Novel Generator

AI-powered novel writing agent with RAG, modular skills, always-on rules, and a web-based chapter editor.

Generate, revise, and finalize novel chapters using Claude, OpenAI, Groq, or local models via Ollama. Context from your world bible is automatically retrieved from a ChromaDB vector store and injected into every prompt.

Features

• RAG pipeline — chunking, embedding, and semantic retrieval via ChromaDB.
• Multi-provider LLM — Claude, OpenAI, Groq, Ollama with streaming support.
• Skills & rules — modular markdown files that shape every generation.
• Web editor — generate, preview, revise, and finalize chapters in the browser.
• Feedback loop — edit drafts, request revisions, then save and auto-index.
• Docker-ready — single command to build and run with persistent volumes.

Project structure

backend/
  config.py          Settings from environment variables
  rag_pipeline.py    ChromaDB embedding, chunking, retrieval
  api_client.py      Unified LLM client (Claude / OpenAI / Groq / Ollama)
  agent.py           Prompt assembly with skill and rule injection
  server.py          FastAPI REST API with SSE streaming
frontend/
  templates/         HTML
  static/            CSS + JS
skills/              Step-by-step generation workflows
rules/               Always-on constraints (tone, style, genre)
prompts/             System and chapter prompt templates
context/             Novel world files (characters, locations, plot)
chapters/            Finalized chapter output
vectorstore/         ChromaDB persistent storage

Quick start

# 1. Install
pip install -r requirements.txt

# 2. Configure
cp .env.example .env
# edit .env — set at least one API key

# 3. Run
python main.py

Open http://localhost:8000, go to the Vector DB tab, click Ingest Context Directory, then start generating chapters.

Docker

cp .env.docker.example .env.docker
# edit .env.docker — set at least one API key

docker compose up -d

Optionally include a local Ollama instance:

docker compose --profile with-ollama up -d

Persistent data lives in data/vectorstore/ and data/chapters/.

See DOCS.md for volume details, API reference, production notes, and customization.

LLM providers

Provider	Env variable	Example models
Claude	ANTHROPIC_API_KEY	claude-sonnet-4-20250514
OpenAI	OPENAI_API_KEY	gpt-4o, gpt-4o-mini
Groq	GROQ_API_KEY	llama-3.3-70b-versatile
Ollama	OLLAMA_BASE_URL	Any local model

Set DEFAULT_LLM_PROVIDER and DEFAULT_MODEL in .env.

License

MIT