Inkos — Autonomous AI Novel Writing Agent

Introduction

Inkos is an autonomous AI writing system that uses multiple specialized agents to draft, review, and revise novels from an initial premise. Rather than generating text in a single pass, it separates the creative process into distinct stages — planning, drafting, auditing, and revision — with optional human review gates between each phase.

What Inkos Does

• Generates full-length novel manuscripts from a premise or outline
• Uses specialized agents for plotting, chapter drafting, continuity checking, and style editing
• Implements human review gates so authors can approve or redirect at each stage
• Supports multiple LLM backends including OpenAI, Anthropic, and local models
• Outputs structured manuscripts with chapter organization and metadata

Architecture Overview

Inkos is a TypeScript application that orchestrates multiple AI agents through a pipeline. The planning agent creates a structured outline, the drafting agent writes chapters sequentially while maintaining context, the audit agent checks for plot holes and continuity errors, and the revision agent polishes prose. Each transition between agents can pause for human review. The system maintains a persistent state so work can resume across sessions.

Self-Hosting & Configuration

• Clone the repository and install Node.js dependencies
• Configure your LLM provider API key in the environment file
• Set creative parameters like genre, tone, and target word count in the config
• Run the pipeline from the CLI or through the web interface
• Output files are saved as structured Markdown with chapter metadata

Key Features

• Multi-agent pipeline separating creative concerns into specialized roles
• Human-in-the-loop review gates for author control
• Continuity auditing that catches plot holes and character inconsistencies
• Session persistence for resuming long writing projects
• Configurable genre, tone, and stylistic preferences

Comparison with Similar Tools

• Sudowrite — commercial AI writing assistant; Inkos is open-source with full pipeline control
• NovelAI — subscription service for AI fiction; Inkos runs locally with any LLM provider
• GPT-Author — single-pass generation; Inkos uses multi-agent review and revision stages
• WriteSonic — marketing-focused AI writing; Inkos is purpose-built for long-form fiction

FAQ

Q: How long of a novel can Inkos generate? A: Length depends on your configuration and LLM context limits. The system handles chapter-by-chapter generation to work within token constraints.

Q: Can I edit the output between stages? A: Yes. Human review gates let you modify content, adjust the outline, or redirect the story before the next agent processes it.

Q: Which languages are supported? A: Inkos supports any language your LLM backend can handle, though it was primarily designed for Chinese and English fiction.

Q: Does it support local models? A: Yes. Any OpenAI-compatible endpoint works, including Ollama and other local inference servers.

Sources

• https://github.com/Narcooo/inkos