LLM Configuration

Edit .env.local or use the Settings page in the UI:

Variable	Description	Example
LLM_API_FORMAT	openai_compat or ollama	openai_compat
LLM_BASE_URL	API endpoint	https://api.openai.com/v1
LLM_MODEL	Model name	gpt-4o
LLM_API_KEY	API key	sk-...

Embedding config is optional — falls back to the LLM config if left empty.

Project Structure

novel-studio/
├── backend/
│   ├── main.py              # FastAPI entry (API sub-app + static files)
│   ├── config.py            # Settings from .env.local
│   ├── api/                 # REST + WebSocket endpoints
│   ├── agents/              # Agent definition files (.md)
│   ├── skills/              # Skill definition files
│   ├── orchestrator/        # Pipeline logic (chapter, outline, review)
│   ├── projects/            # Workspace + persistence layer
│   ├── runtime/             # Session + agent management
│   ├── tools/               # Custom pipeline tools
│   └── vectorstore/         # Embedding + FAISS integration
├── frontend/
│   ├── src/
│   │   ├── pages/           # Projects, Workbench, Chapters, Bible, PromptStudio, Settings
│   │   ├── components/      # ChatPanel, ReviewPanel, ChapterList, MarkdownViewer, etc.
│   │   ├── hooks/           # useChat, usePipeline, useWebSocket
│   │   └── api.ts           # Backend API client
│   └── vite.config.ts       # Dev proxy config
├── pyproject.toml           # Python dependencies
├── setup.sh                 # One-click setup script
├── Dockerfile               # Multi-stage build (supports build-arg mirrors)
├── docker-compose.yml       # Docker Compose config
├── .dockerignore            # Excludes node_modules/dist from build context
└── .env.example             # LLM config template

Nginx Deployment

Example nginx reverse proxy config for production:

server {
    listen 80;
    server_name your-domain.com;

    location / {
        proxy_pass http://127.0.0.1:8080;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }

    location /api/ws {
        proxy_pass http://127.0.0.1:8080;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host $host;
        proxy_read_timeout 600s;
    }
}

Running Tests

.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest backend/tests/ -q

License

MIT